AI Watch Defining Artificial Intelligence

J R C T E C H N I C A L R E P O R T S

AI Watch

Defining Artificial Intelligence

Towards an operational

definition and taxonomy

of artificial intelligence

EUR 30117 EN

This publication is a report by the Joint Research Centre (JRC), the European Commission’s science and knowledge service. It aims to

provide evidence-based scientific support to the European policymaking process. The scientific output expressed does not imply a policy

position of the European Commission. Neither the European Commission nor any person acting on behalf of the Commission is

responsible for the use that might be made of this publication.

Contact information

Address: Edificio Expo, C/ Inca Garcilaso 3, Sevilla 41092, Spain

Email: EC-AI-[email protected].eu

EU Science Hub

https://ec.europa.eu/jrc

JRC118163

EUR 30117 EN

PDF

ISBN 978-92-76-17045-7

ISSN 1831-9424

doi: 10.2760/382730

Luxembourg: Publications Office of the European Union, 2020

The reuse policy of the European Commission is implemented by the Commission Decision 2011/833/EU of 12 December 2011 on the

reuse of Commission documents (OJ L 330, 14.12.2011, p. 39). Except otherwise noted, the reuse of this document is authorised under

the Creative Commons Attribution 4.0 International (CC BY 4.0) licence (https://creativecommons.org/licenses/by/4.0/). This means that

reuse is allowed provided appropriate credit is given and any changes are indicated. For any use or reproduction of photos or other

material that is not owned by the EU, permission must be sought directly from the copyright holders.

All content © European Union 2020, except: Cover image © Sdecoret©AdobeStock, 2018 Peshkova©AdobeStock, 2018

How to cite this report: Samoili, S., López Cobo, M., Gómez, E., De Prato, G., Martínez-Plumed, F., and Delipetrev, B., AI Watch. Defining

Artificial Intelligence. Towards an operational definition and taxonomy of artificial intelligence, EUR 30117 EN, Publications Office of the

European Union, Luxembourg, 2020, ISBN 978-92-76-17045-7, doi:10.2760/382730, JRC118163.

 
i 
Contents 
Foreword ................................................................................................................ 1 
Acknowledgements ..................................................................................................... 2 
Abstract ................................................................................................................. 3 
Executive summary ..................................................................................................... 4 
Introduction ......................................................................................................... 6 
Proposal for a common definition on Artificial intelligence ..................................................... 7 
1  AI definitions .................................................................................................. 7 
2  AI Watch operational definition of AI ........................................................................ 8 
2.1  AI taxonomy ........................................................................................... 9 
2.1.1  Sources .......................................................................................... 9 
2.1.2  AI Watch taxonomy ........................................................................... 11 
2.2  AI keywords ......................................................................................... 14 
2.2.1  Construction process .......................................................................... 14 
2.2.2  Keyword list ................................................................................... 15 
3  Collection of AI definitions and subdomains .............................................................. 17 
AI definitions and subdomains in: policy documents, research and market reports ......................... 29 
1  Policy and institutional perspective: Commission Services; National; International .................... 29 
1.1  European Commission level ........................................................................ 29 
1.1.1  High Level Expert Group on Artificial Intelligence (HLEG), 2019 .......................... 29 
1.1.2  EC Coordinated Plan on AI, 2018 ............................................................ 30 
1.1.3  European AI Strategy: EC Communication - Artificial Intelligence for Europe, 2018..... 31 
1.1.4  EC JRC Flagship report on AI: Artificial Intelligence. A European Perspective, 2018 ..... 32 
1.2  National level: European Union .................................................................... 33 
1.2.1  AI 4 Belgium Report, 2019 ................................................................... 33 
1.2.2  AI National Strategy: Denmark, 2019 ....................................................... 34 
1.2.3  AI National Strategy: France. Monitoring report, 2019 ..................................... 35 
1.2.4  Spanish RDI Strategy in Artificial Intelligence, 2019 ....................................... 37 
1.2.5  AI National Strategy: France (Villani Mission), 2018 ....................................... 38 
1.2.6  AI National Strategy: Germany, 2018 ....................................................... 39 
1.2.7  AI National Strategy: Sweden, 2018 ......................................................... 40 
1.2.8  Report of the Steering Group of the AI Programme: Finland, 2017 ....................... 41 
1.3  National level: non-EU .............................................................................. 42 
1.3.1  Australia’s Ethic Framework, 2019 .......................................................... 42 
1.3.2  US Congressional Research Service, 2019 .................................................. 43 
1.3.3  Working Paper for AI National Strategy: India, 2018 ....................................... 44 

 
ii 
1.3.4  US National Defense Authorization Act, 2018 .............................................. 45 
1.3.5  US Department of Defense, 2018 ........................................................... 46 
1.3.6  National Industrial Strategy: United Kingdom, 2018; 2017 ............................... 47 
1.3.7  AI National Strategy: Japan, 2017 ........................................................... 48 
1.3.8  AI National Strategy: China, 2017 ........................................................... 49 
1.3.9  AI National Strategy: Canada, 2017 ......................................................... 50 
1.4  International Organisations ........................................................................ 51 
1.4.1  OECD, 2019 ................................................................................... 51 
1.4.2  UNESCO, 2019 ................................................................................ 52 
1.4.3  StandICT.eu project, 2019 .................................................................... 53 
1.4.4  OECD, 2018 ................................................................................... 54 
1.4.5  ETSI, 2018 ..................................................................................... 55 
1.4.6  OECD, 2017 ................................................................................... 56 
1.4.7  World Economic Forum, 2017 ............................................................... 57 
1.4.8  ISO, 1993; 1995; 2015 ....................................................................... 58 
2  Research perspective ....................................................................................... 59 
2.1  Tsinghua University, 2018 ......................................................................... 59 
2.2  Kaplan and Haenlein, 2018 ........................................................................ 60 
2.3  Poole et al., 2017; 2010; 1998 .................................................................... 61 
2.4  Kaplan, 2016 ........................................................................................ 63 
2.5  Stone et al.: AI100, 2016 ........................................................................... 64 
2.6  Russel and Norvig, 2010 (3rd edition); 1995 ..................................................... 65 
2.7  Bruner, 2009 ........................................................................................ 66 
2.8  McCarthy, 2007 ..................................................................................... 67 
2.9  Gardner, 1999 ....................................................................................... 68 
2.10  Nakashima, 1999 ................................................................................... 69 
2.11  Nilsson, 1998 ....................................................................................... 70 
2.12  Neisser et al., 1996 ................................................................................. 71 
2.13  Fogel, 1995 ......................................................................................... 72 
2.14  Wang, 1995 ......................................................................................... 73 
2.15  Albus, 1991 ......................................................................................... 74 
2.16  Schank, 1991; 1987 ................................................................................ 75 
2.17  McCarthy, 1988 ..................................................................................... 76 
2.18  Gardner, 1987 ....................................................................................... 77 
2.19  Gardner, 1983 ....................................................................................... 78 
2.20  Newell and Simon, 1976 ........................................................................... 79 
2.21  Minsky, 1969 ........................................................................................ 80 

 
iii 
2.22  McCarthy, 1959 ..................................................................................... 81 
2.23  McCarthy et al., 1955 ............................................................................... 82 
3  Market perspective ......................................................................................... 83 
3.1  CB Insights, 2019 ................................................................................... 83 
3.2  Statista, 2017 ....................................................................................... 84 
3.3  McKinsey, 2017 ..................................................................................... 85 
Conclusions........................................................................................................ 86 
References ............................................................................................................ 87 
List of tables .......................................................................................................... 90 

Foreword

This report is published in the context of AI Watch, the European Commission knowledge service to monitor

the development, uptake and impact of Artificial Intelligence (AI) for Europe, launched in December 2018.

AI has become an area of strategic importance with potential to be a key driver of economic development. AI

also has a wide range of potential social implications. As part of its Digital Single Market Strategy, the

European Commission put forward in April 2018 a European strategy on AI in its Communication "Artificial

Intelligence for Europe" COM(2018)237. The aims of the European AI strategy announced in the

communication are:

● To boost the EU's technological and industrial capacity and AI uptake across the economy, both by

the private and public sectors

● To prepare for socio-economic changes brought about by AI

● To ensure an appropriate ethical and legal framework.

Subsequently, in December 2018, the European Commission and the Member States published a “Coordinated

Plan on Artificial Intelligence”, COM(2018)795, on the development of AI in the EU. The Coordinated Plan

mentions the role of AI Watch to monitor its implementation.

AI Watch monitors European Union’s industrial, technological and research capacity in AI; AI-related policy

initiatives in the Member States; uptake and technical developments of AI; and AI impact. AI Watch has a

European focus within the global landscape. In the context of AI Watch, the Commission works in coordination

with Member States. AI Watch results and analyses are published on the AI Watch Portal

(https://ec.europa.eu/knowledge4policy/ai-watch_en).

From AI Watch in-depth analyses we will be able to understand better European Union’s areas of strength and

areas where investment is needed. AI Watch will provide an independent assessment of the impacts and

benefits of AI on growth, jobs, education, and society.

AI Watch is developed by the Joint Research Centre (JRC) of the European Commission in collaboration with

the Directorate‑General for Communications Networks, Content and Technology (DG CONNECT). This report

addresses the following objectives of AI Watch: Developing an overview and analysis of the European AI

ecosystem.

Acknowledgements

The following researchers constitute the panel of experts that provided valuable comments and useful

critiques for this work (in alphabetical order): Virginia Dignum (Umeå University and High Level Expert Group

on AI), Anders Jonsson (Universitat Pompeu Fabra), Henrik Junklewitz (Joint Research Centre's Cyber & Digital

Citizens' Security Unit), Ramón López de Mántaras (Artificial Intelligence Research Institute (IIIA-CSIC)), José

Orallo (Valencian Research Institute for Artificial Intelligence (Universitat Politècnica de València)), Ignacio

Sánchez (Joint Research Centre's Cyber & Digital Citizens' Security Unit). The authors would also like to

acknowledge the contributions from several colleagues. The authors are grateful to Antonio Puente and

Mariana Popova (DG CNECT) for their comments. In addition, the authors are plenty grateful to Alessandro

Annoni, Paul Desruelle, Gianluca Misuraca, Stefano Nativi and Miguel Vázquez-Prada Baillet (JRC-Digital

Economy Unit) for their useful suggestions and support during the whole process.

Authors

Samoili, Sofia

López-Cobo, Montserrat

Gómez, Emilia

De Prato, Giuditta

Martínez-Plumed, Fernando

Delipetrev, Blagoj

Abstract

This report proposes an operational definition of artificial intelligence to be adopted in the context of AI

Watch, the Commission knowledge service to monitor the development, uptake and impact of artificial

intelligence for Europe. The definition, which will be used as a basis for the AI Watch monitoring activity, is

established by means of a flexible scientific methodology that allows regular revision. The operational

definition is constituted by a concise taxonomy and a list of keywords that characterise the core domains of

the AI research field, and transversal topics such as applications of the former or ethical and philosophical

considerations, in line with the wider monitoring objective of AI Watch. The AI taxonomy is designed to inform

the AI landscape analysis and will expectedly detect AI applications in neighbour technological domains such

as robotics (in a broader sense), neuroscience or internet of things. The starting point to develop the

operational definition is the definition of AI adopted by the High Level Expert Group on artificial intelligence.

To derive this operational definition we have followed a mixed methodology. On one hand, we apply natural

language processing methods to a large set of AI literature. On the other hand, we carry out a qualitative

analysis on 55 key documents including artificial intelligence definitions from three complementary

perspectives: policy, research and industry.

A valuable contribution of this work is the collection of definitions developed between 1955 and 2019, and

the summarisation of the main features of the concept of artificial intelligence as reflected in the relevant

literature.

Executive summary

This report proposes a taxonomy for artificial intelligence (AI) and a list of related keywords, as an operational

definition for AI in the framework of the AI Watch, the Commission’s knowledge service to monitor the

development, uptake and impact of artificial intelligence for Europe. AI Watch aims to monitor the industrial,

technological and research capacity, as well as policy initiatives in the Member States, uptake and technical

developments of AI and its impact. AI Watch has a European focus within the global landscape and covers all

Member States. The established operational definition will be used as a basis for the AI Watch monitoring

activity. The AI taxonomy will assist in the mapping of the AI landscape, and it is expected to detect AI

applications in other technological domains such as robotics (in a wider sense), big data, web technologies,

high performance computing, embedded systems, internet of things, etc.

The need to establish an operational definition proceeds from the absence of a mutually agreed definition

and taxonomy of AI, which would impede the attainment of the wide monitoring objective of AI Watch. It is in

this basis that we propose a multi-perspective analysis to structure the AI taxonomy. In particular we provide

a unique taxonomy that represents and interconnects all the AI subdomains from political, research and

industrial perspective. In this scope, the taxonomy reflects these perspectives and aims to cover the entire AI

landscape, which consists of economic agents with R&D or industrial AI activities. Moreover, considering that

AI is a dynamic field, we propose an iterative method that can be updated over time to capture the rapid AI

evolution, and that provides a taxonomy and a set of keywords. The method consists of the following steps: (i)

qualitative analysis of AI definitions and subdomains emanating from reports with academic, industrial and

policy perspectives, (ii) selection of definition, identification of representative keywords in AI with a natural

language processing method, and taxonomy formation, and (iii) taxonomy and keywords validation.

In the first part of the method, the objective is to collect and analyse the existing definitions and identify the

main subdomains covering all aspects in the AI field. In this scope, we conduct a qualitative analysis in a

selected set of 29 AI policy and institutional reports (including standardisation efforts, national strategies, and

international organisations reports), 23 relevant research publications and 3 market reports, from the

beginning of AI in 1955 until today. AI has been usually described in relation to human intelligence, or

intelligence in general, with many definitions referring to machines that behave like humans or are capable of

actions that require intelligence. Since human intelligence is also difficult to define and measure, and

although there have been different attempts of quantification, the objective definition of something as

subjective and abstract as intelligence, falsely gives the impression of a precision that cannot be obtained. As

a consequence, most definitions found in research, policy or market reports are vague and propose an ideal

target rather than a measurable research concept. The study of the definitions found in literature leads us to

identify four characteristics that are commonly mentioned in AI: i) perception of the environment and real-

world complexity, ii) information processing: collecting and interpreting inputs, iii) decision making, including

reasoning, learning and taking actions; and iv) achievement of pre-defined goals. Taking into consideration

these features, we consider the definition proposed by the HLEG on AI as the starting point in developing the

operational definition in AI Watch: "Artificial intelligence (AI) systems are software (and possibly also

hardware) systems designed by humans that, given a complex goal, act in the physical or digital dimension by

perceiving their environment through data acquisition, interpreting the collected structured or unstructured

data, reasoning on the knowledge, or processing the information, derived from this data and deciding the best

action(s) to take to achieve the given goal. AI systems can either use symbolic rules or learn a numeric model,

and they can also adapt their behaviour by analysing how the environment is affected by their previous

actions." (HLEG, 2019).

Although it may be considered highly technical for different audiences and objectives, it is a very

comprehensive definition, which incorporates the aspects of perception, understanding, interpretation,

interaction, decision making, adaptation to behaviour and achievement of goals, whereas other definitions do

not address them in their entirety. Considering that the HLEG definition is comprehensive, hence highly

technical and detailed, less specialised definitions can be adopted for studies of different objective, such as

enterprise surveys. In this scope, the definitions provided by the EC JRC Flagship report on AI (2018) and the

European AI Strategy (COM(2018) 237 final), or the one to be used in the AI module of the Community survey

on ICT usage and e-commerce in enterprises 2021, are suitable alternatives.

Regarding the AI subdomains, we found that despite the multiple facets of AI, and consequently the lack of a

common definition and taxonomy among research communities, literature or reports, there are a number of

common topics in the definitions and taxonomies analysed for this study. In particular, in order to establish

the taxonomy, in addition to the assessment of the aforementioned set of documents, we explored the

information provided by the official publication of the Association for the Advancement of Artificial

Intelligence (AAAI) (aitopics.org), and several top AI conferences. The proposed AI taxonomy, as a list of

representative core and transversal AI domains and subdomains, will assist us to classify R&D and industrial

agents and their activities. Therefore, it encompasses main theoretical AI scientific areas, and AI related non-

technological issues from industrial and R&D AI activities, as well as ethical and philosophical issues. It

remains linked to the HLEG definition of AI in the context of AI Watch. The proposed taxonomy follows:

AI taxonomy

AI domain

AI subdomain

Core

Reasoning

Knowledge representation

Automated reasoning

Common sense reasoning

Planning

Planning and Scheduling

Searching

Optimisation

Learning

Machine learning

Communication

Natural language processing

Perception

Computer vision

Audio processing

Transversal

Integration and

Interaction

Multi-agent systems

Robotics and Automation

Connected and Automated vehicles

Services

AI Services

Ethics and Philosophy

AI Ethics

Philosophy of AI

Source: Authors' elaboration

To complete the operational definition of AI, a list of keywords representative of the AI subdomains is

established based on a part of the techno-economic segments (TES) analytical approach that will provide at a

later stage an overview of the AI landscape worldwide. The keywords are used in text queries to identify

activities and economic agents relevant to AI, for their further analysis. The list of keywords is the result of a

multi-step process combining a semi-automatic text mining approach, desk research and domain experts'

involvement. More specifically, the top keywords from a vast collection of journals in AI are identified though

text mining in the Scopus database in three different years, from which the most frequent author’s keywords

per year are selected. Similarly, the industrial aspect is addressed by extracting keywords from firms’

descriptions. Subsequently, the initial list of keywords is reviewed by AI experts and a short selection is made.

A topic modelling is then performed, so as to detect the most representative topics and terms without the

involvement of any expert that might induce unintentional bias. The initial list and the list resulting from the

topic modelling step are merged and any redundancies are removed. External domain experts from several AI

subdomains review the list and advise on synonyms that need to be grouped and on targeting subdomains

that may not be sufficiently captured by the initial sources. The taxonomy and list of keywords is then

validated and finalised.

Valuable contributions of this work are: the collection of definitions developed between 1955 and 2019; the

summarisation of the main features of the concept of artificial intelligence as reflected in the relevant

literature; and the development of a replicable process that can provide a dynamic definition and taxonomy of

the AI.

1 Introduction

AI has become an area of strategic importance and been identified as a potential key driver of economic

development as underlined in the European strategy on AI (COM (2018) 237 on Artificial Intelligence for

Europe) and in the related Coordinated Plan (COM(2018)795). Similarly, AI has become a clear target for

national governments resulting in the formulation of national AI strategies. AI Watch is the Commission

knowledge service to monitor the development, uptake and impact of artificial intelligence (AI) for Europe,

launched in December 2018. It will monitor industrial, technological and research capacity, policy initiatives in

the Member States, uptake and technical developments of AI and its impact. AI Watch has a European focus

within the global landscape and covers all Member States.

The aim of this document is to establish an operational definition of AI formed by a concise taxonomy and a

set of keywords that characterise the core and transversal domains of AI. The operational definition is based

on a concrete and inclusive definition of AI. Such a taxonomy and keywords will assist the mapping of the AI

ecosystem of interrelated economic agents, and will allow to describe their technological areas of

specialisation. Also, it will expectedly overlap with other technological domains such as robotics (in a broader

sense), big data, web technologies, high performance computing, embedded systems, internet of things, etc.

The operational definition will be used as a basis for the monitoring activity and will serve as a reference for

the other AI Watch outputs. This objective results from the need to monitor the implementation of the EC

Coordinated Plan on AI on an annual basis, as reflected in the communication

This work has been organised in a three step strategy as follows:

— Review of existing definitions. We review AI definitions found in a selected set of 55 documents: 29 AI

policy and institutional reports, 23 relevant research publications and 3 market reports, in order to

incorporate academic, industrial and corporate perspectives.

— Definition selection, taxonomy formation and representative keywords selection: We then

adopt, based on this review, a general definition of AI and we complement this with a taxonomy and

keywords that characterise the AI domain. The keywords are extracted based on automatic text analysis

of a corpus of AI scientific references, firm-level databases, and industrial activity documents,

complemented by desk research and domain experts' involvement.

— Definition and taxonomy validation: We validate our approach with a small number of AI experts.

Since AI is a dynamic field, the described process is planned to be dynamic. This taxonomy and keywords

building process will be reviewed and iterated in the future to capture the rapid evolution of AI.

The rest of the report is organised as follows: section 2 presents the AI definition, together with its

operationalisation in the form of a taxonomy and list of keywords, as well as a description of the process.

section 3 provides detailed information about the 55 documents analysed, including the source, the text of

the definition and AI subdomains -when available-, contextual information about the source and the

document itself, and the date of publication. Finally, section 4 presents the conclusions drawn from this study.

Footnote 19 of the Annex: "AI Watch developed by the Joint Research Centre will contribute to monitoring AI-related development

and will provide a number of analyses necessary to support the implementation of the European AI initiative. Among others it will

develop AI indexes addressing all dimensions relevant for policy making. Such information will be made available at the AI Watch

portal https://ec.europa.eu/knowledge4policy/ai-watch_en".

2 Proposal for a common definition on Artificial intelligence

To establish an operational AI definition to be adopted in AI Watch, composed by a taxonomy and

representative keywords, we propose a 3-layer approach that allows the dynamic update of all the

aforementioned. This approach consists of the following layers:

i. AI definition selection,

ii. taxonomy formation with core and transversal AI subdomains,

iii. pertinent keyword selection for each subdomain of the taxonomy.

In this scope, we consider 55 documents that address the AI domain from different perspectives,

acknowledging three complementary approaches under which AI is considered:

— the policy and institutional perspective, which is especially relevant for the objective of this work given

the scope in which the AI definition is to be used, focuses on the development of the industry, the

research capacity, and the impact on society of advanced technologies. This approach considers AI as an

instrument for growth and technological development. We have collected and analysed documents from

the European Commission, national strategies and policy documents (European and non-European), as

well as other international institutions such as the OECD, UNESCO, World Economic Forum, ISO, etc.;

— the research perspective, which is the understanding of AI as a research field and its development as a

general purpose technology;

— the market perspective, which has a strong focus on industrial development and assessment of the

economic value and future market prospects.

The simultaneous consideration of the three approaches provides an overview of the past and current

perceptions of AI and how the concept evolves over time. All the collected documents provide an AI definition,

or identify or describe core and transversal AI subdomains, most of the documents present both types of

information. These were analysed in order to identify the main aspects specified as AI features, as well as the

core and transversal AI subdomains, so as to propose an operational definition and taxonomy that is useful

for the objectives of AI Watch. The thorough investigation of the concept from an ontological perspective and

the analysis of the evolution of AI as a concept and research field remain out of the scope of this study. The

details of the explored definitions can be found in section 3. Table 3 offers a collection of the definitions and

AI subdomains covered, as provided in the original documents.

2.1 AI definitions

Despite the increased interest in AI by the academia, industry and public institutions, there is no standard

definition of what AI actually involves. AI has been described by certain approaches in relation to human

intelligence, or intelligence in general. Many definitions refer to machines that behave like humans or are

capable of actions that require intelligence (US NDAA, 2019; Russel and Norvig, 1955; McCarthy, 2007;

Nilsson, 1998; Fogel, 1995; Albus, 1991; Luger and Stubblefield, 1993; Winston, 1992; McCarthy, 1988;

Gardner, 1987; 1983; Newell and Simon, 1976; Bellman, 1978; Minsky, 1969; McCarthy et al., 1955). Since

human intelligence is also difficult to define and measure, and although there have been different attempts

of quantification (Gardner, 1983; 1987; Neisser et al., 1996), the objective definition of something as

subjective and abstract as intelligence (Kaplan, 2016) falsely gives the impression of a precision that cannot

be obtained. As a consequence, most definitions found in research, policy or market reports are vague and

propose an ideal target rather than a measurable research concept.

The oversimplification of the concept of intelligence that is needed in order to define, or even develop, AI is

illustrated by Russel and Norvig (1985; 2010) and emphasised by the High Level Expert Group on Artificial

Intelligence (HLEG, 2019) when focusing on rational AI and hence considering benchmark against an ideal

performance. "A system is rational if it does the “right thing”, given what it knows" (Russel and Norvig, 1985;

2010).

Two activities are especially considered in this study when analysing AI definitions: existing standardisation

efforts, and the contribution of the High-Level Expert Group on Artificial Intelligence.

Standardisation efforts

In order to collect information on the standardisation of AI and its applications, the International Organization

for Standardization (ISO) is included in the analysis. Currently the available AI definitions are found in the

ISO/IEC 2382 of 2015, established in 1993 and 1995. In 2018, in an effort to update these definitions, two

sub committees with six working groups and one study group are formed with the goal to develop 10 AI

standards for ISO/IEC. The ISO/IEC JTC1/SC42 is the first international standards committee identifying the

entire AI ecosystem. JTC1’s scope for SC42 is to become “a systems integration entity to work with other ISO,

IEC and JTC 1 committees looking at AI applications”. Until May 2019, three standards are published with a

different objective; hence an AI definition is not included.

The High-Level Expert Group on Artificial Intelligence

The High-Level Expert Group (HLEG) on Artificial Intelligence has been appointed by the European Commission

with the main aim to support the implementation of the European AI Strategy. This includes the elaboration of

recommendations on future-related policy developments and on ethical, legal and societal issues related to

AI, including socio-economic challenges. The HLEG on AI is composed by 52 representatives from academia,

civil society and industry. The first two outputs of the HLEG on AI are the Ethics Guidelines for Trustworthy

Artificial Intelligence

, and a definition of AI

developed to describe a common understanding of the domain

and its capabilities, and serving as a supporting document for the HLEG's deliverables. The HLEG definition is

considered together with the remaining documents analysed in this study.

Common features in AI definitions

Despite the multiple facets of AI, and consequently the lack of a common definition, there are a number of

commonalities that we observe in the analysed definitions. This expression of common aspects suggests that

they may be considered as the main features of AI:

— Perception of the environment, including the consideration of the real world complexity (HLEG, 2019;

European AI Strategy, 2018; EC JRC Flagship report on AI, 2018; Tsinghua University, 2018; Nakashima,

1999; Nilsson, 1998; Poole et al., 1998; Fogel, 1995; Wang, 1995; Albus, 1991; Newell and Simon, 1976).

— Information processing: collecting and interpreting inputs (in form of data) (HLEG, 2019; European AI

Strategy, 2018; EC JRC Flagship report on AI, 2018; Kaplan and Haenlein, 2018; Tsinghua University,

2018; Nakashima, 1999; Nilsson, 1998; Poole et al., 1998; Wang, 1995).

— Decision making (including reasoning and learning): taking actions, performance of tasks (including

adaptation, reaction to changes in the environment) with certain level of autonomy (HLEG, 2019; OECD,

2019; European AI Strategy 2018; EC JRC Flagship report on AI 2018; Kaplan and Haenlein 2018;

Tsinghua University, 2018; Nilsson, 1998; Poole Mackworth and Goebel, 1998; Fogel, 1995; ISO/IEC 2382-

28, 1995; Wang, 1995; Albus, 1991; Newell and Simon, 1976).

— Achievement of specific goals: this is considered as the ultimate reason of AI systems (HLEG 2019; OECD,

2019; European AI Strategy, 2018; Kaplan and Haenlein, 2018; Poole at al., 1998; Fogel, 1995; Albus,

1991; Newell and Simon, 1976).

2.2 AI Watch operational definition of AI

The proposed AI Watch operational definition is based on a concrete definition taken as a starting point, and is

composed by a concise taxonomy and a set of keywords that characterise the core and transversal domains

of AI. To reach a common understanding on the concept of AI in the framework of AI Watch, it is important

that the starting point is an inclusive definition, hence covering all technological developments and activities

carried out by all types actors that make up the AI ecosystem, whether industrial, research, government

initiatives. Taking into consideration the features that many of the explored definitions share (see Table 3), as

well as the aforementioned objectives, we consider the definition proposed by the HLEG on AI as the

starting point for the development of the operational definition. Although it may be considered highly

technical for different audiences and objectives, it is a very comprehensive definition which incorporates the

aspects of perception, understanding, interpretation, interaction, decision making, adaptation to behaviour and

achievement of goals, whereas other definitions do not address them in their entirety:

ec.europa.eu/newsroom/dae/document.cfm?doc_id=58477

ec.europa.eu/newsroom/dae/document.cfm?doc_id=56341

HLEG definition of AI

"Artificial intelligence (AI) systems are software (and possibly also hardware) systems designed by humans(2)

that, given a complex goal, act in the physical or digital dimension by perceiving their environment through

data acquisition, interpreting the collected structured or unstructured data, reasoning on the knowledge, or

processing the information, derived from this data and deciding the best action(s) to take to achieve the given

goal. AI systems can either use symbolic rules or learn a numeric model, and they can also adapt their

behaviour by analysing how the environment is affected by their previous actions."

Other suitable definitions targeted to alternative uses

Considering that the HLEG definition is comprehensive, hence highly technical and detailed, less specialised

definitions can be adopted for studies of different objective, such as enterprise surveys. In this scope, the

definitions provided by the EC JRC Flagship report on AI (2018) (see detailed reference in subsection 3.1.1.4)

and the European AI Strategy (COM(2018) 237 final, see subsection 3.1.1.3) are suitable alternatives:

EC JRC Flagship report on AI

“AI is a generic term that refers to any machine or algorithm that is capable of observing its environment,

learning, and based on the knowledge and experience gained, taking intelligent action or proposing decisions.

There are many different technologies that fall under this broad AI definition. At the moment, ML

techniques

are the most widely used.”

European AI Strategy

"Artificial Intelligence refers to systems that display intelligent behaviour by analysing their environment and

taking action — with some degree of autonomy — to achieve specific goals."

The latter has been considered by Eurostat as the starting point for the development of a definition that will

be included in the AI module of the Community Survey on ICT Usage and e-Commerce in Enterprises 2021

upon approval.

Community survey on ICT usage and e-commerce in enterprises 2021

“Artificial intelligence refers to systems that use technologies such as: text mining, computer vision, speech

recognition, natural language generation, machine learning, deep learning to gather and/or use data to predict,

recommend or decide, with varying levels of autonomy, the best action to achieve specific goals.”

2.2.1 AI taxonomy

The proposed taxonomy addresses political, research and industrial perspectives and aims to cover and

classify the AI landscape, which consists of economic agents with R&D or industrial AI related activities.

Therefore, this taxonomy is able to detect correspondingly a wide range of core AI related scientific

subdomains (e.g. knowledge representation and reasoning, machine learning) and transversal topics such as

applications of the former (e.g. robots, automated vehicles, etc.) or ethical and philosophical considerations.

The taxonomy is presented as a reduced list of abstract high level domains and their related subdomains.

These are meant to encompass the main theoretical AI branches, as well as AI related non-technological

issues. The AI subdomains are represented by a list of keywords (see subsection 2.2.2), these will enable us to

capture the AI activities carried out by economic agents, for further analysis of the AI landscape from a

techno-economic perspective.

2.2.1.1 Sources

The AI field allows several classification approaches and corresponding divisions in specific subdomains or

topics. It should be noted again that there is no commonly agreed AI taxonomy among research communities,

literature or reports, given the rapid evolution of this knowledge domain and varied perspectives from which

AI is considered. For this part of the study, we have analysed existing taxonomies and attempts to disentangle

the AI knowledge domain. We have explored the following sources:

(

)

Machine Learning

— AITopics

: this website is an official publication of the Association for the Advancement of Artificial

Intelligence (AAAI), presenting in ordered way information about AI. The information gathered covers

different dimensions: research (through journals and conferences), AI applications, authors; and different

types of sources: papers, news, tweets, etc. The documents analysed are tagged -combining machine

learning with subject matter expert knowledge- and classified according to two main dimensions:

technological and industrial, that is considering the economic sector in which AI is developed and/or used.

We focus on the technology break down provided by this source. The following AI related fields are

considered by AITopics under AI: Assistive technologies, Cognitive science, Games, Human-centered

computing, Machine learning, Natural language, Representation & reasoning, Robots, Speech, Systems &

languages, Vision, together with other less technology related : Challenges,. Issues, History, Science

fiction, The future

— Specialised conferences: we explore the top AI conferences in order to identify submission groups as

proxies of the main current in research sub-fields. The following conference submission groups have been

considered:

o AAAI

: AI and the Web, Applications, Cognitive Modeling, Cognitive Systems, Computational

Sustainability and AI, Game Theory and Economic Paradigms, Game Playing and Interactive

Entertainment, Heuristic Search and Optimization, Human-AI Collaboration, Human-Computation

and Crowd Sourcing, Humans and AI, Knowledge Representation and Reasoning, Machine

Learning Applications, Machine Learning Methods, Multiagent Systems, Natural Language

Processing (NLP) and Knowledge Representation, NLP and Machine Learning, NLP and Text

Mining, Planning and Scheduling, Reasoning under Uncertainty, Robotics, Search and Constraint

Satisfaction, Vision.

o International Joint Conferences on Artificial Intelligence (IJCAI):

 2009: Agent-based and Multi-agent Systems, Multidisciplinary Topics and Applications,

Robotics and Vision, Natural-Language Processing, Knowledge Representation,

Reasoning and Logic, Constraints, Satisfiability, and Search, Planning and Scheduling,

Uncertainty in AI, Machine Learning, Web and Knowledge-based Information Systems.

 2018: Agent-based and Multi-agent Systems, Computer Vision, Constraints and SAT,

Heuristic Search and Game Playing, Humans and AI, Knowledge Representation and

Reasoning, Machine Learning, Machine Learning Applications, Multidisciplinary Topics

and Applications, Natural Language Processing, Planning and Scheduling, Robotics,

Uncertainty in AI.

— Documents analysed for this study (section 3): We also acknowledge the AI subdomains mentioned in the

policy, research and market reports. A summary of the main AI subdomains listed in all the documents

follow (subsection 2.3). Additionally, we analysed the taxonomy and keywords developed by the Working

Group drafting the Spanish strategy on AI

: Machine Learning, Natural Language Processing, Computer

Vision and Perception, Knowledge Representation and Reasoning, Multiagent Systems, Data Science,

Other.

— Moreover, this top-down approach is complemented with a bottom-up approach that converges with the

taxonomy. In the early results of this approach we used a natural language processing method (topic

modelling) to unbiasedly identify thematic subdomains in a collection of more than 64 thousand

industrial and R&D AI activities. This resulted in the identification of six thematic subdomains (machine

learning, computer vision, natural language processing, connected and automated vehicles, robotics, and

AI services), which have a correspondence in the proposed taxonomy.

(

)

aitopics.org

(

)

Additionally, some of the documents collected are classified under other main technological fields such as Architecture, Enterprise

Application, Information Management, Sensing and signal processing, among others.

(5)

AAAI 2018

(

)

Coordinated by Professor Ramón López de Mántaras (IIIA-CSIC), who acted also as advisor in this study.

2.2.1.2 AI Watch taxonomy

In accordance to the HLEG, AI techniques and sub-disciplines can be grouped under two big strands regarding

the systems' capabilities: (i) reasoning and decision making, (ii) and learning and perception. The first group of

capabilities includes the transformation of data into knowledge, by transforming real world information into

something understandable and usable by machines, and making decisions following an organised path of

planning, solution searching and optimisation. This strand covers the AI subdomains of Knowledge

representation and reasoning (usually making use of symbolic rules to represent and infer knowledge) and

Planning (including Planning & Scheduling, Searching, and Optimisation). The second group of capabilities

develops in absence of symbolic rules, and involves learning -meaning the extraction of information, and

problem solving based on structured or unstructured perceived data (written and oral language, image, sound,

etc.)-, adaptation and reaction to changes, behavioural prediction, etc. This second strand covers AI sub-fields

related to learning, communication and perception, such as Machine learning, Natural language processing,

and Computer vision.

The academic approach followed by the HLEG is to be complemented by considering the wider monitoring

objective of AI Watch, namely to capture and measure the AI landscape that involves multifarious economic

agents and complementary approaches, considering also the impact on society. Consequently, the taxonomy

proposed is based on the main AI domains identified by the HLEG and is expanded to cover additional

dimensions:

— the concept of rational agents, as entities that make decisions and act in relation to its environment,

including interaction with other agents;

— research and industrial developments, and other AI applications such as cloud service models offered by

service companies to accelerate AI uptake;

— other noteworthy aspects related to AI, but not necessarily technology related, arise as important subjects

in policy documents and the social debate: ethical considerations such as transparency, explainability,

accountability, fairness and safety, as well philosophical matters involving the deepest nature of AI and

its evolution.

Taking into consideration the above, we propose the following AI domains and subdomains as characterising

the AI field. They are divided into core and transversal domains, the former referring to the fundamental

goals of AI, the latter not specifically related to a particular academic discipline or area of knowledge, but as

issues common to all the core domains.

Table 1. AI domains and subdomains constituting one part of the operational definition of AI

AI taxonomy

AI domain

AI subdomain

Core

Reasoning

Knowledge representation

Automated reasoning

Common sense reasoning

Planning

Planning and Scheduling

Searching

Optimisation

Learning

Machine learning

Communication

Natural language processing

Perception

Computer vision

Audio processing

Transversal

Integration and

Interaction

Multi-agent systems

Robotics and Automation

Connected and Automated vehicles

Services

AI Services

Ethics and Philosophy

AI Ethics

Philosophy of AI

Source: Authors' elaboration

It is noteworthy that the suggested domains and subdomains are related, and not disjoint, subsets of AI. This

ensues from the nature of the AI field that embraces intertwined applications and theoretical advancements,

with fuzzy boundaries. For instance, the fact that machine learning is exploited in either computer vision,

audio processing or natural language processing does not prevent them from being separate research fields,

considered by top AI conferences topics and related literature (see subsection 2.2.1.1). At the same time,

computer vision and natural language processing are in turn embedded in more complex applications, such as

virtual personal assistants or robotic platforms. Following this continuum, we consider theoretical

advancements in one end of the taxonomy and industrial applications in the other one. We should also stress

that the defined subdomains may not be fully AI-driven. For instance, while mechanical robots do not always

incorporate AI techniques, robotics is considered as a relevant domain impacted by recent developments in AI

techniques.

In conclusion, the AI Watch taxonomy is not meant to constitute a rigid classification, but a comprehensive

collection of areas that represents AI from our three target perspectives: policy, research and industry.

In the following, we succinctly describe the above domains and subdomains, highlighting their identification in

different AI national strategies and reports.

Domain: Reasoning

Subdomains: Knowledge representation; Automated reasoning; Common sense reasoning

The domain of reasoning tackles the way machines transform data into knowledge, or infer facts from data.

Several classifications address knowledge representation and automated reasoning as a field of AI, to

describe the process of justifying (reasoning) the available data and information, provide solutions and

represent them efficiently, based on a set of symbolic rules (HLEG, 2019; Spanish RDI Strategy in Artificial

Intelligence, 2019; National Strategy: France Monitoring Report, 2019; CB Insights, 2019; AI National Strategy:

Germany, 2018; Working Paper for National Strategy: India, 2018; ETSI, 2018; National Strategy: France

(Villani Mission), 2018; AI National Strategy: China, 2017; McCarthy, 2007; Nilsson, 1998).

Domain: Planning

Subdomains: Planning and Scheduling; Searching; Optimisation

The main purpose of automated planning concerns the design and execution of strategies (e.g., an organised

set of actions) to carry out some activity, and typically performed by intelligent agents, autonomous robots

and unmanned vehicles. Unlike classical control and classification problems, the solutions are complex and

must be discovered and optimised in the multidimensional space. (HLEG, 2019; Spanish RDI Strategy in

Artificial Intelligence, 2019; National Strategy: France Monitoring Report, 2019; CB Insights, 2019; AI National

Strategy: Germany, 2018; McCarthy, 2007).

Domain: Learning

Subdomains: Machine Learning (ML)

By learning, we refer to the ability of systems to automatically learn, decide, predict, adapt and react to

changes, improving from experience, without being explicitly programmed. ML is widely included in the vast

majority of efforts to identify AI categories, as the basic algorithmic approach to achieve AI regardless the

type of learning, namely reinforcement, supervised, semi-supervised, unsupervised (HLEG, 2019; Spanish RDI

Strategy in Artificial Intelligence, 2019; StandICT.eu project, 2019; National Strategy: Denmark, 2019; National

Strategy: France Monitoring report, 2019; Australia’s Ethic Framework Dawson et al., 2019; US Congressional

Research Service, 2019; CB Insights, 2019; EC JRC Flagship report on AI, 2018; AI National Strategy: Germany,

2018; OECD, 2018; Tsinghua University, 2018; Working Paper for AI National Strategy: India, 2018; National

Industrial Strategy: UK, 2018; 2017; AI National Strategy: France (Villani Mission), 2018; US Department of

Defense, 2018; OECD, 2017; McKinsey, 2017; Stone et al.: AI100, 2016; McCarthy, 2007).

Domain: Communication

Subdomains: Natural Language Processing (NLP)

NLP, as the main task of communication, refers to the machine’s ability to identify, process, understand

and/or generate information in written and spoken human communications. It is considered as an AI

subdomain from several national strategies and AI experts, encompassing applications such as text

generation, text mining, classification, and machine translation (HLEG, 2019; Spanish RDI Strategy in Artificial

Intelligence, 2019; National Strategy: Denmark, 2019; National Strategy: France Monitoring report, 2019; CB

Insights, 2019; EC JRC Flagship report on AI, 2018; OECD, 2018; Tsinghua University, 2018; Working Paper for

AI National Strategy: India, 2018; National Strategy: France (Villani Mission), 2018; US Department of

Defense, 2018; AI National Strategy: Japan, 2017; AI National Strategy: China, 2017; McKinsey, 2017; Stone

et al.: AI100, 2016; McCarthy, 2007)

Domain: Perception

Subdomains: Computer vision; Audio processing

Perception refers to systems’ ability to become aware of their environment through the senses: vision,

hearing, manipulation. etc., being vision and hearing the most developed areas in AI. Computer vision (CV)

refers to activities that identify human faces and objects in digital images, as part of object-class detection. It

is identified as one of the essential scientific fields with parts belonging to machine perception and, thus, AI. It

is usually referred to as image pattern recognition for specific tasks, or as in a broader sense as machine

vision, with applications on face and body recognition, video content recognition, 3D reconstruction, public

safety and security, health etc. (HLEG, 2019; Spanish RDI Strategy in Artificial Intelligence, 2019; National

Strategy: Denmark, 2019; Australia’s Ethic Framework Dawson et al., 2019; US Congressional Research

Service, 2019; CB Insights, 2019; EC JRC Flagship report on AI, 2018; AI National Strategy: Germany, 2018;

Tsinghua University, 2018; Working Paper for AI National Strategy: India, 2018; OECD, 2018; US Department

of Defense, 2018; AI National Strategy: Japan, 2017; OECD, 2017; McKinsey, 2017; Stone et al.: AI100, 2016;

McCarthy, 2007). Audio processing refers to AI systems allowing the perception or generation (synthesis) of

audio signals, including speech, but also other sound material (e.g. environmental sounds, music). Speech or

voice recognition, audio processing or sound technologies are also often proposed to be archived as an AI

subdivision (AI4Belgium Report, 2019; COM(2018) 237 final; EC JRC Flagship report on AI, 2018; OECD, 2017,

2018; Tsinghua University, 2018; Working Paper for AI National Strategy: India, 2018; AI National Strategy:

Japan, 2017; McCarthy, 2007).

Domain: Integration and Interaction

Subdomains: Multi-agent systems; Robotics and Automation; Connected and Automated vehicles

(CAVs)

The transversal domain of Integration and Interaction addresses the combination of perception, reasoning,

action, learning and interaction with the environment, as well as characteristics such as distribution,

coordination, cooperation, autonomy, interaction and integration.. Robotics and Automation refers to activities

related to application and research of the technological intelligent tools to assist or substitute human activity,

or to enable actions that are not humanly possible (e.g. medical robots), to optimize technical limitations,

labour or production costs. The CAVs subdomain regards technologies of autonomous vehicles, connected

vehicles and driver assistance systems, considering all automation levels and all communication technologies

(V2X). Multi-agent systems, Unmanned systems (CAVs, drones), as well as robotics and process automation

(Application programming interface (API), robotic process automation for industrial, social and other uses) are

also mentioned as separate intrinsic subdivisions of AI (HLEG, 2019; Spanish RDI Strategy in Artificial

Intelligence, 2019; UNESCO, 2019; Australia’s Ethic Framework, 2019; National Strategy: Denmark, 2019;

National Strategy: France Monitoring report, 2019; US Congressional Research Service, 2019; CB Insights,

2019; EC JRC Flagship report on AI, 2018; COM(2018) 237 final; AI National Strategy: Germany, 2018;

Tsinghua University, 2018; Working Paper for AI National Strategy: India, 2018; National Industrial Strategy:

UK, 2018; 2017; National Strategy: France (Villani Mission), 2018; Statista 2017; McKinsey, 2017; AI National

Strategy: Japan, 2017; AI National Strategy: China, 2017; Stone et al.: AI100, 2016).

Domain: Services

Subdomains: AI Services

The transversal domain of AI services refers to any infrastructure, software and platform (e.g., cognitive

computing, ML frameworks, bots and virtual assistants, etc.) provided as (serverless) services or applications,

possibly in the cloud, which are available off the shelf and executed on demand, reducing the management of

complex infrastructures. In this regard, cloud computing services are often presented when describing the AI

landscape(US NDAA, 2019; Chinese National Strategy, 2017). Infrastructure as a Service (IaaS) is the basis of

cloud computing, providing access and management of virtual resources such as servers, storage, operating

systems and networking. Subsequently, cloud platforms (or Platform as a Service (PaaS)) are service products

of cloud applications, and can be used within Software as a Service (SaaS) architectures, which are cloud

applications and adaptive intelligence software (HLEG, 2019; Spanish RDI Strategy in Artificial Intelligence,

2019; US Department of Defense, 2018; Tsinghua University, 2018; Working Paper for AI National Strategy:

India, 2018; AI National Strategy: China, 2017; Statista, 2017; McKinsey, 2017).

Domain: Ethics & Philosophy

Subdomains: AI Ethics; Philosophy of AI

Philosophical and ethical issues associated with AI are proliferating and rising citizens’ attention and

governments’ policy interest as intelligent systems become widespread. The ethics of AI is considered as a

transversal subdomain, as AI advances and applications in different areas should ensure compliance with

ethical principles and values, including applicable regulation. Given the impact on human beings and society,

establishing trust in AI is the focus of several frameworks and initiatives by policy bodies and institutions

(HLEG, 2019; OECD, 2019; StandICT.eu project, 2019; National Strategy: France Monitoring report, 2019;

Australia’s Ethic Framework Dawson et al., 2019; EC Coordinated Action Plan on AI, 2018; European AI

Strategy: EC Communication; National Strategy: France (Villani Mission), 2018; Artificial Intelligence for

Europe, 2018).

2.2.2 AI keywords

In order to fulfil its objective as a monitoring tool, one of the outputs of AI Watch will be to provide an

overview of the worldwide landscape of AI. This effort will be conducted by applying the Techno-economic

segments (TES) analytical approach developed by the EC JRC to the AI field (Samoili S., Righi R., Cardona M.,

López Cobo M., Vázquez-Prada Baillet M., and De Prato G., 2020). This methodology is developed to map

technological (and non-technological) domains that do not correspond to standard classifications (e.g.

industries, products), and that are pervasive and cross-sectoral. It is conceived as an analytical framework and

replicable methodology to analyse and describe the dynamics of specific TES ecosystems, by exploiting

different types of factual data including non-official heterogeneous sources. The initial stages of the TES

analytical approach are presented in Samoili S., Righi R., Lopez-Cobo M., Cardona M., and De Prato G. (2019),

and a first application to the AI domain is shown in De Prato G., López Cobo., M., Samoili S., Righi R., Vázquez-

Prada Baillet, M., and Cardona M. (2019) and Samoili et al. (2019). The first step of the TES methodology

addresses the definition of a techno-economic segment, e.g. AI, followed by its operationalisation through a

list of keywords. The keywords are used in text queries to identify activities and economic agents relevant to

the technology under study, AI in this case, for further analysis.

2.2.2.1 Construction process

The domains and subdomains that are selected as characterising AI are represented by a list of keywords.

These, as the AI subdomains, cover different aspects, such as methods, algorithmic approaches, applications,

products, research areas, etc., but also address aspects such as ethics and philosophy, not as an intrinsic part

of AI, but rather as an application of ethical principles and philosophical concerns to AI. The list of keywords is

built in a multi-step process combining a semi-automatic text mining approach, desk research and domain

experts' involvement:

1. Identification of top keywords in the research domain

(a) Seed articles: First we conduct a selection of a seed subset of scientific articles where the term

"artificial intelligence" is present in the title, keywords or abstract of the publication. This first step is

run on all articles available in the Scopus Database in three different years (2005, 2009 and 2017).

The consideration of the time dimensions allows capturing recently coined terms, as well as others

that are consolidated, or even some that fell into disuse but that were important terms in the past.

(b) Expansion to cover articles not triggered by the technology term: In view of expanding the set of

investigated documents, and not limiting the analysis to the papers containing the keyword "artificial

intelligence", we consider all articles published in the journals in which the articles identified in step

1.(a) are found. In this step, 137 specialised journals are considered, while generalist journals and the

ones centred in other scientific fields are ignored. For instance, the journal “Engineering Applications

of Artificial Intelligence” would be selected, while “Physics of Life Reviews” would not, even if the

latter has published some AI related articles.

during the three referred years. The selected AI related articles amount to 25 600: 2 907 published

in 2005, 12 706 in 2009 and 9 987 in 2017. The number of different keywords included in these

papers totals 57 850. The first draft list of keywords is composed by a selection from the 300 most

frequent author's keywords per year, from which generic terms are removed.

2. Identification of keywords in the industrial dimension of the technology

In order to cover terms reflecting the recent industrial developments and AI applications, we also take into

consideration sources of industrial activity. To that end, we have analysed and extracted relevant terms from

companies’ activities descriptions. Since an equivalent to author's keywords is not available from firms'

descriptions, we obtain the most frequent terms (unigrams, bigrams and trigrams) and manually inspect their

relevancy in order to incorporate them to the draft list built in step 1.

3. Initial keyword selection

The list of candidate terms, sorted by relevance based on their frequency of occurrence, is then reviewed by

in-house researchers and a short selection is made. Terms are grouped when synonyms, very similar terms

and different spellings are found, then the groups are reduced to a single term per groups. Terms appearing in

both sub-lists are prioritised.

4. Selection of keywords through topic modelling

We consider the most representative terms from the six AI subdomains identified from topic modelling on a

corpus of 64 thousand documents of R&D and industrial activity. The subdomains are identified by applying

semantic clustering with the Latent Dirichlet Allocation (LDA) model, a generative hierarchical mixed-

membership model for discrete data (Blei & Laerty, 2009; Blei et al., 2003; Papadimitriou et al., 2000). The

model returns the most probable topics that best represent the corpus, without the involvement of any expert

to avoid unintentional bias. Only the labelling of topics is done manually. The most relevant keywords of each

of the six topics are also considered, and redundancies with terms already included in the list, removed.

5. Validation by a panel of experts in several AI subdomains

An in-house pool of researchers made a selection that was reviewed by external domain experts in several AI

areas. The advice for improvement targeted the expansion of the frontiers considered, namely the inclusion of

domains and related terms not so well captured by the research and industrial sources analysed so far.

6. Final review and selection of list of keywords per domain

As a consequence of the review in step 5, areas such as Knowledge representation and reasoning or AI ethics

and their corresponding related terms were introduced. The final taxonomy was then depicted and the final

keyword list defined. Valuable inputs in this process were: the terms describing the submission groups in top

AI conferences, the term frequencies observed in AITopics, and the terms produced by the Spanish Working

Group on AI responsible for the drafting of the Spanish strategy.

2.2.2.2 Keyword list

Table 2 presents the keywords identified as most relevant within each AI subdomain comprising the

taxonomy. This list of keywords is designed to map and model AI activities in a broad sense. The keywords are

presented grouped in the broad categories identified in the taxonomy, which, as explained in detail in

subsection 2.2.1, are not mutually exclusive. This keyword list is intended to be dynamically updated according

to new technological developments in core and transversal domains, and to agree with alternative proposals.

The rationale for building the list of keywords is to determine, in a practical way, the boundaries of the

ecosystem of economic agents active in AI. In practical terms, the list of keywords will be used taking into

account additional considerations. For instance, in order to avoid as much as possible the occurrence of false

positives, i.e., the incorrect identification as AI of activities that are not AI related, a reduced list of intrinsic-AI

keywords is used to query the data sources to identify the relevant active agents in the TES ecosystem.

Furthermore, some of the remaining keywords are considered only after conditioning its co-occurrence with

some of the core AI terms (these are the non-intrinsic AI keywords). Examples of intrinsic-AI terms used as

standalone terms to identify activities are: deep learning, face recognition, swarm intelligence and

unsupervised learning. Terms that are only used in combination with intrinsic-AI terms include, for instance:

accountability, classification, clustering, cognitive system, industrial robot, service robot and social robot, since

these non-intrinsic terms could be used in a non-AI context.

Table 2. Most relevant keywords within each AI domain

AI domain

AI subdomain

Keyword

Reasoning

Knowledge

representation;

Automated reasoning;

Common sense

reasoning

case-based reasoning

inductive programming

causal inference

information theory

causal models

knowledge representation & reasoning

common-sense reasoning

latent variable models

expert system

semantic web

fuzzy logic

uncertainty in artificial intelligence

graphical models

Planning

Planning and

Scheduling;

Searching;

Optimisation

bayesian optimisation

hierarchical task network

constraint satisfaction

metaheuristic optimisation

evolutionary algorithm

planning graph

genetic algorithm

stochastic optimisation

gradient descent

Learning

Machine learning

active learning

feature extraction

adaptive learning

generative adversarial network

adversarial machine learning

generative model

adversarial network

multi-task learning

anomaly detection

neural network

artificial neural network

pattern recognition

automated machine learning

probabilistic learning

automatic classification

probabilistic model

automatic recognition

recommender system

bagging

recurrent neural network

bayesian modelling

recursive neural network

boosting

reinforcement learning

classification

semi-supervised learning

clustering

statistical learning

collaborative filtering

statistical relational learning

content-based filtering

supervised learning

convolutional neural network

support vector machine

data mining

transfer learning

deep learning

unstructured data

deep neural network

unsupervised learning

ensemble method

Communication

Natural language

processing

chatbot

natural language generation

computational linguistics

machine translation

conversation model

question answering

coreference resolution

sentiment analysis

information extraction

text classification

information retrieval

text mining

natural language understanding

Perception

Computer vision

action recognition

object recognition

face recognition

recognition technology

gesture recognition

sensor network

image processing

visual search

image retrieval

Audio processing

computational auditory scene

analysis

sound synthesis

music information retrieval

speaker identification

sound description

speech processing

sound event recognition

speech recognition

sound source separation

speech synthesis

AI domain

AI subdomain

Keyword

Integration

and Interaction

Multi-agent systems

agent-based modelling

negotiation algorithm

agreement technologies

network intelligence

computational economics

q-learning

game theory

swarm intelligence

intelligent agent

Robotics and

Automation

cognitive system

robot system

control theory

service robot

human-ai interaction

social robot

industrial robot

Connected and

Automated vehicles

autonomous driving

self-driving car

autonomous system

unmanned vehicle

autonomous vehicle

Services

AI Services

ai application

intelligence software

ai benchmark

intelligent control

ai competition

intelligent control system

ai software toolkit

intelligent hardware development

analytics platform

intelligent software development

big data

intelligent user interface

business intelligence

internet of things

central processing unit

machine learning framework

computational creativity

machine learning library

computational neuroscience

machine learning platform

data analytics

personal assistant

decision analytics

platform as a service

decision support

tensor processing unit

distributed computing

virtual environment

graphics processing unit

virtual reality

AI Ethics and

Philosophy

AI Ethics

accountability

safety

explainability

security

fairness

transparency

privacy

Philosophy of AI

artificial general intelligence

weak artificial intelligence

strong artificial intelligence

narrow artificial intelligence

Source: Authors' elaboration

2.3 Collection of AI definitions and subdomains

Table 3 presents the keywords identified as most relevant within each AI subdomains comprising the

operational definition. The domains included in the summary table are mentioned in the collected documents

as categories or applications. The documents are ordered in descending chronological order and then by

alphabetical order within each section of: policy and institutional (European Commission level, national level

and international organisations), research and market. For longer descriptions of the AI definitions,

explanations, context, etc., see the individual subsections of section 3.

Table 3. Summary of definitions and subdomains or applications referred to in the collected documents.

Source

AI Definition

Reasoning;

Planning

Learning

Communication

Perception

Integration

and

Interaction

Services

Ethics

and

Philosophy

Other/ NA

AI reference definition for AI Watch

HLEG, 2019

“Artificial intelligence (AI) systems are software (and possibly also hardware)

systems designed by humans

that, given a complex goal, act in the physical or

digital dimension by perceiving their environment through data acquisition,

interpreting the collected structured or unstructured data, reasoning on the

knowledge, or processing the information, derived from this data and deciding the

best action(s) to take to achieve the given goal. AI systems can either use symbolic

rules or learn a numeric model, and they can also adapt their behaviour by

analysing how the environment is affected by their previous actions.”



Policy and institutional approaches

European Commission Level

EC Coordinated Action

Plan on AI, 2018

“Artificial Intelligence refers to systems that display intelligent behaviour by

analysing their environment and taking action — with some degree of autonomy —

to achieve specific goals.”



European AI Strategy: EC

Communication, Artificial

Intelligence for Europe,

2018

“Artificial intelligence (AI) refers to systems that display intelligent behaviour by

analysing their environment and taking actions – with some degree of autonomy –

to achieve specific goals.”



EC JRC Flagship report on

AI: Artificial Intelligence.

A European Perspective,

2018

“AI is a generic term that refers to any machine or algorithm that is capable of

observing its environment, learning, and based on the knowledge and experience

gained, taking intelligent action or proposing decisions. There are many different

technologies that fall under this broad AI definition. At the moment, ML techniques

are the most widely used.”



National level: European Union

Source

AI Definition

Reasoning;

Planning

Learning

Communication

Perception

Integration

and

Interaction

Services

Ethics

and

Philosophy

Other/ NA

AI4Belgium Report, 2019

Reference to the European AI Strategy definition (section 3.1.1.3):

'According to the European Commission: “AI refers to systems that display

intelligent behaviour by analysing their environment and taking actions – with

some degree of autonomy – to achieve specific goals. AI-based systems can be

purely software-based, acting in the virtual world (e.g. voice assistants, image

analysis software, search engines, speech and face recognition systems) or AI can

be embedded in hardware devices (e.g. advanced robots, autonomous cars, drones

or Internet of Things applications).” '



AI National Strategy:

Denmark, 2019

"Artificial intelligence is systems based on algorithms (mathematical formulae)

that, by analysing and identifying patterns in data, can identify the most

appropriate solution. The vast majority of these systems perform specific tasks in

limited areas, e.g. control, prediction and guidance. The technology can be designed

to adapt its behaviour by observing how the environment is influenced by previous

actions."



AI National Strategy:

France. Monitoring report,

2019

Unofficial translation:

A theoretical and practical interdisciplinary field, with objective the understanding

of the cognitive and thinking mechanisms, and their imitation by a material and

software device, for assistance or substitution purposes of human activities.

The AI definition used is reported to be the one of Russel and Norvig, 1995.



Spanish RDI Strategy in

Artificial Intelligence,

2019

“AI can be defined as the Science and Engineering that allows the design and

programming of machines capable of carrying out tasks that require intelligence.

Rather than achieving general intelligence, current AI focuses on what is known as

specific AI, which is producing very important results in many fields of application

such as natural language processing or artificial vision; however, from a scientific

and basic and applied research point of view, general AI remains the major

objective to be achieved, that is, creating an ecosystem with intelligent

multitasking systems.”



AI National Strategy:

France (Villani Mission),

2018

"AI has always been envisioned as an evolving boundary, rather than a settled

research field. Fundamentally, it refers to a programme whose ambitious objective

is to understand and reproduce human cognition; creating cognitive processes

comparable to those found in human beings. Therefore, we are naturally dealing

with a wide scope here, both in terms of the technical procedures that can be

employed and the various disciplines that can be called upon: mathematics,

information technology, cognitive sciences, etc. There is a great variety of

approaches when it comes to AI: ontological, reinforcement learning, adversarial



Source

AI Definition

Reasoning;

Planning

Learning

Communication

Perception

Integration

and

Interaction

Services

Ethics

and

Philosophy

Other/ NA

learning and neural networks, to name just a few."

AI National Strategy:

Germany, 2018

"In highly abstract terms, AI researchers can be assigned to two groups: “strong”

and “weak” AI. “Strong” AI means that AI systems have the same intellectual

capabilities as humans, or even exceed them. “Weak” AI is focused on the solution

of specific problems using methods from mathematics and computer science,

whereby the systems developed are capable of self-optimisation. To this end,

aspects of human intelligence are mapped and formally described, and systems

are designed to simulate and support human thinking."



National Industrial

Strategy: UK, 2018; 2017



AI National Strategy:

Sweden, 2018

"There is no one single, clear-cut or generally accepted definition of artificial

intelligence, but many definitions. In general, however, AI refers to intelligence

demonstrated by machines."



Report of the Steering

Group of the AI

Programme: Finland,

2017

"Artificial intelligence refers to devices, software and systems that are able to

learn and to make decisions in almost the same manner as people. Artificial

intelligence allows machines, devices, software, systems and services to function

in a sensible way according to the task and situation at hand."



National level: non-EU

Australia’s Ethic

Framework, 2019

"A collection of interrelated technologies used to solve problems autonomously

and perform tasks to achieve defined objectives without explicit guidance from a

human being "



US Congressional

Research Service, 2019



Source

AI Definition

Reasoning;

Planning

Learning

Communication

Perception

Integration

and

Interaction

Services

Ethics

and

Philosophy

Other/ NA

Working Paper for AI

National Strategy: India,

2018

“AI refers to the ability of machines to perform cognitive tasks like thinking,

perceiving, learning, problem solving and decision making. Initially conceived as a

technology that could mimic human intelligence, AI has evolved in ways that far

exceed its original conception. With incredible advances made in data collection,

processing and computation power, intelligent systems can now be deployed to

take over a variety of tasks, enable connectivity and enhance productivity.”



US National Defense

Authorization Act, 2018

“1. Any artificial system that performs tasks under varying and unpredictable

circumstances without significant human oversight, or that can learn from

experience and improve performance when exposed to data sets.

2. An artificial system developed in computer software, physical hardware, or other

context that solves tasks requiring human-like perception, cognition, planning,

learning, communication, or physical action.

3. An artificial system designed to think or act like a human, including cognitive

architectures and neural networks.

4. A set of techniques, including machine learning that is designed to approximate

a cognitive task.

5. An artificial system designed to act rationally, including an intelligent software

agent or embodied robot that achieves goals using perception, planning, reasoning,

learning, communicating, decision-making, and acting.”



US Department of

Defense, 2018



AI National Strategy:

Japan, 2017



AI National Strategy:

China, 2017



Source

AI Definition

Reasoning;

Planning

Learning

Communication

Perception

Integration

and

Interaction

Services

Ethics

and

Philosophy

Other/ NA

AI National Strategy:

Canada, 2017



International Organisations

OECD, 2019

"An AI system is a machine-based system that can, for a given set of human-

defined objectives, make predictions, recommendations, or decisions influencing

real or virtual environments. AI systems are designed to operate with varying

levels of autonomy."



UNESCO, 2019



StandICT.eu project,

2019



OECD, 2018

“AI can make business more productive, improve government efficiency and relieve

workers of mundane tasks. It can also address many of our most pressing global

problems, such as climate change and wider access to quality education and

healthcare…This combination of interdisciplinary origins, wavering trajectories, and

recent commercial success make "artificial intelligence" a difficult concept to

define and measure…The term itself is used interchangeably both as the still-

faraway goal of true machine intelligence and as the currently available

technology powering today’s hottest startups”



ETSI, 2018

“Computerized system that uses cognition to understand information and solve

problems.”

NOTE 1: ISO/IEC 2382-28 "Information technology -- Vocabulary" defines AI as "an

interdisciplinary field, usually regarded as a branch of computer science, dealing

with models and systems for the performance of functions generally associated

with human intelligence, such as reasoning and learning".

NOTE 2: In computer science AI research is defined as the study of "intelligent

agents": any device that perceives its environment and takes actions to achieve its



Source

AI Definition

Reasoning;

Planning

Learning

Communication

Perception

Integration

and

Interaction

Services

Ethics

and

Philosophy

Other/ NA

goals.

NOTE 3: This includes pattern recognition and the application of machine learning

and related techniques.

NOTE 4: Artificial Intelligence is the whole idea and concepts of machines being

able to carry out tasks in a way that mimics the human intelligence and would be

considered "smart".

OECD, 2017

“Artificial Intelligence (AI) is a term used to describe machines performing human-

like cognitive functions (e.g. learning, understanding, reasoning or interacting). It

has the potential to revolutionise production as well as contribute to tackling

global challenges related to health, transport and the environment.”



World Economic Forum,

2017

“Artificial intelligence (AI) is the software engine that drives the Fourth Industrial

Revolution. Its impact can already be seen in homes, businesses and political

processes. In its embodied form of robots, it will soon be driving cars, stocking

warehouses and caring for the young and elderly. It holds the promise of solving

some of the most pressing issues facing society, but also presents challenges such

as inscrutable “black box” algorithms, unethical use of data and potential job

displacement. As rapid advances in machine learning (ML) increase the scope and

scale of AI’s deployment across all aspects of daily life, and as the technology

itself can learn and change on its own, multistakeholder collaboration is required

to optimize accountability, transparency, privacy and impartiality to create trust.”

“Artificial intelligence (AI) or self-learning systems is the collective term for

machines that replicate the cognitive abilities of human beings. Within the broader

technological landscape, predictive maintenance in the cognitive era has the

potential to transform global production systems.”



ISO, 1993; 1995; 2015

“Branch of computer science devoted to developing data processing systems that

perform functions normally associated with human intelligence, such as reasoning,

learning, and self-improvement” (2121393: ISO, Al: term, abbreviation and

definition standardized by ISO/IEC [ISO/IEC 2382-1:1993])

“Interdisciplinary field, usually regarded as a branch of computer science, dealing

with models and systems for the performance of functions generally associated

with human intelligence, such as reasoning and learning” (2123769: term,

abbreviation and definition standardized by ISO/IEC [ISO/IEC 2382-28:1995])

“Capability of a functional unit to perform functions that are generally associated

with human intelligence such as reasoning and learning” (2123770: term,

abbreviation and definition standardized by ISO/IEC [ISO/IEC 2382-28:1995])



Source

AI Definition

Reasoning;

Planning

Learning

Communication

Perception

Integration

and

Interaction

Services

Ethics

and

Philosophy

Other/ NA

Research approach

Tsinghua University,

2018

“AI machines do not necessarily have to obtain intelligence by thinking like a

human and that it is important to make AI solve problems that can be solved by a

human brain. Brain science and brainlike intelligence research and machine-

learning represented by deep neural networks represent the two main

development directions of core AI technologies, with the latter referring to the use

of specific algorithms to direct computer systems to arrive at an appropriate

model based on existing data and use the model to make judgment on new

situations, thus completing a behavior mechanism…In general, the artificial

intelligence we know today is based on modern algorithms, supported by historical

data, and forms artificial programs or systems capable of perception, cognition,

decision making and implementation like humans.”



Kaplan and Haenlein,

2018

“Artificial intelligence (AI)—defined as a system’s ability to correctly interpret

external data, to learn from such data, and to use those learnings to achieve

specific goals and tasks through flexible adaptation.”



Poole et al., 2017; 2010;

1998

“Artificial intelligence (AI) is the established name for the field we have defined as

computational intelligence (CI), Computational intelligence is the study of the

design of intelligent agents. An agent is something that acts in an environment—it

does something. Agents include worms, dogs, thermostats, airplanes, humans,

organizations, and society. An intelligent agent is a system that acts intelligently:

What it does is appropriate for its circumstances and its goal, it is flexible to

changing environments and changing goals, it learns from experience, and it

makes appropriate choices given perceptual limitations and finite computation.”

“Artificial intelligence, or AI, is the field that studies the synthesis and analysis of

computational agents that act intelligently.

An agent is something that acts in an environment; it does something. Agents

include worms, dogs, thermostats, airplanes, robots, humans, companies, and

countries.”



Source

AI Definition

Reasoning;

Planning

Learning

Communication

Perception

Integration

and

Interaction

Services

Ethics

and

Philosophy

Other/ NA

Kaplan, 2016

“There is little agreement about what intelligence is. …there is scant reason to

believe that machine intelligence bears much relationship to human intelligence, at

least so far.”

“There are many proposed definitions on AI …most are roughly aligned around the

concept of creating computer programs or machines capable of behavior we would

regard as intelligent if exhibited by humans.”



Stone et al.: AI100, 2016

““Intelligence” remains a complex phenomenon whose varied aspects have

attracted the attention of several different fields of study, including psychology,

economics, neuroscience, biology, engineering, statistics, and linguistics. Naturally,

the field of AI has benefited from the progress made by all of these allied fields.

For example, the artificial neural network, which has been at the heart of several

AI-based solutions was originally inspired by thoughts about the flow of

information in biological neurons.”



Russel and Norvig, 2010

(3rd edition); 1995

Four categories of AI are presented and eight definitions of earlier literature.

The categories are regarding thought processes, reasoning, human and rational

behaviour. For more detailed information please refer to subsection 3.2.6.



Bruner, 2009

“…any and all systems that process information must be governed by specifiable

"rules" or procedures that govern what to do with inputs. It matters not whether it

is a nervous system, or the genetic apparatus that takes instruction from DNA and

then reproduces later generations, or whatever. This is the ideal of artificial

intelligence (AI), so-called.”



McCarthy, 2007

“It is the science and engineering of making intelligent machines, especially

intelligent computer programs. It is related to the similar task of using computers

to understand human intelligence, but AI does not have to confine itself to

methods that are biologically observable.”

“Intelligence is the computational part of the ability to achieve goals in the world.

Varying kinds and degrees of intelligence occur in people, many animals and some

machines.”



Gardner, 1999

“A biopsychological potential to process information that can be activated in a

cultural setting to solve problems or create products that are of value in a culture.”



Source

AI Definition

Reasoning;

Planning

Learning

Communication

Perception

Integration

and

Interaction

Services

Ethics

and

Philosophy

Other/ NA

Nakashima, 1999

“Intelligence is the ability to process information properly in a complex

environment. The criteria of properness are not predefined and hence not available

beforehand. They are acquired as a result of the information processing.”



Nilsson, 1998

“Artificial Intelligence (AI), broadly (and somewhat circularly) defined, is concerned

with intelligent behavior in artefacts. Intelligent behavior, in turn, involves

perception, reasoning, learning, communicating, and acting in complex

environments.”



Neisser et al., 1996

The article introduces in the AI definition the notions of adapting to the

environment, reasoning, learning etc. through a human intelligence definition, with

multiple dimensions, due to biologically inspired processes.

“Individuals differ from one another in their ability to understand complex ideas, to

adapt effectively to the environment, to learn from experience, to engage in

various forms of reasoning, to overcome obstacles by taking thought.

Concepts of intelligence are attempts to clarify and organise this complex set of

phenomena.”



Fogel, 1995

“Any system…that generates adaptive behaviour to meet goals in a range of

environments can be said to be intelligent.”



Wang, 1995

Intelligence is “the ability for an information processing system to adapt to its

environment with insufficient knowledge and resources.”



Albus, 1991

“…the ability of a system to act appropriately in an uncertain environment, where

appropriate action is that which increases the probability of success, and success

is the achievement of behavioral subgoals that support the system’s ultimate

goal.”



Schank, 1991; 1987

“AI suffers from a lack of definition of its scope. One way to attack this problem is

to attempt to list some features that we would expect an intelligent entity to have.

None of these features would define intelligence, indeed a being could lack any

one of them and still be considered intelligent. Nevertheless each attribute would

be an integral part of intelligence in its way. ...They are communication, internal

knowledge, world knowledge, intentionality, and creativity.”

“AI's primary goal is to build an intelligent machine. The second goal is to find out

about the nature of intelligence.”

“Intelligence means getting better over time.”



Source

AI Definition

Reasoning;

Planning

Learning

Communication

Perception

Integration

and

Interaction

Services

Ethics

and

Philosophy

Other/ NA

McCarthy, 1988

“The goal of artificial intelligence (A.I.) is machines more capable than humans at

solving problems and achieving goals requiring intelligence. There has been some

useful success, but the ultimate goal still requires major conceptual advances and

is probably far off.

There are three ways of attacking the goal. The first is to imitate the human

nervous system. The second is to study the psychology of human intelligence. The

third is to understand the common sense world in which people achieve their goals

and develop intelligent computer programs. This last one is the computer science

approach.”



Gardner, 1987

AI “seeks to produce, on a computer, a pattern of output that would be considered

intelligent if displayed by a human being”.

Schlinger (1992) mentions that this book also refers that “AI is viewed as a way of

testing a particular theory of how cognitive processes might work. That theory is

the popular information-processing model of cognition. Where AI researchers

disagree, according to Gardner, is how literally to interpret the thinking metaphor.

For example, some take what John Searle calls the "weak view" of AI, wherein

computer programs are simply a means for testing theories of how humans might

carry out cognitive operations. The weak view of AI is synonymous with modern

cognitive psychology.”



Gardner, 1983

“Artificial intelligence is commonly defined by referencing definitions of human

intelligence, as in Minsky’s definition.

In contrast to the standard approach of measuring one kind of intelligence (as in

standard IQ tests), Gardner (cognitive scientist) offers an eight-dimensional

definition to disentangle the oversimplification of intelligence's measurement.

In particular, he proposed multiple conceptions of intelligence, not only logical-

mathematical, linguistic, but also spatial, musical, bodily-kinaesthetic, personal.”



Newell and Simon, 1976

“By “general intelligent action” we wish to indicate the same scope if intelligence

as we see in human action: that in any real situation behavior appropriate to the

ends of the system and adaptive to the demands of the environment can occur,

within some limits of speed and complexity.”



Minsky, 1969

AI is “the science of making machines do things that would require intelligence if

done by men”.



Source

AI Definition

Reasoning;

Planning

Learning

Communication

Perception

Integration

and

Interaction

Services

Ethics

and

Philosophy

Other/ NA

McCarthy, 1959

The author, one of the founding father of AI, proposes that common sense

reasoning ability is key to AI.

“A program has common sense if it automatically deduces for itself a sufficiently

wide class of immediate consequences of anything it is told and what it already

knows.”



McCarthy et al., 1955

“..every aspect of learning or any other feature of intelligence can in principle be so

precisely described that a machine can be made to simulate it. An attempt will be

made to find how to make machines use language, form abstractions and

concepts, solve kinds of problems now reserved for humans, and improve

themselves.

…the artificial intelligence problem is taken to be that of making a machine behave

in ways that would be called intelligent if a human were so behaving.”



Market approach

CB Insights, 2019



Statista, 2017

“Artificial Intelligence (AI) essentially refers to computing technologies that are

inspired by the ways people use their brains and nervous systems to reason and

make decisions, but typically operate quite differently.”



McKinsey, 2017



Source: Authors’ elaboration.

3 AI definitions and subdomains in: policy documents, research and

market reports

3.1 Policy and institutional perspective: Commission Services; National;

International

3.1.1 European Commission level

3.1.1.1 High Level Expert Group on Artificial Intelligence (HLEG), 2019

Source

HLEG Definition of AI

Text of the

definition

“Artificial intelligence (AI) systems are software (and possibly also hardware) systems

that, given a complex goal, act in the physical or digital

dimension by perceiving their environment through data acquisition, interpreting the

collected structured or unstructured data, reasoning on the knowledge, or processing

the information, derived from this data and deciding the best action(s) to take to

achieve the given goal. AI systems can either use symbolic rules or learn a numeric

model, and they can also adapt their behaviour by analysing how the environment is

affected by their previous actions.”

Humans design AI systems directly, but they may also use AI techniques to optimise

their design.

Subdomains

As a scientific discipline, AI includes several approaches and techniques, such as

machine learning (of which deep learning and reinforcement learning are specific

examples), machine reasoning (which includes planning, scheduling, knowledge

representation and reasoning, search, and optimization), and robotics (which includes

control, perception, sensors and actuators, as well as the integration of all other

techniques into cyber-physical systems).

Context

The High-Level Expert Group (HLEG) on Artificial Intelligence has been appointed by

the European Commission, with main aim to support the implementation of the

European AI Strategy. This includes the elaboration of recommendations on future-

related policy development and on ethical, legal and societal issues related to AI,

including socio-economic challenges.

The HLEG on AI is composed by 52 representatives from academia, civil society and

industry.

The first two outputs of the HLEG on AI are the Ethics Guidelines for Trustworthy

Artificial Intelligence, and the definition on AI presented here, developed as a

supporting document for the HLEG's deliverables.

Date of

publication/

release

8 April 2019

Comments

This definition builds on the definition used in the EC Communication 'Artificial

Intelligence for Europe'.

A disclaimer warns about the oversimplification undergone for the development of

the definition and the consideration of AI capabilities and research areas.

3.1.1.2 EC Coordinated Plan on AI, 2018

Source

EC. Coordinated Plan on AI. COM(2018) 795 final and Annex

Text of the

definition

“Artificial Intelligence refers to systems that display intelligent behaviour by analysing

their environment and taking action — with some degree of autonomy — to achieve

specific goals. We are using AI on a daily basis, for example to block email spam or

speak with digital assistants.

Growth in computing power, availability of data and progress in algorithms have

turned AI into one of the most important technologies of the 21st century.”

Subdomains

“Medicine (...improve diagnoses and develop therapies for diseases for which none

exist yet)

Environment (...reduce energy consumption by optimising resources; it can contribute

to a cleaner environment by lessening the need for pesticides; it can help improve

weather prediction and anticipate disasters)

Finance & employment (AI will be the main driver of economic and productivity

growth and will contribute to the sustainability and viability of the industrial base in

Europe)”

Context

After the adoption of the European AI Strategy in April 2018, the Coordinated Action

Plan proposes joint actions for closer and more efficient cooperation between

Member States, Norway, Switzerland and the Commission in four key areas:

increasing investment, making more data available, fostering talent and ensuring

trust. Its main aim is to foster the development and use of AI in Europe.

Date of

publication/

release

07 December 2018

Comments

“The Commission is increasing investments in AI under the research and innovation

framework programme Horizon 2020 to EUR 1.5 billion in the period 2018-2020,

constituting a 70% increase compared to 2014-2017.”

“For the next MFF, the Commission has proposed to dedicate at least EUR 1 billion per

year from Horizon Europe and the Digital Europe Programme to AI.”

The definition is the same as in the European AI Strategy.

3.1.1.3 European AI Strategy: EC Communication - Artificial Intelligence for Europe, 2018

Source

EC Communication from the Commission to the European Parliament, the European

Council, the Council, the European Economic and Social Committee and the

Committee of the Regions. Artificial Intelligence for Europe. COM(2018) 237 final

{SWD(2018) 137 final}.

Text of the

definition

“Artificial intelligence (AI) refers to systems that display intelligent behaviour by

analysing their environment and taking actions – with some degree of autonomy – to

achieve specific goals.

AI-based systems can be purely software-based, acting in the virtual world (e.g. voice

assistants, image analysis software, search engines, speech and face recognition

systems) or AI can be embedded in hardware devices (e.g. advanced robots,

autonomous cars, drones or Internet of Things applications).We are using AI on a

daily basis, e.g. to translate languages, generate subtitles in videos or to block email

spam.

Many AI technologies require data to improve their performance. Once they perform

well, they can help improve and automate decision making in the same domain. For

example, an AI system will be trained and then used to spot cyber-attacks on the

basis of data from the concerned network or system.”

Subdomains

Context

EC AI COMM is prepared in order to set a European initiative on AI. In particular the

aims are (i) to promote EU’s technological and industrial capacity and AI uptake

across the economy, (ii) anticipate socio-economic changes driven by AI and adapt

accordingly, (iii) form a suitable ethical and legal framework

Date of

publication/

release

25 April 2018

Comments

3.1.1.4 EC JRC Flagship report on AI: Artificial Intelligence. A European Perspective, 2018

Source

Craglia M. (Ed.), Annoni A., Benczur P., Bertoldi P., Delipetrev P., De Prato G., Feijoo C.,

Fernandez Macias E., Gomez E., Iglesias M., Junklewitz H, López Cobo M., Martens B.,

Nascimento S., Nativi S., Polvora A., Sanchez I., Tolan S., Tuomi I., Vesnic Alujevic L.,

Artificial Intelligence - A European Perspective, EUR 29425 EN, Publications Office,

Luxembourg, 2018, ISBN 978-92-79-97217-1, doi:10.2760/11251, JRC113826

Text of the

definition

“AI is a generic term that refers to any machine or algorithm that is capable of

observing its environment, learning, and based on the knowledge and experience

gained, taking intelligent action or proposing decisions. There are many different

technologies that fall under this broad AI definition. At the moment, ML techniques

are the most widely used.”

Subdomains

Machine learning methods; Connected and automated vehicles (CAVs); Speech

recognition & NLP; Face recognition

Context

Date of

publication/

release

2018

Comments

3.1.2 National level: European Union

3.1.2.1 AI 4 Belgium Report, 2019

Source

AI 4 Belgium Coalition, AI 4 Belgium Report

Text of the

definition

Reference to the European AI Strategy definition (section 3.1.1.3):

'According to the European Commission: “AI refers to systems that display intelligent

behaviour by analysing their environment and taking actions – with some degree of

autonomy – to achieve specific goals. AI-based systems can be purely software-

based, acting in the virtual world (e.g. voice assistants, image analysis software,

search engines, speech and face recognition systems) or AI can be embedded in

hardware devices (e.g. advanced robots, autonomous cars, drones or Internet of

Things applications).” '

Subdomains

A structured taxonomy is not presented, although some AI subdomains are

mentioned: machine learning, NLP, chatbots, automation.

Identification of application domains: healthcare, environment, mobility, autonomous

driving, smart cities, fraud detection,

Context

The report is drafted with manifold objectives: bringing AI to the top of the political

agenda and public debate, stimulate a human-centred approach to AI, and providing

a first version of an overarching Belgian AI Strategy.

It provides a number of recommendations covering areas such as: education and

skills, innovation, data strategy, boost AI adoption by the private and public sectors.

Date of

publication/

release

2019

Comments

The report declares the need of an investment of at least EUR 1 billion by 2030.

3.1.2.2 AI National Strategy: Denmark, 2019

Source

Danish Government: Ministry of Finance and Ministry of Industry, Business and

Financial Affairs. Strategy for Denmark’s Digital Growth.

Larosse J. (Vanguard Initiatives Consult&Creation) for DG CNECT. Analysis of National

Initiatives on Digitising European Industry. Denmark: Towards a Digital Growth

Strategy - MADE.

Text of the

definition

“Artificial intelligence is systems based on algorithms (mathematical formulae) that,

by analysing and identifying patterns in data, can identify the most appropriate

solution. The vast majority of these systems perform specific tasks in limited areas,

e.g. control, prediction and guidance. The technology can be designed to adapt its

behaviour by observing how the environment is influenced by previous actions.

Artificial intelligence is used in a number of areas, e.g. search engines, voice and

image recognition, or to support drones and self-driving cars. Artificial intelligence

can be a crucial element to increase productivity growth and to raise the standard of

living in the years to come.”

Danish center for artificial intelligence (DCKAI), part of the Alexandra Institute-Center

for artificial intelligence:

“Artificial intelligence is an experimental science: You use your customer data to build

a model, but you test the model continuously to see if there are alternative and

better algorithms. It will be improved, the more it is being used and the larger the

database it has.

Artificial intelligence requires that you have access to large datasets, which will be

provided by the centre. Since Denmark consists of mainly small and medium-sized

companies, you could fear that they will lose the race, as they do not have the same

opportunities for developing new solutions. Lack of data does not pose the same

problem for large organisations such as IBM, Google and Amazon.”

Subdomains

Priority areas of AI use are reported on: healthcare, energy and utilities, agriculture,

transport, with focus also on big data, cybersecurity, cloud technologies.

AI applications are mentioned as language understanding, voice and image

recognition, machine learning methods, ethics, cybersecurity, robotics, drones, self-

driving cars.

Context

National Strategy: Denmark

The strategy’s objectives are: (i) to introduce common ethical and human-based

principles for AI, (ii) to promote Denmark’s AI attractiveness through research and

development on AI, (iii) to increase Danish businesses growth with AI use and

development, and (iv) improve significantly public services through AI use.

Date of

publication/

release

March 2019

Comments

Three AI institutes in DK:

- The Alexandra Institute – Center for artificial intelligence.

- DELTA (part of FORCE Technology from 01.01.2017)

- Danish Technological Institute (Ibiz-center)

1 billion DKK from 2018 to 2025, and afterwards 75 million DKK per year. More

Investment numbers available in the report.

3.1.2.3 AI National Strategy: France. Monitoring report, 2019

Source

Commission des affaires européennes. Gattolin A., Kern C., Pellevat C., Ouzoulias P.

Rapport d'information sur la stratégie européenne pour l'intelligence artificielle.

Intelligence artificielle : l'urgence d'une ambition européenne.

Text of the

definition

Unofficial translations follow:

- Annexe 3:

"Champ interdisciplinaire théorique et pratique qui a pour objet la compréhension de

mécanismes de la cognition et de la réflexion, et leur imitation par un dispositif

matériel et logiciel, à des fins d’assistance ou de substitution à des activités

humaines.

Attention : Cette publication annule et remplace celle du Journal officiel du 22

septembre 2000."

A theoretical and practical interdisciplinary field, with the objective of understanding

the cognitive and thinking mechanisms, and their imitation by a material and

software device, for assistance or substitution purposes of human activities.

Attention: This publication cancels and replaces this of the Journal officiel of

September 22nd 2000.

- Première Partie I.A.:

The AI definition used is reported to be one of Russel and Norvig, 1995:

"…l’étude des méthodes permettant aux ordinateurs de se comporter intelligemment…

l’IA inclut des tâches telles que l’apprentissage, le raisonnement, la planification, la

perception, la compréhension du langage et la robotique… ces technologies visent à

réaliser par l’informatique des tâches cognitives réalisées traditionnellement par les

êtres humains."

…the study of methods allowing to the computer to behave intelligently…AI includes

tasks as learning, reasoning, planning, perception, language comprehension and

robotics…these technologies aim to achieve with computer science cognitive tasks

that are traditionally achieved by human beings.

" Ce qu’on appelle intelligence artificielle est donc plus aujourd’hui un prolongement

de l’intelligence humaine qu’une forme autonome d’intelligence. C’est pourquoi

Charles-Édouard Bouée, PDG du cabinet Roland Berger, préfère parler d’intelligence

humaine augmentée plutôt que d’intelligence artificielle."

What it is called AI today is more an extension of human intelligence than an

autonomous form of intelligence.

This is why Charles-Édouard Bouée, CEO of Roland Berger consultancy firm, prefers to

talk about augmented human intelligence than AI.

Subdomains

The subdomains defined by Russel and Norvig, 1995:

learning, reasoning, planning, perception, understanding, language comprehension,

robotics

Context

Monitoring technical report of the French Senate to follow the objective set by the

national plan “AI for Humanity”.

Date of

publication/

release

31 January 2019

Comments

Annexe 2 (budget) from La stratégie nationale de recherche en intelligence

artificielle, 28.11.2018

- From EU: 1’5 billion € in the framework of the H2020 programme until 2020, and

for the next MFF 1 billion € per year are proposed in research on AI, as part of the

H2020, with objective to release 20 billion € of investments each year over 2020-

2030.

- From France: 1’5 billion € in AI, from which 700 million for research.

- 5’000 researchers in AI, 250 research groups, 35 master degrees specialised in AI,

300 start-ups specialised in AI [Ministère de lʼEnseignement supérieur, de la

Recherche et de lʼInnovation, date accessed 07.03.2019]

Six axis of the AI strategy:

1. Interdisciplinary AI institutes

2. Attract and keep talents

3. Support AI projects (100 million € until 2022. Since 2018 22 million € to 61

projects.)

4. Reinforce the computation means (HPC) (170 million € until 2022)

5. Reinforce private-public research partnerships (65 million € will be invested by the

state by 2022 to bring the total volume of projects to at least 130 million €, 65

millions € additional to other programs and institutes.)

6. Reinforce bilateral, European, international collaborations with Germany, Europe,

and the world.

3.1.2.4 Spanish RDI Strategy in Artificial Intelligence, 2019

Source

Spanish Ministry of Science, Innovation and Universities, Spanish RDI Strategy in

Artificial Intelligence

Text of the

definition

“AI can be defined as the Science and Engineering that allows the design and

programming of machines capable of carrying out tasks that require intelligence.

Rather than achieving general intelligence, current AI focuses on what is known as

specific AI, which is producing very important results in many fields of application

such as natural language processing or artificial vision; however, from a scientific and

basic and applied research point of view, general AI remains the major objective to be

achieved, that is, creating an ecosystem with intelligent multitasking systems.”

Subdomains

Listing the areas that the Spanish academic and scientific communities are active,

the following AI areas are mentioned:

machine learning, heuristic optimization, planning, automatic deduction, ontologies,

logic and reasoning, big data, natural language processing, artificial vision, robotics,

multi-agent systems, recommender systems, man-machine cooperation, agent-based

modelling.

Moreover the following applications are mentioned in the strategic sectors of:

health, agriculture, creative industry, industry based on experience, services, energy

and environmental sustainability, as part of the AI for society.

AI ethics are also among the strategy’s priorities, in order to avoid discrimination.

Context

AI RDI Strategy: Spain

The priorities of the strategy are to: (i) achieve organisational structure, (ii) establish

strategic areas, (iii) facilitate knowledge transfer, (iv) plan actions in AI

education/training, (v), develop a digital data ecosystem, (vi) analyse AI ethics.

Date of

publication/

release

2019

Comments

3.1.2.5 AI National Strategy: France (Villani Mission), 2018

Source

Parliamentary Mission (Villani Mission): Villani C., Schoenauer M., Bonnet Y., Berthet C.,

Cornut A.-C., Levin F., Rondepierre B. For A Meaningful Artificial Intelligence Towards A

French And European Strategy (Donner un sens à l'intelligence artificielle : pour une

stratégie nationale et européenne).

Text of the

definition

“AI has always been envisioned as an evolving boundary, rather than a settled

research field. Fundamentally, it refers to a programme whose ambitious objective is

to understand and reproduce human cognition; creating cognitive processes

comparable to those found in human beings. Therefore, we are naturally dealing with

a wide scope here, both in terms of the technical procedures that can be employed

and the various disciplines that can be called upon: mathematics, information

technology, cognitive sciences, etc. There is a great variety of approaches when it

comes to AI: ontological, reinforcement learning, adversarial learning and neural

networks, to name just a few.”

“...this technology [AI] represents much more than a research field: it determines our

capacity to organize knowledge and give it meaning, it increases our decision-making

capabilities and our control over these systems and, most notably, it enables us to

capitilize on the value of data.”

“A meaningful AI finally implies that AI should be explainable: explaining this

technology to the public so as to demystify it—and the role of the media is vital from

this point of view—but also explaining artificial intelligence by extending research into

explicability itself. AI specialists themselves frequently maintain that significant

advances could be made on this subject.”

Subdomains

APIs; Text Data Mining (including computer processes that “involve extracting

knowledge from texts or databases according to criteria of novelty or similarity”);

CAVs; Health (pre-diagnosis); Robotics; Components Industry Adapted to AI

Context

This report was assigned as a parliamentary mission by the Prime Minister É. Philippe,

and led by C. Villani, with aim to create a national strategy that, among other aims,

will make France a leader in AI. The report analyses different AI aspects: political,

economic, research, employment, ethics, social cohesion. There are separate annexes

for each of the domains of particular interest for France: education, health,

agriculture, transport, defense and security.

The AI definition presented in this fiche is used for the national strategy.

Date of

publication/

release

29 March 2018

Comments

The French Strategy for AI is also called as the “AI for Humanity” plan. The world AI

leaders are mentioned:

“The current colossi of artificial intelligence—the United States and China—and the

emerging economies in that field (Israel, Canada and the United Kingdom in

particular) have sometimes developed or are still developing in radically different

ways. France and Europe will notnecessarily need to launch their own ‘European style

Google’ to secure a place on the international stage.

The United States and China are at the forefront of this technology and their

investments far exceed those made in Europe. Canada, the United Kingdom and,

especially, Israel hold key positions in this emerging ecosystem. Considering that

France and Europe can already be regarded as “cybercolonies” in many aspects, it is

essential that they resist all forms of determinism by proposing a coordinated

response at European level.”

The role of Europe in robotics is discussed as having all the necessary to lead in this

subdomain, “whether in terms of industrial robotics, for example, or agricultural

robotics.”

Budget: 1.5 billion euros on AI during the next 5 years. More funding details are

mentioned in the report.

3.1.2.6 AI National Strategy: Germany, 2018

Source

Federal Government. Artificial Intelligence Strategy.

Text of the

definition

It is stated that a generally valid or consistently used by all stakeholders AI definition

does not exist. The AI definition used for the Federal Government’s AI Strategy is

based on the following understanding of AI:

“In highly abstract terms, AI researchers can be assigned to two groups: “strong” and

“weak” AI. “Strong” AI means that AI systems have the same intellectual capabilities

as humans, or even exceed them. “Weak” AI is focused on the solution of specific

problems using methods from mathematics and computer science, whereby the

systems developed are capable of self-optimisation. To this end, aspects of human

intelligence are mapped and formally described, and systems are designed to

simulate and support human thinking.”

Subdomains

“Weak” AI approach:

deductive reasoning systems; knowledge-based systems methods and software;

pattern analysis and recognition; robotics (autonomous systems); multimodal human-

machine interaction

Context

National Strategy: Germany

The aims of the strategy are: (i) to promote Germany’s and Europe’s leading role in

AI, (ii) ensure a responsible AI development and use, (iii) integrate AI in society. In the

framework of science and innovation promotion, an organisation specialised in AI is

established (German Research Center of Artificial Intelligence - DFKI).

Date of

publication/

release

November 2018

Comments

It is mentioned that Germany’s Government will use AI to solve specific problems,

namely the “weak” approach will be adopted. (For examples of “weak”/ “narrow”,

“strong”/ “general” AI see OECD, 2018; McCarty, 2007; Gardner, 1987)

Budget:

500 million € in the AI strategy for 2019 and the following years, up to 3 billion € by

2025.

100 additional professorships in AI.

In the last 30 years, the German government has provided just €500 million in state

aid for AI-related research. [Handelsblatt, 07.2018, date accessed 07.03.2019]

3.1.2.7 AI National Strategy: Sweden, 2018

Source

Government Offices of Sweden: Ministry of Enterprise and Innovation. National

Approach to AI (N2018.36).

Text of the

definition

Sweden’s innovation strategy approach to the AI definition is used:

“There is no one single, clear-cut or generally accepted definition of artificial

intelligence, but many definitions. In general, however, AI refers to intelligence

demonstrated by machines. Vinnova (Sweden’s innovation agency) (2018) (Artificiell

intelligens i svenskt näringsliv och samhälle. (Artificial intelligence in Swedish

business and society). Interim report 12 February 2018, Reg. no 2017-05616.”

Moreover the breadth of the field is recognised, which “encompasses many

technologies, not least machine learning and deep learning.”

Subdomains

Context

National Strategy: Sweden

The strategy’s goal is promote the Sweden’s role as an AI leader using AI, in order to

strengthen the country’s welfare and competitiveness.

Date of

publication/

release

2018

Comments

3.1.2.8 Report of the Steering Group of the AI Programme: Finland, 2017

Source

Ministry of Economic Affairs and Employment. Finland’s Age of Artificial Intelligence.

Text of the

definition

“Artificial intelligence refers to devices, software and systems that are able to learn

and to make decisions in almost the same manner as people. Artificial intelligence

allows machines, devices, software, systems and services to function in a sensible

way according to the task and situation at hand.”

The absence of a widely accepted definition is stated.

Subdomains

Context

Second interim report of the Steering Group of the Artificial Intelligence Programme

appointed by the Ministry of Economic Affairs and Employment.

The Finnish government is expected to implement the recommendations as

government policy.

Eight key actions are mentioned that are expected to promote Finland to leader in AI:

“1. Enhancement of business competitiveness through the use of AI

2. Effective utilisation of data in all sectors

3. Ensure AI can be adopted more quickly and easily

4. Ensure top-level expertise and attract top experts

5. Make bold decisions and investments

6. Build the world’s best public services

7. Establish new models for collaboration

8. Make Finland a frontrunner in the age of AI”

Date of

publication/

release

18 December 2017

Comments

US companies and innovation hubs are found to be leading in AI applications.

Chinese government is promoting AI development.

A SWOT analysis for Finland is provided. It is found that in Finland the use of AI: will

improve public sector’s efficiency, society and education will be significantly affected,

as well as other sectors. Moreover enterprise-driven ecosystems are promoted to

improve AI implementation.

3.1.3 National level: non-EU

3.1.3.1 Australia’s Ethic Framework, 2019

Source

Dawson, D. and Schleiger, E., Horton, J., McLaughlin, J., Robinson, C., Quezada, G.,

Scowcroft, J., Hajkowicz S. Artificial Intelligence: Australia’s Ethics Framework. Data61

CSIRO, Australia.

Text of the

definition

“A collection of interrelated technologies used to solve problems autonomously and

perform tasks to achieve defined objectives without explicit guidance from a human

being.”

“This definition of AI encompasses both recent, powerful advances in AI such as

neural nets and deep learning, as well as less sophisticated but still important

applications with significant impacts on people, such as automated decision systems.”

The categorisation between “narrow” and “general” AI is mentioned. The “narrow AI”

performs specific functions. The “general AI” “is comparable to human intelligence

across a range of fields”.

In the country’s plan on innovation and science (Innovation and Science Australia

2017, Australia 2030: prosperity through innovation, Australian Government,

Canberra), AI is defined as follows in the acronyms, abbreviations and glossary part:

“Computer systems that are able to perform tasks normally requiring human

intelligence.”

Subdomains

Algorithms; mechanical systems (robots, autonomous vehicles, etc.)

The following ethical principles that should be applied in AI are mentioned:

1. The benefits of any AI systems are greater that its costs.

2. Minimise negative harmful and deceitful outcomes to humans.

3. Regulatory and legal compliance to all relevant obligations, regulations and laws

national and international.

4. Peoples’ private data protection.

5. Ensure fair treatment of human individuals, communities or groups.

6. For transparency reasons, people will be informed when an algorithm is applied,

and which information it uses for decision-making.

7. In the case that an algorithm affects a person, an efficient process should be

ensured, so that the person can “challenge the use or output of the algorithm”.

8. People and organisations that create and implement an AI algorithm are

accountable for its impact.

These could be considered as potential subdivisions of the AI ethics subdomain.

Context

Governmental discussion paper on AI ethics to ensure a responsible development and

application of AI in Australia. The focus is set on “narrow AI”, as “general AI” is not

seen as a likely prospect by 2030.

Date of

publication/

release

4 March 2019

Comments

Australia does not presently have an AI national strategy. A Technology Roadmap, a

Standards Framework, and a national Ethics Framework are planned. An AU$29.9

million investment was announced to promote AI development in Australia. Currently

AI and automation are included in the national Innovation Strategy (Australia 2030:

Prosperity Through Innovation, 2017), in the Victorian All-Party Parliamentary Group

on Artificial Intelligence (VAPPGAI, March 2018), and the Digital Economy Strategy

(September 2017).

3.1.3.2 US Congressional Research Service, 2019

Source

US Congressional Research Service. Artificial Intelligence and National Security.

Text of the

definition

The absence of a commonly accepted definition is again stated. Among the reasons

are the numerous and diverse approaches of research in AI. The report is using NDAA,

2018 definition of AI.

Subdomains

Narrow AI notion is used, with all current AI systems being assigned to this category.

This includes:

machine learning, image recognition, IoT, autonomous/ human-supervises/semi-

autonomous weapon system, robot.

Automated systems are defined as the superset that includes AI, robots and

autonomous systems, which intersect each other.

Context

Report prepared for the US Congress.

Date of

publication/

release

30 January 2019

Comments

A US AI national strategy is not yet signed, but on February 2019 an executive order

was signed to establish the American AI Initiative.

This is expected to include aims to promote to AI research, R&D and workforce

development, while proposing an international engagement.

Older reports are the following:

- Preparing for the Future of Artificial Intelligence. October 2016: Recommendations

on AI regulations, automation, ethics, fairness, security and publicly funded R&D.

- National Artificial Intelligence Research and Development Strategic Plan. October

2016: Strategic plan outline for publicly funded R&D in AI.

- Artificial Intelligence, Automation, and the Economy. December 2016: The impact of

automation, the benefits and the costs of AI were studied, in order to provide policy

recommendations.

3.1.3.3 Working Paper for AI National Strategy: India, 2018

Source

National Strategy for Artificial Intelligence #AIFORALL

Text of the

definition

“AI refers to the ability of machines to perform cognitive tasks like thinking,

perceiving, learning, problem solving and decision making. Initially conceived as a

technology that could mimic human intelligence, AI has evolved in ways that far

exceed its original conception. With incredible advances made in data collection,

processing and computation power, intelligent systems can now be deployed to take

over a variety of tasks, enable connectivity and enhance productivity.”

Three different ways of categorising AI are also offered:

(a) weak vs. strong AI: “weak AI describes “simulated” thinking”, namely “a system

which appears to behave intelligently, but doesn't have any kind of consciousness

about what it's doing”,

(b) narrow vs. general AI: “narrow AI describes an AI that is limited to a single task or

a set number of tasks”

which it surpasses human intelligence, if it ever does”.

Subdomains

Three main categories of AI technologies are identified:

(i) sense: computer vision; audio processing;

(ii) comprehend: natural language processing; knowledge representation

(iii) act: machine learning; expert systems

Virtual agents, cognitive robotics, speech and identity analytics, recommendation

systems, and data visualisation are presented as AI solutions.

Context

National strategy: India, Discussion Paper

Healthcare, agriculture, education, smart cities and infrastructure, smart mobility and

transportation, are identified as the areas that AI would be beneficial in covering

societal needs. The report is intended as an initiator of an evolving AI national

strategy.

Date of

publication/

release

June 2018

Comments

3.1.3.4 US National Defense Authorization Act, 2018

Source

US National Defense Authorization Act for Fiscal Year 2019.

Text of the

definition

In section 238 it is mentioned:

“1. Any artificial system that performs tasks under varying and unpredictable

circumstances without significant human oversight, or that can learn from experience

and improve performance when exposed to data sets.

2. An artificial system developed in computer software, physical hardware, or other

context that solves tasks requiring human-like perception, cognition, planning,

learning, communication, or physical action.

3. An artificial system designed to think or act like a human, including cognitive

architectures and neural networks.

4. A set of techniques, including machine learning that is designed to approximate a

cognitive task.

5. An artificial system designed to act rationally, including an intelligent software

agent or embodied robot that achieves goals using perception, planning, reasoning,

learning, communicating, decision-making, and acting.”

Subdomains

Context

Federal Law that specifies the policies, budget and expenditure of the US Department

of Defense for 2019.

Date of

publication/

release

3 January 2018

Comments

3.1.3.5 US Department of Defense, 2018

Source

US Department of Defense, Govini. Artificial intelligence, big data and cloud

taxonomy.

Text of the

definition

Subdomains

Learning and Intelligence: modeling and simulation, DL, ML, NLP, data mining;

Advanced Computing: super-computing, neuromorphic engineering, quantum

computing;

AI systems: virtual reality, computer vision, virtual agents

Cloud service models are also mentioned (IaaS, PaaS, SaaS).

Context

US Department of Defense (DoD) report to analyse the critical to AI technologies

critical, and the vendor landscape and performance within the 25 sub-segments that

are found.

Date of

publication/

release

2018

Comments

Govini is a US big data and analytics firm contracted by the DoD. DoD considers AI as

a “technological cornerstone” for its Third Offset Strategy.

3.1.3.6 National Industrial Strategy: United Kingdom, 2018; 2017

Source

HM Government: Department for Business, Energy & Industrial Strategy, Department

for Digital, Culture, Media & Sport. Industrial Strategy. Artificial Intelligence Sector

Deal.

HM Government: Department for Business, Energy & Industrial Strategy. Industrial

Strategy. Building a Britain fit for the future.

Text of the

definition

Subdomains

Machine learning and robotics are mentioned as parts of examples for the uses that

the strategy aims to achieve, without further indications of AI subdomains.

Context

National Strategy: UK

The strategy’s aims are to position UK as global leader in AI based on ideas, people,

infrastructure, business environment, communities across the UK.

Date of

publication/

release

April 2018

November 2017

Comments

Budget:

- £20 million in AI applications for the services sector

- £93 million for robotics with multiple uses

- £20 million to stimulate among other ways the AI uptake

- £300 million for AI research funding, £83 million for AI grants, £42 million for the

expansion of the Alan Turing Institute, with £30 million from private funding.

3.1.3.7 AI National Strategy: Japan, 2017

Source

Strategic Council for AI Technology. Artificial Intelligence Technology Strategy.

Text of the

definition

Subdomains

Vision; virtual reality (VR); autonomous driving; robots; natural language processing;

image recognition; voice recognition/synthesis; prediction

Context

National Strategy: Japan

AI is seen a set of valuable services with a roadmap for its development in three

phases: (i) the use and application of data-driven AI, (ii) the public use of AI and data,

(iii) the creation of ecosystems through multi-domains connections.

A strong focus is set on the data management, the academia-industry collaborations,

the technological and system development and system development for AI start-ups

and their matching with large corporations or financial institutions

Date of

publication/

release

31 March 2017

Comments

Japan was among the first countries that developed a national AI strategy. The

strategy presents AI’s development phases for Japan. The strategy combines US and

Chinese aims.

3.1.3.8 AI National Strategy: China, 2017

Source

China’s State Council. Next Generation Artificial Intelligence Development Plan (AIDP).

Original report. Translated report.

Text of the

definition

Subdomains

Knowledge computing engines and knowledge service technology;

Cross-medium analytical reasoning technology;

Swarm intelligence technology;

Autonomous unmanned systems;

Intelligent virtual reality modelling technology;

Intelligent computing chips and systems;

Natural language processing technology;

Support platforms of the aforementioned (Autonomous Unmanned System Support

Platforms, AI Basic Data and Security Detection Platforms, etc.)

Context

National Strategy: China

Date of

publication/

release

20 July 2017

Comments

China shows a significant interest in the foreign AI developments and among the

conclusions of the strategy is the focus that is set on achieving world-leading levels

in AI and reduce foreign dependence [China State Council. Made in China 2025].

“AI has become a new focus of international competition. AI is a strategic technology

that will lead in the future; the world’s major developed countries are taking the

development of AI as a major strategy to enhance national competitiveness and

protect national security…”

“…by 2030, China’s AI theories, technologies, and applications should achieve

worldleading levels, making China the world’s primary AI innovation center…”

Budget:

“…the intelligent application of a complete industrial chain and high-end industrial

clusters, with AI core industry scale exceeding 1 trillion RMB, and with the scale of

related industries exceeding 10 trillion RMB.”

3.1.3.9 AI National Strategy: Canada, 2017

Source

Pan-Canadian Artificial Intelligence Strategy

Text of the

definition

Subdomains

Context

National strategy: Canada.

It has four goals:

1. to increase the number of AI researchers and graduates in Canada,

2. to form three AI centres of scientific excellence (Alberta Machine Intelligence

Institute (AMII, Edmonton), Vector Institute (Toronto), Mila (Montreal)),

3. to develop thought leadership on the economic, ethical, policy, and legal

implications of AI,

4. to support Canada’s AI research community.

Part of the strategy is the collaboration of the Canadian Institute for Advanced

Research (CIFAR), which leads the strategy, with the Canadian government and the

three new AI centres of scientific excellence.

Date of

publication/

release

2017

Comments

Canada was the first country that released an AI national strategy.

Budget: $125-million investment in AI research and innovation in Canada.

On 7

of June 2018 Canada and France published a joint statement on AI. The

announcement included their common aim, namely to encourage the development of

AI while anticipating any impacts with coordinated efforts. The materialisation of this

aim would be an international study group consisted of internationally recognised

experts in science, industry and civil society, together with policymakers. It is set to

identify opportunities and challenges ensuing from AI, and provide an inclusive

mechanism “for sharing multidisciplinary analysis, foresight and coordination

capabilities in the area of artificial intelligence”.

3.1.4 International Organisations

3.1.4.1 OECD, 2019

Source

OECD, Recommendation of the Council on Artificial Intelligence, OECD/LEGAL/0449

Text of the

definition

"An AI system is a machine-based system that can, for a given set of human-defined

objectives, make predictions, recommendations, or decisions influencing real or virtual

environments. AI systems are designed to operate with varying levels of autonomy."

Subdomains

Ethics: inclusive growth, sustainable development and well-being; human-centered

values and fairness; transparency and explainability; robustness, security and safety;

accountability

Context

Under the OECD Legal instruments, this document presents a number of

recommendations to promote innovation on AI based on ethical principles and

respecting human rights and democratic values.

Date of

publication/

release

22 May 2019

Comments

3.1.4.2 UNESCO, 2019

Source

UNESCO. Principles for AI: Towards a Humanistic Approach? A Global Conference

Text of the

definition

Subdomains

Rapid technological advancements in artificial intelligence (AI) – as well as other

evolving technologies such as robotics, big data analytics, and the Internet of Things

– are changing the way we learn, work and live together.

Context

Conference on AI principles

Date of

publication/

release

04 March 2019

Comments

Presently no definition is found reported by UNESCO.

3.1.4.3 StandICT.eu project, 2019

Source

Supporting European Experts Presence in International Standardisation Activities in

ICT (StandICT.eu). ICT standards and ongoing work at International level in the AI field

– a Landscape analysis

Text of the

definition

Subdomains

Themes/challenges/areas:

personalised AI; trustworthiness; ethics; AI security; transparency of autonomous

systems; AI usage; wellbeing metrics; big data; AI foundational standards; AI

governance; computational approaches; health; transparency of data processing;

conceptualisation and specification of domain knowledge

Context

Project funded by H2020 for ICT standardisation, ICT Technical specifications, cloud

computing, 5G communications, IoT, cybersecurity, data technologies.

Aim: description of the ICT standards, ongoing work at international level and

landscape analysis.

Date of

publication/

release

24 February 2019

Comments

“End of 2018 two sub committees” (JTC1 SC42, JTC1 SC27 WG4) “with 6 working

groups” (JTC1 SC42 JWG1, JTC1 SC42 WG1, JTC1 SC42 WG2, JTC1 SC42 WG3, JTC1

SC42 WG4, JTC1 SC42 WG5) “and 1 study group” (JTC1 SC42 Study Group 1) “with

the goal to develop 10 AI standards are active in ISO/IEC.”

More details on each expected standard: p. 39.

3 standards are published (2 are stated in the report):

1. ISO/IEC 20546:2019 Information technology -- Big data -- Overview and

vocabulary. ISO/IEC JTC1/SC42 working groups for standardisation in the area of AI

published on 28.02.2019 the ISO/IEC 20546:2019. It requires a fee to be

downloaded, however from the preview it can be seen that it is an overview and

vocabulary only on “Information technology – Big data”, without any mention to AI

definitions.

2. ISO/IEC TR 20547-2:2018 Information technology -- Big data reference

architecture -- Part 2: Use cases and derived requirements

Until 03.05.2019 the ISO/IEC WD 22989 on Artificial Intelligence -- Concepts and

terminology Standard and/or project under the direct responsibility of ISO/IEC

JTC1/SC42 is reported “under development”, and more specifically in the

“Preparatory” phase.

3. ISO/IEC TR 20547-5:2018 Information technology -- Big data reference

architecture -- Part 5: Standards roadmap

ISO/IEC JTC1/SC42 is the first international standards committee looking at the entire

AI ecosystem. JTC1’s scope for SC42 is to become a systems integration entity to

work with other ISO, IEC and JTC 1 committees looking at AI applications.

Among the reported community and industrial activities are mentioned (some

involved in ISO/IEC JTC1/SC42):

- the multi-stakeholder platform on ICT standardisation (MSP)

- European AI Alliance steered by High-Level Expert Group on AI (AI HLEG)

- Fraunhofer Cluster of Excellence “Cognitive Internet Technologies” (CCIT)

- Big Fata Value Association (BDVA)

More projects are mentioned for the development of standards on other aspects of AI

(e.g. P7006 - Standard for Personal Data Artificial Intelligence (AI) Agent, P7008

Standard for Ethically Driven Nudging for Robotic, Intelligent and Autonomous

Systems, P7010 - Wellbeing Metrics Standard for Ethical Artificial Intelligence and

Autonomous Systems, IEEE Ethically Aligned Design version 2, et. al.)

3.1.4.4 OECD, 2018

Source

OECD Directorate for Science, Technology and Innovation, Committee on Industry,

Innovation and Entrepreneurship. Identifying and Measuring Developments in Artificial

Intelligence. DSTI/CIIE/ WPIA(2018)4

Text of the

definition

“AI is neither science fiction nor a science project. There was universal agreement that

artificial intelligence already provides beneficial applications that are used every day

by people worldwide. Going forward, conference participants suggested that the

development and uses of AI systems should be guided by principles that will promote

well-being and prosperity while protecting individual rights and democracy.

A consensus emerged that the fast-paced and far-reaching changes from AI offer

dynamic opportunities for improving the economic and social sectors. AI can make

business more productive, improve government efficiency and relieve workers of

mundane tasks. It can also address many of our most pressing global problems, such

as climate change and wider access to quality education and healthcare.

…

This combination of interdisciplinary origins, wavering trajectories, and recent

commercial success make "artificial intelligence" a difficult concept to define and

measure.

The term itself is used interchangeably both as the still-faraway goal of true machine

intelligence and as the currently available technology powering today’s hottest

startups” (p.5)

Subdomains

Machine learning (including deep learning); statistics, mathematics and computational

methods; specific fields and applications such as: text mining; image recognition;

biology machine vision; speech recognition; machine translation (weak AI or Artificial

Narrow Intelligence) (pp.4-5)

Context

Policy Document that proposes an approach to identify and measure AI developments

in science, technological developments, and software.

Date of

publication/

release

12 October 2018

Comments

Methods used: topic modelling to subdivide AI-codes, and find key development fields

and applications.

Among other sources used: GitHub, patents, Scopus.

3.1.4.5 ETSI, 2018

Source

ETSI GR ENI 004 v.1.1.1. Experiential Network Intelligence (ENI); Terminology for Main

Concepts in ENI

Text of the

definition

“Computerized system that uses cognition to understand information and solve

problems.”

NOTE 1: ISO/IEC 2382-28 "Information technology -- Vocabulary" defines AI as "an

interdisciplinary field, usually regarded as a branch of computer science, dealing with

models and systems for the performance of functions generally associated with

human intelligence, such as reasoning and learning".

NOTE 2: In computer science AI research is defined as the study of "intelligent

agents": any device that perceives its environment and takes actions to achieve its

goals.

NOTE 3: This includes pattern recognition and the application of machine learning and

related techniques.

NOTE 4: Artificial Intelligence is the whole idea and concepts of machines being able

to carry out tasks in a way that mimics the human intelligence and would be

considered "smart".

Subdomains

Mention of only two fields:

“Knowledge reasoning: field of artificial intelligence that uses a set of knowledge

bases and a given knowledge representation to reason about the information

available

NOTE: Typically, this is used to validate data as well as predict or infer new

information from existing information.

Knowledge representation: field of artificial intelligence that represents data and

information in a form that a computerized system can use.”

Context

The Experiential Networked Intelligence (ENI) ETSI Industry Specification Group (ISG)

of the European Telecommunications Standards Institute (ETSI) published a document

on the main concepts in ENI.

Date of

publication/

release

June 2018

Comments

European Telecommunications Standards Institute (ETSI) is the recognized regional

standards body addressing telecommunications, broadcasting and other electronic

communications networks and services. It is a not-for-profit organization, part of the

European Standards Organization (ESO).

It uses the ISO/IEC 2382-28 AI definitions, and works on the standardised use of AI

applications.

3.1.4.6 OECD, 2017

Source

OECD. Science, Technology and Industry Scoreboard 2017. The Digital

Transformation.

Text of the

definition

“Artificial Intelligence (AI) is a term used to describe machines performing human-like

cognitive functions (e.g. learning, understanding, reasoning or interacting). It has the

potential to revolutionise production as well as contribute to tackling global

challenges related to health, transport and the environment.”

Subdomains

Technologies that embed AI; large capacity analysis and storage; information

communication devices; mobile communication; imaging and sound technology; ICT

security; measurement; high-speed computing and network; medical technology

Context

Policy document with indicators regarding the impact of digital transformation on

science, innovation, the economy, work and society.

Date of

publication/

release

2017

Comments

Global rankings and technological map are available in the report.

3.1.4.7 World Economic Forum, 2017

Source

World Economic Forum

WEF. 2017. Impact of the Fourth Industrial Revolution on Supply Chains.

Text of the

definition

“Artificial intelligence (AI) is the software engine that drives the Fourth Industrial

Revolution. Its impact can already be seen in homes, businesses and political

processes. In its embodied form of robots, it will soon be driving cars, stocking

warehouses and caring for the young and elderly. It holds the promise of solving

some of the most pressing issues facing society, but also presents challenges such as

inscrutable “black box” algorithms, unethical use of data and potential job

displacement. As rapid advances in machine learning (ML) increase the scope and

scale of AI’s deployment across all aspects of daily life, and as the technology itself

can learn and change on its own, multistakeholder collaboration is required to

optimize accountability, transparency, privacy and impartiality to create trust.”

“Artificial intelligence (AI) or self-learning systems is the collective term for machines

that replicate the cognitive abilities of human beings. Within the broader

technological landscape, predictive maintenance in the cognitive era has the potential

to transform global production systems.”

Subdomains

Context

Policy conference and white paper on how production and supply chain will be

affected by new technological developments, including AI.

Date of

publication/

release

2017

Comments

Prepared in collaboration with the German logistics association BVL International.

3.1.4.8 ISO, 1993; 1995; 2015

Source

ISO/IEC 2382:2015

Text of the

definition

“Branch of computer science devoted to developing data processing systems that

perform functions normally associated with human intelligence, such as reasoning,

learning, and self-improvement” (2121393: ISO, Al: term, abbreviation and definition

standardized by ISO/IEC [ISO/IEC 2382-1:1993])

“Interdisciplinary field, usually regarded as a branch of computer science, dealing with

models and systems for the performance of functions generally associated with

human intelligence, such as reasoning and learning” (2123769: term, abbreviation

and definition standardized by ISO/IEC [ISO/IEC 2382-28:1995])

“Capability of a functional unit to perform functions that are generally associated

with human intelligence such as reasoning and learning” (2123770: term,

abbreviation and definition standardized by ISO/IEC [ISO/IEC 2382-28:1995])

Subdomains

Context

International Organization for Standardization (ISO)

Date of

publication/

release

2015

Comments

The definitions imply the “general” AI classification; they refer to performance of

human functions: reasoning, learning etc.

3.2 Research perspective

3.2.1 Tsinghua University, 2018

Source

China Institute for Science and Technology Policy at Tsinghua University. AI

Development Report.

Text of the

definition

“AI machines do not necessarily have to obtain intelligence by thinking like a human

and that it is important to make AI solve problems that can be solved by a human

brain. Brain science and brainlike intelligence research and machine-learning

represented by deep neural networks represent the two main development directions

of core AI technologies, with the latter referring to the use of specific algorithms to

direct computer systems to arrive at an appropriate model based on existing data

and use the model to make judgment on new situations, thus completing a behavior

mechanism.

While only limited progress has been made in the first direction, tremendous strides

have been taken in the second direction so much that machine learning has not only

become the main paradigm of AI technology but been equated by some with AI

itself. In general, the artificial intelligence we know today is based on modern

algorithms, supported by historical data, and forms artificial programs or systems

capable of perception, cognition, decision making and implementation like humans.”

Subdomains

Technical Dimensions of AI Enterprise Identification:

- Speech: speech recognition, speech synthesis, speech interaction, speech

evaluation, human-machine dialogue, voiceprint recognition

- Vision: biometrics (face recognition, iris recognition, fingerprint recognition, vein

recognition, etc.) affective computing, emotion recognition, expression recognition,

behavior recognition, gesture recognition, body recognition, video content recognition,

object and scene recognition, mobile vision, optical character recognition (OCR),

handwriting recognition, SLAM, spatial recognition, 3D reconstruction etc.

- Natural Language Processing: natural language interaction, natural language

understanding, semantic understanding, machine translation, text mining (semantic

analysis, semantic computing, classification, clustering), information extraction,

human-machine interaction

- Basic algorithm and platform: machine learning, deep learning, open source

framework, open platform

- Basic hardware: chips, lidars, sensors, etc.

- Basic enabling technology: cloud computing, big data

Product and Industry Dimensions:

- Intelligent robotics: industrial robotics, service robotics, personal/ home robotics

- Smart driving: Intelligent driving, driverless driving, autonomous driving, assisted

driving, advanced driver assistance system (ADAS), laser radar, ultrasonic radar,

millimetre wave radar, GPS positioning, high-precision map, vehicle chip, human-car

interaction, etc.

- Drone: consumer drones, professional drones

- AI+: Finance, insurance, judiciary administration, entertainment, tourism, healthcare,

education, logistics and warehousing, smart home, smart city, network security, video

surveillance, commerce, human resources, corporate services

Context

The report captures in multiple dimensions the Chinese and worldwide AI ecosystem.

It was firstly presented during the World Peace Forum.

Date of

publication/ release

July 2018

Comments

Budget:

150 billion dollars by 2030, more than 50 billion euros in AI research by 2025

3.2.2 Kaplan and Haenlein, 2018

Source

Kaplan, A. and Haenlein, M. Siri, Siri, in my hand: Who’s the fairest in the land? On the

interpretations, illustrations, and implications of artificial intelligence

Text of the

definition

“Artificial intelligence (AI)—defined as a system’s ability to correctly interpret external

data, to learn from such data, and to use those learnings to achieve specific goals

and tasks through flexible adaptation.”

Subdomains

Context

The article states that AI is different from concepts as IoT and big data.

It introduces in the definition the notions of interpretation of the environment

(external data), learning, achievement of goals/tasks etc.

Refers to AI through stages: artificial narrow/general/super intelligence.

Date of

publication/

release

2018

Comments

3.2.3 Poole et al., 2017; 2010; 1998

Source

Poole, D., Mackworth, A., and Goebel, R. (1998). Computational Intelligence: A Logical

Approach. Oxford University Press, New York.

Poole, D., Mackworth. A. (2010). Artificial Intelligence Foundations of Computer Agents

Poole, D., Mackworth A. (2017). Artificial Intelligence: Foundations of Computational

Agents, second edition

Text of the

definition

1998: “Artificial intelligence (AI) is the established name for the field we have defined

as computational intelligence (CI), Computational intelligence is the study of the

design of intelligent agents. An agent is something that acts in an environment—it

does something. Agents include worms, dogs, thermostats, airplanes, humans,

organizations, and society. An intelligent agent is a system that acts intelligently:

What it does is appropriate for its circumstances and its goal, it is flexible to changing

environments and changing goals, it learns from experience, and it makes appropriate

choices given perceptual limitations and finite computation.”

2010, 2017: “Artificial intelligence, or AI, is the field that studies the synthesis and

analysis of computational agents that act intelligently.

An agent is something that acts in an environment; it does something. Agents include

worms, dogs, thermostats, airplanes, robots, humans, companies, and countries.”

“We are interested in what an agent does; that is, how it acts. We judge an agent by

its actions…

An agent acts intelligently when:

• what it does is appropriate for its circumstances and its goals, taking into account

the short-term and long-term consequences of its actions

• it is flexible to changing environments and changing goals

• it learns from experience

• it makes appropriate choices given its perceptual and computational limitations”

“A computational agent is an agent whose decisions about its actions can be

explained in terms of computation. That is, the decision can be broken down into

primitive operations that can be implemented in a physical device. This computation

can take many forms. In humans this computation is carried out in “wetware”; in

computers it is carried out in “hardware.” Although there are some agents that are

arguably not computational, such as the wind and rain eroding a landscape, it is an

open question whether all intelligent agents are computational.

All agents are limited. No agents are omniscient or omnipotent. Agents can only

observe everything about the world in very specialized domains, where “the world” is

very constrained. Agents have finite memory. Agents in the real world do not have

unlimited time to act.”

The central scientific goal of AI “is to understand the principles that make intelligent

behavior possible in natural or artificial systems. This is done by:

• the analysis of natural and artificial agents

• formulating and testing hypotheses about what it takes to construct intelligent

agents and

• designing, building, and experimenting with computational systems that perform

tasks commonly viewed as requiring intelligence.

As part of science, researchers build empirical systems to test hypotheses or to

explore the space of possible designs. These are quite distinct from applications that

are built to be useful for an application domain.”

“The definition is not for intelligent thought alone. We are only interested in thinking

intelligently insofar as it leads to more intelligent behavior. The role of thought is to

affect action.”

“The central engineering goal of AI is the design and synthesis of useful, intelligent

artifacts. We actually want to build agents that act intelligently. Such agents are

useful in many applications.”

Subdomains

Context

Book

Date of

publication/

release

2017; 2010; 1998

Comments

Endorsed by McCarthy (among the founders of AI in McCarthy, J. What is Artificial

Intelligence. (2007))

It equals AI to computational intelligence, and it is focused on the definitions of

agents, actions, reaction to the environment, learning....

3.2.4 Kaplan, 2016

Source

Kaplan, J. Artificial Intelligence What everyone needs to know.

Text of the

definition

“There is little agreement about what intelligence is. …there is scant reason to believe

that machine intelligence bears much relationship to human intelligence, at least so

far.”

“There are many proposed definitions on AI …most are roughly aligned around the

concept of creating computer programs or machines capable of behavior we would

regard as intelligent if exhibited by humans.”

He suggests that McCarthy's definition, although sensible, is deeply flawed [section 1

Defining AI, p.1], as it is difficult to define and/or measure human intelligence.

“Our cultural predilection for reducing things to numeric measurements that facilitate

direct comparison often creates a false patina of objectivity and precision.”

Subdomains

Context

The book offers a definition of AI based on the juxtaposition between human and

computer intelligence. It is highlighted that the mono-dimensional quantification of

human intelligence and other simplified approaches to define AI are inadequate.

Date of

publication/

release

2016

Comments

The author is a Lecturer and Research Affiliate at Stanford University.

To disentangle the oversimplification of intelligence's quantification, a proposal was

made by a cognitive scientist [Gardner H., 1999] to approach intelligence in eight

dimensions.

3.2.5 Stone et al.: AI100, 2016

Source

Stone, P., Brooks, R., Brynjolfsson, E., Calo, R., Etzioni, O., Hager, G., Hirschberg, J.,

Kalyanakrishnan, S., Kamar, E., Kraus, S., Leyton-Brown, K., Parkes, D., Press, W.,

Saxenian, A.L, Shah, J., Tambe, M., and Teller, A. Artificial Intelligence and Life in 2030.

One Hundred Year Study on Artificial Intelligence: Report of the 2015-2016 Study

Panel, Stanford University, Stanford, CA.

Text of the

definition

““Intelligence” remains a complex phenomenon whose varied aspects have attracted

the attention of several different fields of study, including psychology, economics,

neuroscience, biology, engineering, statistics, and linguistics. Naturally, the field of AI

has benefited from the progress made by all of these allied fields. For example, the

artificial neural network, which has been at the heart of several AI-based solutions

[1,2]

was originally inspired by thoughts about the flow of information in biological

neurons

[3]

.”

[1] Gerald Tesauro, “Practical Issues in Temporal Difference Learning,” Machine

Learning, no. 8 (1992): 257—77.

[2] David Silver, Aja Huang, Chris J. Maddison, Arthur Guez, Laurent Sifre, George van

den Driessche, Julian Schrittwieser, Ioannis Antonoglou, Veda Panneershelvam, Marc

Lanctot, Sander Dieleman, Dominik Grewe, John Nham, Nal Kalchbrenner, Ilya

Sutskever, Timothy Lillicrap, Madeleine Leach, Koray Kavukcuoglu, Thore Graepel, and

Demis Hassabis, “Mastering the game of Go with deep neural networks and tree

search,” Nature 529 (2016): 484—489.

[3] W. McCulloch and W. Pitts, W., “A logical calculus of the ideas immanent in nervous

activity,” Bulletin of Mathematical Biophysics, 5 (1943): 115—133.

Subdomains

Trends: large scale machine learning, deep learning, reinforcement learning, robotics,

computer vision, natural language processing, collaborative systems, crowdsourcing

and human computation, algorithmic game theory and computational social choice,

IoT, neuromorphic computing.

Applications in domains: transportation, home service robots, healthcare, education,

low-resource communities, public safety and security, employment and workplace,

entertainment

Context

Investigation of the AI field, started publishing periodic reports in 2014. Analysis of AI

impact on “people, their communities and society”, in view of other fields that can

affect the AI evolution (science, engineering, computing systems).

Date of

publication/

release

September 2016

Comments

Policy projections in the report available.

There is not a definition per se, but a reference to different disciplines interested in

and interrelated with AI.

3.2.6 Russel and Norvig, 2010 (3rd edition); 1995

Source

Russel, S. and Norvig, P. Artificial Intelligence. A Modern Approach.

Text of the

definition

In Figure 1.1 of the book eight definitions are mentioned:

“Eight definitions of AI, laid out along two dimensions. The definitions on top are

concerned with thought processes and reasoning, whereas the ones on the bottom

address behavior. The definitions on the left measure success in terms of fidelity to

human performance, whereas the ones on the right measure against an ideal

performance measure, called rationality. A system is rational if it does the “right

thing,” given what it knows.

Historically, all four approaches to AI have been followed, each by different people

with different methods. A human-centered approach must be in part an empirical

science, involving observations and hypotheses about human behavior. A rationalist

approach involves a combination of mathematics and engineering. The various

groups have both disparaged and helped each other. Let us look at the four

approaches in more detail.”

Subdomains

Context

Leading book in the AI field. Introduces the idea of a human-centered approach to AI

versus a pragmatic computational approach.

Date of

publication/

release

1995

2010 (3rd edition http://aima.cs.berkeley.edu/)

Comments

Endorsed by McCarthy (among the founders of AI in McCarthy, J. What is Artificial

Intelligence. (2007))

Norvig is an AI leading researcher, Director of Research at Google Inc. He is also an

AAAI Fellow and councillor of the Association for the Advancement of Artificial

Intelligence. He was head of the Computational Sciences Division (now the Intelligent

Systems Division) at NASA Ames Research Center, for research and development in

the areas of autonomy and robotics, automated software engineering and data

analysis, neuro-engineering, collaborative systems research, and simulation-based

decision-making.

3.2.7 Bruner, 2009

Source

Bruner J. Culture, Mind and Education. Contemporary theories of learning.

Text of the

definition

“…any and all systems that process information must be governed by specifiable

"rules" or procedures that govern what to do with inputs. It matters not whether it is a

nervous system, or the genetic apparatus that takes instruction from DNA and then

reproduces later generations, or whatever. This is the ideal of artificial intelligence

(AI), so-called.”

Subdomains

Context

The book chapter offers an AI definition relating it to human intelligence. This

definition includes the notion of having rules or procedures leading to decisions.

Date of

publication/

release

2009

Comments

3.2.8 McCarthy, 2007

Source

McCarthy, J. What is Artificial Intelligence.

Text of the

definition

“It is the science and engineering of making intelligent machines, especially intelligent

computer programs. It is related to the similar task of using computers to understand

human intelligence, but AI does not have to confine itself to methods that are

biologically observable.”

“Intelligence is the computational part of the ability to achieve goals in the world.

Varying kinds and degrees of intelligence occur in people, many animals and some

machines.”

Subdomains

Branches:

logical AI; search; pattern recognition; representation; inference; common sense

knowledge and reasoning; learning from experience; planning; epistemology; ontology;

heuristics; genetic programming

Applications: game playing; speech recognition; understanding natural language;

expert systems; heuristic classification

Context

The article uses the notion of the achievement of goals. Refers to different kinds of

intelligence. Implicit reference to general AI / strong AI.

Date of

publication/

release

2007

Comments

McCarthy is among the founding fathers of AI.

3.2.9 Gardner, 1999

Source

Gardner H. Intelligence Reframed: Multiple Intelligences for the 21st Century, pp.33-

Text of the

definition

“A biopsychological potential to process information that can be activated in a

cultural setting to solve problems or create products that are of value in a culture.”

Subdomains

Context

Book revisiting the multiple human intelligences.

Date of

publication/

release

1999

Comments

This book is the revision of the 1983 book.

3.2.10 Nakashima, 1999

Source

H. Nakashima. AI as complex information processing. Minds and machines, 9:57–80.

Text of the

definition

“Intelligence is the ability to process information properly in a complex environment.

The criteria of properness are not predefined and hence not available beforehand.

They are acquired as a result of the information processing.”

Subdomains

Context

The article presents a definition that includes the notions of information processing

and complex environment.

Date of

publication/

release

1999

Comments

3.2.11 Nilsson, 1998

Source

Nilsson, N.J. Artificial intelligence: a new synthesis. Morgan Kaufmann Publishers, Inc.

Text of the

definition

“Artificial Intelligence (AI), broadly (and somewhat circularly) defined, is concerned

with intelligent behavior in artifacts. Intelligent behavior, in turn, involves perception,

reasoning, learning, communicating, and acting in complex environments.”

Subdomains

Context

The book introduces in the definition the notions of complex environment, reasoning,

learning, communicating etc.

Date of

publication/

release

1998

Comments

“AI has as one of its long-term goals the development of machines that can do these

things as well as humans can, or possibly, even better. Another goal of AI is to

understand this kind of behavior whether it occurs in machines or in humans or other

animals.”

Endorsed by McCarthy (among the founders of AI in McCarthy, J. What is Artificial

Intelligence. (2007)).

3.2.12 Neisser et al., 1996

Source

Neisser U., Boodoo G., Bouchard T.J., Boykin A.W., Brody N., Ceci S.J., Halpern D.F.,

Loehlin J.C., Perloff R., Sternberg R.J., and Urbina S. Intelligence: Knows and Unknowns

Text of the

definition

On human Intelligence:

“Individuals differ from one another in their ability to understand complex ideas, to

adapt effectively to the environment, to learn from experience, to engage in various

forms of reasoning, to overcome obstacles by taking thought.

Concepts of intelligence are attempts to clarify and organise this complex set of

phenomena.”

Subdomains

Context

The article introduces in the AI definition the notions of adapting to the environment,

reasoning, learning etc. A human intelligence definition is used to approach AI, due to

biologically inspired processes. Multiple intelligences approach.

Date of

publication/

release

1996

Comments

Cited among others by Yang, 2013

3.2.13 Fogel, 1995

Source

D. B. Fogel. Review of computational intelligence: Imitating life. Proc. of the IEEE,

83(11).

Evolutionary Computation: Toward a New Philosophy of Machine Intelligence.

Text of the

definition

“Any system…that generates adaptive behaviour to meet goals in a range of

environments can be said to be intelligent.”

Subdomains

Context

The article includes the notions of adaptive behaviour, environment, and achieving

goals.

Date of

publication/

release

1995

Comments

He is a pioneer in evolutionary computation.

He is currently Chief Scientist at Trials.ai, and holds other founding positions at

Natural Selection, Inc., Color Butler, Inc., and Effect Technologies, Inc., the maker of

the patented EffectCheck sentiment analysis software tool. Advisor for several AI

companies in the areas of B2B lead generation, logistics, and employee retention, as

well as other areas.

3.2.14 Wang, 1995

Source

Wang P. On the working definition of intelligence. Center for Research on Concepts

and Cognition, Indiana University.

Text of the

definition

Intelligence is “the ability for an information processing system to adapt to its

environment with insufficient knowledge and resources.”

Subdomains

Context

Technical Report.

The definition includes the notions of information processing, adaptation to the

environment, and insufficiency of knowledge/resources.

Date of

publication/

release

1995

Comments

3.2.15 Albus, 1991

Source

J. S. Albus. Outline for a theory of intelligence. IEEE Trans. Systems, Man and

Cybernetics, 21(3):473–509.

Text of the

definition

“…the ability of a system to act appropriately in an uncertain environment, where

appropriate action is that which increases the probability of success, and success is

the achievement of behavioral subgoals that support the system’s ultimate goal.”

Subdomains

Context

article

Date of

publication/

release

1991

Comments

From Wikipedia:

He was an American engineer, Senior NIST Fellow and founder and former chief of

the Intelligent Systems Division of the Manufacturing Engineering Laboratory at the

National Institute of Standards and Technology (NIST). Albus made contributions to

cerebellar robotics, developed a two-handed manipulator system known as the

Robocrane (a crane-like variation on the Stewart platform idea), among other

contributions.

The definition includes the notions of environment, actions and achieving goals.

3.2.16 Schank, 1991; 1987

Source

Schank R.C. What is AI, Anyway? AI Magazine, 8 (4), aaai.org

R. Schank. Where’s the AI? AI magazine, 12(4):38–49, 1991

Text of the

definition

“AI suffers from a lack of definition of its scope. One way to attack this problem is to

attempt to list some features that we would expect an intelligent entity to have. None

of these features would define intelligence, indeed a being could lack any one of

them and still be considered intelligent. Nevertheless each attribute would be an

integral part of intelligence in its way. ...They are communication, internal knowledge,

world knowledge, intentionality, and creativity.”

“AI's primary goal is to build an intelligent machine. The second goal is to find out

about the nature of intelligence.”

“Intelligence means getting better over time.”

Subdomains

Context

article

Date of

publication/

release

1987

Comments

Roger Carl Schank is an American artificial intelligence theorist, cognitive

psychologist, learning scientist, educational reformer, and entrepreneur. Beginning in

the late 1960s, he pioneered conceptual dependency theory and case-based

reasoning, both of which challenged cognitivist views of memory and reasoning.

3.2.17 McCarthy, 1988

Source

McCarthy, J. The Logic and Philosophy of Artificial intelligence

Text of the

definition

“The goal of artificial intelligence (A.I.) is machines more capable than humans at

solving problems and achieving goals requiring intelligence. There has been some

useful success, but the ultimate goal still requires major conceptual advances and is

probably far off.

There are three ways of attacking the goal. The first is to imitate the human nervous

system. The second is to study the psychology of human intelligence. The third is to

understand the common sense world in which people achieve their goals and develop

intelligent computer programs. This last one is the computer science approach.”

Subdomains

Context

Date of

publication/

release

1988

Comments

McCarthy is among the founding fathers of AI.

3.2.18 Gardner, 1987

Source

Gardner, H. The mind's new science: A history of the cognitive revolution. Basic books.

Text of the

definition

AI “seeks to produce, on a computer, a pattern of output that would be considered

intelligent if displayed by a human being”.

Schlinger (1992) mentions that this book also refers that “AI is viewed as a way of

testing a particular theory of how cognitive processes might work. That theory is the

popular information-processing model of cognition. Where AI researchers disagree,

according to Gardner, is how literally to interpret the thinking metaphor. For example,

some take what John Searle calls the "weak view" of AI, wherein computer programs

are simply a means for testing theories of how humans might carry out cognitive

operations. The weak view of AI is synonymous with modern cognitive psychology.”

Subdomains

Context

book

Date of

publication/

release

1987

Comments

3.2.19 Gardner, 1983

Source

Gardner, H. Frames of Mind; The Theory of Multiple Intelligences. New York, NY: basic

Books.

Text of the

definition

Artificial intelligence is commonly defined by referencing definitions of human

intelligence, as in Minsky’s definition.

In contrast to the standard approach of measuring one kind of intelligence (as in

standard IQ tests), Gardner (cognitive scientist) offers an eight-dimensional definition

to disentangle the oversimplification of intelligence's measurement.

In particular, he proposed multiple conceptions of intelligence, not only logical-

mathematical, linguistic, but also spatial, musical, bodily-kinaesthetic, personal.

Subdomains

Context

book

Date of

publication/

release

1983

Comments

This definition of intelligence is more used to approximate the definition of AI in

terms of aim and processes.

Gardner is a cognitive developmental psychologist, among the pioneers trying to

quantify human intelligence in more than one dimension (another is Robert

Sternberg), introducing the notion of multiple intelligences.

Before his study, human intelligence was mono-semantic and was quantified as such

in intelligence quotient (IQ) points.

The multiple intelligences approach is a better fit to the oversimplification of one

intelligence, and is used to describe why the definition of AI is not easy. (see J. Kaplan

2016, Artificial Intelligence What everyone needs to know, section 1 Defining AI)

3.2.20 Newell and Simon, 1976

Source

Newell, A., Simon, H. A. Computer science as empirical enquiry: Symbols and search.

Communications of the ACM 19, 3:113–126.

Text of the

definition

“By “general intelligent action” we wish to indicate the same scope if intelligence as

we see in human action: that in any real situation behavior appropriate to the ends of

the system and adaptive to the demands of the environment can occur, within some

limits of speed and complexity.”

Subdomains

Context

article

Date of

publication/

release

1976

Comments

Simon was a pioneer in the field of artificial intelligence, creating with A. Newell the

Logic Theory Machine (1956) and the General Problem Solver (GPS) (1959) systems.

The GPS system is considered as the first knowledge representation approach [Newell

and Simon, 1961].

Newell was a researcher in computer science and cognitive psychology at the RAND

Corporation and at Carnegie Mellon University’s School of Computer Science, Tepper

School of Business, and Department of Psychology.

They founded an artificial intelligence laboratory at Carnegie Mellon University and

produced a series of important programs and theoretical insights throughout the late

fifties and sixties.

The definition includes the notions of real situation, goal (ends of the system),

adaptation to the environment, and complexity.

3.2.21 Minsky, 1969

Source

Minsky, M. L. Semantic information processing. Cambridge, MA: MIT Press

Text of the

definition

AI is “the science of making machines do things that would require intelligence if

done by men”.

Subdomains

Context

PhD Thesis of one of the first cognitive scientists approaching AI as human

intelligence.

Date of

publication/

release

1969

Comments

Marvin Minsky was Toshiba Professor of Media Arts and Sciences and Donner

Professor of Electrical Engineering and Computer Science at MIT.

He was a cofounder of the MIT Media Lab and a consultant for the One Laptop Per

Child project.

Definition based on general intelligence.

3.2.22 McCarthy, 1959

Source

McCarthy, J. Programs with Common Sense.

Text of the

definition

Proposes that common sense reasoning ability is key to AI.

“A program has common sense if it automatically deduces for itself a sufficiently

wide class of immediate consequences of anything it is told and what it already

knows.”

Subdomains

Context

Date of

publication/

release

1959

Comments

“Probably the first paper on logical AI, i.e. AI in which logic is the method of

representing information in computer memory and not just the subject matter of the

program. It may also be the first paper to propose common sense reasoning ability as

the key to AI.”

McCarthy is among the founding fathers of AI and it is cited as the one who coined

the term “artificial intelligence”.

AI used for deductive reasoning.

3.2.23 McCarthy et al., 1955

Source

McCarthy, J., Minsky, M. L., Rochester, N., Shannon, C.E. A Proposal For The Dartmouth

Summer Research Project On Artificial Intelligence

Text of the

definition

“..every aspect of learning or any other feature of intelligence can in principle be so

precisely described that a machine can be made to simulate it. An attempt will be

made to find how to make machines use language, form abstractions and concepts,

solve kinds of problems now reserved for humans, and improve themselves.

…the artificial intelligence problem is taken to be that of making a machine behave in

ways that would be called intelligent if a human were so behaving.”

Subdomains

Context

Founding proposal and conference for initiation of AI studies

Date of

publication/

release

31 August 1955

Comments

Founding fathers and conference of AI.

AI as a machine that does what humans do (strong AI concept)

3.3 Market perspective

3.3.1 CB Insights, 2019

Source

CB Insights. Artificial Intelligence Trends

Text of the

definition

Subdomains

AI trends are reported:

conversational agents, cyber threat hunting, drug discovery, predictive maintenance,

e-commerce search, medical imaging & diagnostics, edge computing, facial

recognition, open source frameworks, synthetic training data, back office automation,

language translation, anti-counterfeit, check-out free retail, auto claims processing,

advanced healthcare biometrics, clinical trial enrolment, next-gen prosthetics, capsule

networks, GANs, federated learning, network optimization, reinforcement learning,

autonomous navigation, crop monitoring,

with the following applications:

computer vision, natural language processing/synthesis, predictive intelligence,

architecture, infrastructure

Context

market report

Date of

publication/

release

2019

Comments

Clustering method for figure p.3 is not extensively presented.

The trends are reported using the CB Insights NexTT framework, which is explained

as:

INDUSTRY ADOPTION (y-axis): Signals include momentum of startups in the space,

media attention, customer adoption (partnerships, customer, licensing deals).

MARKET STRENGTH (x-axis): Signals include market sizing forecasts, quality and

number of investors and capital, investments in R&D, earnings transcript

commentary, competitive intensity, incumbent deal making (M&A, strategic

investments).

3.3.2 Statista, 2017

Source

Statista Report. Artificial Intelligence

Text of the

definition

“Artificial Intelligence (AI) essentially refers to computing technologies that are

inspired by the ways people use their brains and nervous systems to reason and

make decisions, but typically operate quite differently.”

Subdomains

Applications:

Automotive (autonomous driving, cloud computing).

Healthcare (early diagnosis and preventing healthcare, surgical assistance, recovery

and rehabilitation, drug discovery, precision medicine and personal genetics,

healthcare robotics: direct patient care robots(surgical robots, exoskeletons,

prosthetics), indirect patient care robots(pharmacy, delivery, disinfection), home

healthcare robots).

Education (intelligent tutoring, science simulation, personalised learning,

resources/courses, educational games).

Finance (Wealth Management, Insurance, Fraud Detection, Banking, Personal Finance

Management).

Entertainment (Movies, Games, Advertising, Personalised Content, Music).

Context

Market report with definitions for machine learning, robotics (including

subcategories), artificial neural networks.

Date of

publication/

release

2017

Comments

3.3.3 McKinsey, 2017

Source

McKinsey Global Institute. Artificial Intelligence. The next digital Frontier?

Text of the

definition

Subdomains

Computer vision; natural language; machine learning; autonomous vehicles; smart

robotics; virtual agents

Context

Discussion Paper on AI landscape, investment and expenditures in AI

Date of

publication/

release

June 2017

Comments

The global AI landscape, expenditure and investment are discussed, with analysis by

technological subcategories, affected sectors in the value chain (leaders, followers,

adopters etc.). More detailed information is provided for leading countries.

4 Conclusions

The absence of a formal commonly agreed AI definition demanded the development of a process to establish

a reference AI definition, and its subsequent operationalisation into a taxonomy and representative keywords,

which can be adopted in the AI Watch framework and used in mapping and monitoring activities. The

proposed iterative process includes three perspectives: policy and institutional, research, and market, in order

to acquire a comprehensive overview about the AI domain. The AI definition adopted by the High Level Expert

Group on AI is used as a baseline definition. It is selected based on the review of 55 relevant documents

covering AI policy and institutional reports (including standardisation efforts, national strategies, and

international organisations reports), research publications and market reports. An exhaustive list of the

collected documents can be found in the report. The proposed operational definition is composed by a concise

taxonomy characterising the core domains of the AI research field and transversal topics; and a list of

keywords representative of such taxonomy. As AI is a dynamic field, we propose an iterative method that can

be updated over time to capture the rapid AI evolution.

While the baseline definition will be used as the general AI Watch definition of AI, the operational definition

has a more functional use. Both the taxonomy and the list of keywords are essential to identify, map and

characterise the worldwide AI landscape, one of the monitoring goals of AI Watch. The keywords are used in

the initial phase to capture the relevant AI activities and the economic agents behind them. The main utility of

the taxonomy is to classify AI activities, and will assist in the mapping of the AI landscape and the

classification of economic agents’ areas of specialisation. Different uses of the keyword list are possible. A

narrow use of the list, i.e. selecting only intrinsic-AI terms, allows to identify relevant AI activities, with an

expected low proportion of false positives. When the objective is the categorisation of AI-related activities, a

more comprehensive list is more suitable, in order to classify activities in their corresponding taxonomy

domains.

Valuable contributions of this work are: the collection of definitions developed between 1955 and 2019; the

summarisation of the main features of the concept of artificial intelligence as reflected in the relevant

literature; and the development of a replicable process that can provide a dynamic definition and taxonomy of

the AI.

References

AI 4 Belgium Coalition (2019), AI 4 Belgium Report

Blei, D.M., Lafferty, J.D.: Topic models. Text mining: classification, clustering, and applications 10(71) (2009)

34.

Blei, D.M., Ng, A.Y., Jordan, M.I.: Latent Dirichlet Allocation. Journal of Machine Learning research 3(Jan) (2003)

993-1022

Bruner J. (2009), Culture, Mind and Education. Contemporary theories of learning.

CB Insights (2019), Artificial Intelligence Trends

China Institute for Science and Technology Policy at Tsinghua University (2018), AI Development Report.

China’s State Council (2017), Next Generation Artificial Intelligence Development Plan (AIDP).

CIFAR (2017), Pan-Canadian Artificial Intelligence Strategy

Commission des affaires européennes. Gattolin A., Kern C., Pellevat C., Ouzoulias P. (2019), Rapport

d'information sur la stratégie européenne pour l'intelligence artificielle. Intelligence artificielle : l'urgence d'une

ambition européenne.

Craglia M. (Ed.), Annoni A., Benczur P., Bertoldi P., Delipetrev P., De Prato G., Feijoo C., Fernandez Macias E.,

Gomez E., Iglesias M., Junklewitz H, López Cobo M., Martens B., Nascimento S., Nativi S., Polvora A., Sanchez I.,

Tolan S., Tuomi I., Vesnic Alujevic L., Artificial Intelligence - A European Perspective, EUR 29425 EN,

Publications Office, Luxembourg, 2018, ISBN 978-92-79-97217-1, doi:10.2760/11251, JRC113826

D. B. Fogel. (1995), Evolutionary Computation: Toward a New Philosophy of Machine Intelligence.

D. B. Fogel. (1995), Review of computational intelligence: Imitating life. Proc. of the IEEE, 83(11).

Danish Government: Ministry of Finance and Ministry of Industry, Business and Financial Affairs (2019),

Strategy for Denmark’s Digital Growth.

Dawson, D. and Schleiger, E., Horton, J., McLaughlin, J., Robinson, C., Quezada, G., Scowcroft, J., Hajkowicz S.

(2019), Artificial Intelligence: Australia’s Ethics Framework. Data61 CSIRO, Australia.

De Prato G., López Cobo., M., Samoili S., Righi R., Vázquez-Prada Baillet, M., and Cardona M., The AI Techno-

Economic Segment Analysis. Selected Indicators, EUR 29952 EN, Publications Office of the European Union,

Luxembourg, 2019, ISBN 978-92-76-12584-6, doi:10.2760/576586, JRC118071

EC Communication from the Commission to the European Parliament, the European Council, the Council, the

European Economic and Social Committee and the Committee of the Regions. Artificial Intelligence for Europe.

COM(2018) 237 final {SWD(2018) 137 final}.

EC. Coordinated Plan on AI. COM(2018) 795 final and Annex

ETSI (2018), ETSI GR ENI 004 v.1.1.1. Experiential Network Intelligence (ENI); Terminology for Main Concepts in

ENI

Federal Government (2018), Artificial Intelligence Strategy.

Gardner, H. (1999), Intelligence Reframed: Multiple Intelligences for the 21st Century, pp.33-34.

Gardner, H. (1987), The mind's new science: A history of the cognitive revolution. Basic books.

Gardner, H. (1983), Frames of Mind; The Theory of Multiple Intelligences. New York, NY: basic Books.

General Secretariat of Scientific Policy Coordination of the Ministry of Science, Innovation and Universities and

to the Artificial Intelligence Task Force (GTIA, Grupo de Trabajo de Inteligencia Artificial) (2019), Spanish RDI

Strategy in Artificial Intelligence

Government Offices of Sweden: Ministry of Enterprise and Innovation (2018), National Approach to AI

(N2018.36).

H. Nakashima (1999), AI as complex information processing. Minds and machines, 9:57–80.

High Level Expert Group on Artificial Intelligence (2019), A definition of AI: Main capabilities and disciplines.

HM Government: Department for Business, Energy & Industrial Strategy (2017) Industrial Strategy. Building a

Britain fit for the future.

HM Government: Department for Business, Energy & Industrial Strategy, Department for Digital, Culture,

Media & Sport (2018) Industrial Strategy. Artificial Intelligence Sector Deal.

ISO/IEC 2382:2015

J. S. Albus. (1991), Outline for a theory of intelligence. IEEE Trans. Systems, Man and Cybernetics, 21(3):473–

509.

Kaplan, A. and Haenlein, M. (2018), Siri, Siri, in my hand: Who’s the fairest in the land? On the interpretations,

illustrations, and implications of artificial intelligence.

Kaplan, J. (2016), Artificial Intelligence What everyone needs to know.

Larosse J. (Vanguard Initiatives Consult&Creation) for DG CNECT (2019), Analysis of National Initiatives on

Digitising European Industry. Denmark: Towards a Digital Growth Strategy - MADE.

McCarthy, J. (2007) What is Artificial Intelligence.

McCarthy, J. (1988), The Logic and Philosophy of Artificial intelligence

McCarthy, J. (1959), Programs with Common Sense.

McCarthy, J., Minsky, M. L., Rochester, N., Shannon, C.E. (1955), A Proposal For The Dartmouth Summer

Research Project On Artificial Intelligence

McKinsey Global Institute (2017, Artificial Intelligence. The next digital Frontier?

Minsky, M. L. (1969), Semantic information processing. Cambridge, MA: MIT Press

Neisser U., Boodoo G., Bouchard T.J., Boykin A.W., Brody N., Ceci S.J., Halpern D.F., Loehlin J.C., Perloff R.,

Sternberg R.J., and Urbina S. (1996), Intelligence: Knows and Unknowns.

Newell, A., Simon, H. A. (1976), Computer science as empirical enquiry: Symbols and search. Communications

of the ACM 19, 3:113–126.

Nilsson, N.J. (1998), Artificial intelligence: a new synthesis. Morgan Kaufmann Publishers, Inc.

NITI Aayog (2018), National Strategy for Artificial Intelligence #AIFORALL

OECD (2019), Recommendation of the Council on Artificial Intelligence, OECD/LEGAL/0449

OECD (2018), Directorate for Science, Technology and Innovation, Committee on Industry, Innovation and

Entrepreneurship. Identifying and Measuring Developments in Artificial Intelligence. DSTI/CIIE/ WPIA(2018)4

OECD (2017), Science, Technology and Industry Scoreboard 2017. The Digital Transformation.

Papadimitriou, C.H., Raghavan, P., Tamaki, H., Vempala, S.: Latent semantic indexing: A probabilistic analysis.

Journal of Computer and System Sciences 61(2) (2000) 217-235

Parliamentary Mission (Villani Mission): Villani C., Schoenauer M., Bonnet Y., Berthet C., Cornut A.-C., Levin F.,

Rondepierre B. (2018), For A Meaningful Artificial Intelligence Towards A French And European Strategy

(Donner un sens à l'intelligence artificielle : pour une stratégie nationale et européenne).

Poole, D., Mackworth, A. (2017), Artificial Intelligence: Foundations of Computational Agents, second edition.

Poole, D., Mackworth, A. (2010), Artificial Intelligence Foundations of Computer Agents.

Poole, D., Mackworth, A., and Goebel, R. (1998). Computational Intelligence: A Logical Approach. Oxford

University Press, New York.

Russel, S. and Norvig, P. (2010), Artificial Intelligence. A Modern Approach.

Samoili S., Righi R., Cardona M., López Cobo M., Vázquez-Prada Baillet M., and De Prato G., TES analysis of AI

Worldwide Ecosystem in 2009-2018, EUR 30109 EN, Publications Office of the European Union, Luxembourg,

2020, ISBN 978-92-76-16661-0, doi:10.2760/85212, JRC120106.

Samoili S., Righi R., Lopez-Cobo M., Cardona M., and De Prato G. (2019) Unveiling Latent Relations in the

Photonics Techno-Economic Complex System. In: Cagnoni S., Mordonini M., Pecori R., Roli A., Villani M. (eds)

Artificial Life and Evolutionary Computation. WIVACE 2018. Communications in Computer and Information

Science, vol 900. Springer, Cham

Schank R. (1991), Where’s the AI? AI magazine, 12(4):38–49, 1991

Schank R.C. (1987), What is AI, Anyway? AI Magazine, 8 (4), aaai.org

StandICT.eu project (2019), Supporting European Experts Presence in International Standardisation Activities

in ICT (StandICT.eu). ICT standards and ongoing work at International level in the AI field – a Landscape

analysis

Statista (2017),Statista Report 2017. Artificial Intelligence

Stone, P., Brooks, R., Brynjolfsson, E., Calo, R., Etzioni, O., Hager, G., Hirschberg, J., Kalyanakrishnan, S., Kamar,

E., Kraus, S., Leyton-Brown, K., Parkes, D., Press, W., Saxenian, A.L, Shah, J., Tambe, M., and Teller, A. (2016),

Artificial Intelligence and Life in 2030. One Hundred Year Study on Artificial Intelligence: Report of the 2015-

2016 Study Panel, Stanford University, Stanford, CA.

Strategic Council for AI Technology (2017), Artificial Intelligence Technology Strategy.

UNESCO (2019). Principles for AI: Towards a Humanistic Approach? A Global Conference

US Congressional Research Service (2019), Artificial Intelligence and National Security.

US Department of Defense, Govini (2018), Artificial intelligence, big data and cloud taxonomy.

US National Defense (2018) Authorization Act for Fiscal Year 2019.

Wang P. (1995), On the working definition of intelligence. Center for Research on Concepts and Cognition,

Indiana University.

World Economic Forum (2017), WEF. 2017. Impact of the Fourth Industrial Revolution on Supply Chains.

List of tables

Table 1. AI domains and subdomains constituting one part of the operational definition of AI .................11

Table 2. Most relevant keywords within each AI domain ...........................................................16

Table 3. Summary of definitions and subdomains or applications referred to in the collected documents. ...18

GETTING IN TOUCH WITH THE EU

In person

All over the European Union there are hundreds of Europe Direct information centres. You can find the address of the centre

nearest you at: https://europa.eu/european-union/contact_en

On the phone or by email

Europe Direct is a service that answers your questions about the European Union. You can contact this service:

- by freephone: 00 800 6 7 8 9 10 11 (certain operators may charge for these calls),

- at the following standard number: +32 22999696, or

- by electronic mail via: https://europa.eu/european-union/contact_en

FINDING INFORMATION ABOUT THE EU

Online

Information about the European Union in all the official languages of the EU is available on the Europa website at:

https://europa.eu/european-union/index_en

EU publications

You can download or order free and priced EU publications from EU Bookshop at: https://publications.europa.eu/en/publications.

Multiple copies of free publications may be obtained by contacting Europe Direct or your local information centre (see

https://europa.eu/european-union/contact_en).

KJ-NA-30117-EN-N

doi:10.2760/382730

ISBN 978-92-76-17045-7