map a flow of intelligence activity map b matching ... · perspectives (fig. 2). the four maps are...

The Japanese Societ y for Ar tific ial Intell igence, AI Map task force

Map AFlow of intel l igence act iv i t y

Map BMatching technology andappl icat ions wi th the nex t target

Map CConnect iv i t y of fundamentalswi th methods and appl icat ions

Map DAI research is d iverseThe vast f ront ier of AI

01

Artificial intelligence (AI) research has expanded greatly this century, making it difficult to capture the whole picture. Therefore, we have created the ‘AI Map’ series of maps to provide sketches of this picture to help new AI researchers with their future work as well as researchers in other fields who would like to use AI (Fig. 1).

There are many areas of AI research. These areas are intricately linked and developing rapidly so it is difficult to fit the relevance of all research fields into a single map without contradiction. Therefore, for the beta version of the AI Map, we created four maps that capture AI research from four different perspectives (Fig. 2).

The four maps are beta versions and we plan to improve them further. We also expect that volunteers will produce additional maps. For example, experts in specific fields will create partially detailed maps as well as tutorials, and these maps will be linked to the beta versions.

As an introduction, we describe the four maps and illustrate their usage.

Map A focuses on the process of intelligence. The concept of intelligence as an input/output process flow is shared by many AI researchers, and research on each step of this process is ongoing. This map is intended to be used to develop fundamen-tal research to realize complex processing, or to deconstruct

intellectual processing. In addition, the viewpoints of individual intelligence and group intelligence are included in this map.

Map B shows the relationship between technologies and applications. Many AI studies research fundamental technolo-gies with limited targets. This map shows representative pairs of technologies and objects. Because there have been many successful studies in which the target shifted, the next successful area may center on individual keywords. The map is intended to show how to shift research targets as well as the peripheral technology used for applications.

Map C focuses on the foundations that AI research rests on and AI’s various applications. AI research is a highly interdisci-plinary field broadly based on natural sciences, humanities, and social sciences. In addition, the map shows important areas of application. This map is intended as a reference when reviewing fundamental research or when exploring new applications.

Map D shows AI researchers’ various answers to the question “What is intelligence?”. One researcher answered “learning, recognition, and prediction”. Other research is pursuing various aspects of intelligence, such as “inference, knowledge, and language” as well as “discovery, search, and creation”, and how these aspects mutually affect one another. This map is intended to show the spread of AI research and the depth of the research field. The frontier of AI research is broad.

© The Japanese Society for Artificial Intelligence

02


Figure 1. Main target users of AI Map Beta

Interdisciplinaryresearchers

Target users

Beginners

Figure 2. Relationship between the whole of AI research and AI Map Beta

Flow of intelligence processing

What is intelligence?

Successful applications and their development trends

What kind of applicationand basis are there?

AB

CD

H

G

E

K

I

ML

N

O

P

Map AMap D

Map BMap C

A

03

Flow of intelligence activity

AI research sees human activity as a flow consisting of a combination of many intellec-tual activities. There are research fields that correspond to each step in this flow. Humans perceive and interpret the visual image, pay attention to the required informa-tion, evaluate the information based on the selected information, form an intention, and decide a series of operation sequences. For example, let’s consider a fellow research-er who approaches while holding out his right hand. I recognize the right hand approaching and identify the person as non-Japanese. In addition, his expression is friendly. I remember that there is a custom of shaking hands in foreign countries. I combine the recognition, and construct a series of actions, such as putting out my right hand, smiling while making eye contact, and shaking his hand.

AI also needs to work with humans by communicating with the people around it, and this involves many areas of research. For example, one area of research studies the interaction and dialogue between humans and robots with physical bodies.

In addition, many new research fields are emerging that examine how humans view AI. Research is also required on the appropriate use of AI, and includes evaluating AI’s reliability and operability.

Novices can learn about applications and activities related to their academic fields. For those who are already researching a certain field, the map can show related AI research themes and highlight possible partners for collaboration.

Internalenvironment

Externalenvironment

Objective

Evaluation

Interpretation

Perception

Intention formation

Operation selection

Execution

Robot, AgentHuman, Interaction,

Information

Relationship between AI and people

AI elemental technologies

Discovery, Search

Prediction

Creation

Knowledge, Language

Body, Motion

Learning

Cognition

Brain science

Cognitive scienceMedia

Artificial GeneralIntelligence

Reasoning

AI Integration

Evolution, Growth

04

Internalenvironment

Externalenvironment

Robot, Agent

Human, Interaction, Information


Discovery, Search

Creation

Knowledge, Language

Body, Motion

Prediction

Learning

Cognition

Brain science

Cognitive scienceMedia

Reasoning

AI elemental technologies

AI Integration


Evolution, Growth

Objective

Evaluation

Interpretation

Perception

Intention formation

Operation selection

Execution

Cognitive architecture

Machine Learning

Behavior estimation

Artificial neural network

Commonsense

Evolutionary comp. Artificial life

Genetic Algorithm

Inductive reasoningHypothetical reasoning

Deductive inference

Natural languageprocessing

Information Retrieval Affordance

Understanding of meaning

Concept learning Knowledge acquisition

Knowledge discovery

AI for medical care Agriculture and AI

Real estate and AI

Computational social sciences

Transportation, Logistics and AI

Automated driving system

Financial AIIndustrial AI Food and AI

Materials informatics

AI ethics

Legal AI

AI lifecycle

ExplainabilityMachine LearningEngineering

Privacy preservingdata mining

Software engineering

Fairness of AI

AI reliability

AI evaluation/validation

Information visualization

Green AI

Social infrastracture and AI

Fuzzy Logic

Prediction

Summarization

Serach

Transfer learning

Deep Learning Reinforcement learning

Anomaly detection

Clustering

Image recognition

Embodiment

Robot

Multi agent system

Human Interface

Kansei

Onomatopoeia Brain science

Cognitive science

Cognitive neuroscience

Knowledge sharing

InteractionBehavioral economics

Recommendation system

Machine translation

Expert systemMedia synthesisTuring test

Intelligent tutoring system

Game AI

CreativityMusic and AI

Intelligent user interface

HAI

HRI

Human computation

Speech recognition

Signal recognition Computer vision

SegmentationGesture recognition

Touch, tasteand smell recognition

Heuristics

Knowledge graph

TheoremprovingComputational neuroscience

Ontology

Knowledge base

Logic programming

Planning

Robotics

KinematicsSkill science

Subsumptionarchitecture

Semantic web

Singularity

Swarmvintelligence

AGI

GAN

Bayesian inference

Triggers for Behavior Change

AI ethicsAI ethics

ReliabilityReliability

Causal inference

Evaluationand verification


05

Matching technology and applications

with the next target

Application

Technology

Cognitive science Problem solving Knowledge,Reasoning

Learning

Web

Natural language

Image

Sound

Agent

Cyber Physical

System

Multimedia

Monomedia

Society and AI

Human

Singularity

AI research has produced a large number of technologies that are general-purpose with no specific target. However, during development, targets have been defined within a certain range, and realization methods and fundamental technologies have been developed for them. For example, image targets have evolved from simple signal recognition to complex image recognition, and then to generative adversarial network (GAN).

In addition, technological development in a specific field has stimulated related technical fields, and has created new technology-target pairs. Therefore, the area around coordinates that is currently producing a large number of new technologies could be developed by changing the target or changing the technical goal.

For example, as a hypothesis, in recent years the practicality of multi-agent systems has been rapidly improved by increases in computing power and the preparation of large amounts of data. Looking at surrounding areas on the map, scheduling is likely to be closer to problem solving, and as the target moves closer to knowledge and reasoning, new technological advances may emerge in multi-agent systems (e.g., the Semantic Web).

B


Robot

Kansei

Market design

Game AI

Computer vision


Learning

Web

Natural language

Image

Sound

Agent

Cyber Physical

06

Technology

Application

Rule-based system

常識推論

因果推論

SAT CSP

Math. optimization Information visualization

Text Mining

Data mining

Auction

Scheduling

Society and AI

System

Multimedia

Monomedia

Human


Computational neuroscience

Onomatopoeia

Signal recognition

Understanding of meaning

Logic programming

Planning

SerachFuzzy Logic

Theoremproving

Heuristics

Skill science

Embodiment



Human computation

Commonsense

Causal inference

HRI

Genetic Algorithm

Swarmvintelligence

Knowledge base

Ontology

Multi agent systemEvolutionary comp.

Artificial life

Inductive reasoning

Hypothetical reasoning

Deductive inference

Image recognition

Speech recognition GAN

Deep Learning

Summarization

Interaction Concept learning


SegmentationWeb mining

ClusteringPredictionSemantic web

Knowledge graph

Anomaly detection

Knowledge discovery

Knowledge acquisitionKnowledge sharingGesture recognition

Data science

Transfer learning

Reinforcement learning

Behavior estimation

Expert system

Bayesian inference

AGI

Media synthesis

Music and AI


AI evaluation/validation

AI ethics

Fairness of AI

AI reliability Singularity

Explainability

AI lifecycle

CreativityLegal AI

Triggers for Behavior ChangeIntelligent user interface

Machine translation

Information Retrieval


HAI

07

Connectivity of fundamentals

with methods and applications

Application

Method

Basis

Industrial Application

Game Medical and BiologyRobotics

Human Interface, Interaction

Social computing Agent

Natural Language Processing Multimedia

AI Basics/Theory

Machine Learning/Data Mining / Text Mining Metaheuristics

Knowledge expressionLogic / Reasoning

Knowledge ManagementSemantic Technology

3rd

Major classification Middle classification

2nd

1st

3rd

2nd

1st

3rd

2nd

1stMap C shows how the roots and branches of AI research spread widely.

AI is built on many fundamental disciplines, including mathematics, statistics, logic, cognitive science, and neuroscience. Of course, it is impossible to master all these subjects before you start AI research. However, it may be useful to return to the basics before aiming to go beyond the horizon of current research. Along with reading the latest papers and comparing and evaluating the latest libraries on GitHub, it is also worthwhile spending time studying the fundamentals.

Map C shows the scope of AI applications, which is rapidly expanding. Applications will probably extend to every aspect of human society as practicability improves. As a starting point, the map shows the application areas that are now booming. For example, AI is used in various fields such as financial technology, medicine, real estate, music, and agriculture. In addition, as the applications expand, new technical issues and social issues are emerging, including ethics, credibility, and explainability. These are fed back as learning into the foundations of AI, and the large tree of AI research spreads further.

Major classification

C

08

3rd layer

2nd layer

1st layer

Social infrastructure

Basic science

EconomicProductionLifeCulture

Industrial infrastructure

Application

Method

Basis

Probability / Statistics* An example of interpretation : Inductive inference is created based on probability / statistics and mathematical science, machine learning technology develops, prediction technology advances social infrastructure AI, and contributes to infrastructure maintenance, environmental conservation, and economic activity development

Philosophy Mathematical science Data Science Computer science Psychology Biology

Peripheral research area

ArchitectureBig data analysisORMath. optimizationMathimatical statisticsMathimatical logic

Crowd Sourcing

Major classification

Knowledge management and inheritance

Society

Robotics

Social Computing

Agent

Machine Learning / Data Mining / Text Mining

User interface

Industry/Infrastructure Medical and biology

Management

Industrial application

Environment

EducationTransportation and logistics

Entertainment and art

Natural language processing

Multimedia

Deep LearningMetaheuristics

Artificial intelligence basics / theory

Knowledge engineeringKnowledge expression

Logic and reasoning

Semantic technology

Creativeactivity

Livingassistance

Data market designKnowledge management


AI for decision making and consensus building

SAT CSP

AI-based security

AI for Management

Food and AI

AI for Investment

AI teacherAI for Transportation, Logistics

Data Engineering

Applied science

Behavior estimation

Evolutionary comp.Genetic Algorithm

Concept learning Knowledge discovery


Materials informatics

AI ethics


AI reliability AI evaluation/validation

Green AI

Fuzzy LogicPrediction

Clustering

Image recognition

Embodiment

Multi agent system

Human Interface

Onomatopoeia

Brain scienceCognitive science

Interaction

Game AI

Creativity

Music and AI

HAIHRI

Speech recognitionSignal recognition

Segmentation

Gesture recognitionTouch, taste

and smell recognition

Knowledge graphOntology Knowledge base

Logic programming

Kinematics

Skill science


Semantic web

Swarmvintelligence

AGIGAN



Legal AI AI for medical care Agriculture and AIReal estate and AI

Industrial AI


Computational social sciences

Social infrastracture and AI

Transfer learning Anomaly detectionReinforcement learning Machine translation

Computer vision Media synthesis

Privacy preservingdata mining

Human computation

AI lifecycleInductive reasoningTheoremprovingDeductive inference Bayesian inference

Commonsense Knowledge acquisition


Expert system

Kansei


D

09

AI research is diverse

The vast frontier of AI

Reasoning, Knowledge, Language

Discovery, Search,Creation

Evolution, Life, Growth

Human, Interaction, Emotion

AI Frontier

Body, Robot, Motion

Scope of AI research

Learning, Cognition, Prediction

In AI research, there are various approaches for realizing mechanical intelligence. The ultimate goal of the "AI frontier" is the realization of intelligence comparable to or beyond that of humans and other living things, and its integration into society. In the surrounding area of the map, there are multiple viewpoints with different ways of thinking about intelligence that continue to be studied in depth, and each has made steady scientific and technological progress. Furthermore, AI research is related to many other research fields, and through close coordination with these other fields it can split from or fuse with other fields, opening up new horizons.

For example, in this map, the area of “reasoning, knowledge, and language” (also called “adult intelligence”) is shown in the upper right of “learning, recognition, and prediction”. Humans can use words, build and share knowledge, and make various inferences. Some of these processes are formalized in AI research and have theoreti-cal explanations and practical applications. In this area, new technologies and research directions are being developed with data-driven approaches. In addition, research is beginning to show that language and reasoning influence recognition itself.

To keep the map in a single plane, two main adjacent areas were used. However, in practice, fusion with the opposite region across the AI frontier is also popular. For example, the relationship between “reasoning, knowledge, and language” and “human, interaction, and emotion” is deep, and “onomatopoeia" is a research field located between these two opposite fields. In addition, image generation using GAN which is an application of deep learning, is a fusion of “learning, recognition, and prediction” and “discovery, search, and creation”. Future AI research may have great potential for merging areas that do not have deep links.

AI Fundamentals

AIFrontier

DiscoverySearchCreation

ReasoningKnowledgeLanguage

LearningCognitionPrediction

BodyRobotMotion

Human,InteractionEmotion

EvolutionLifeGrowth

AI research is developing in conjunction with a number of academic disciplines around it. The closer to the center, the more AI-specific or unresolved / undefined problems.

Reasoning, Knowledge, LanguageDiscovery, Search,Creation



Body, Robot, Motion

Learning, Cognition, Prediction

Scheduling

SAT CSP

Information visualization

Auction

Data mining

Web mining

Data market design Rule-based system

Foundations of mathematics Mathimatical logic Semantics Formal language theory

Mathematical programming

Graph theory

Game theory

Big data analysis

Database

Mechatronics

Body psychology

Sports brain science

Biology

Developmental science

Developmental psychology

Math. optimization

Simulation

OR

Data science

Mathimatical statistics

Object oriented

AI Frontier

10


Machine Learning

Behavior estimation

Commonsense

Evolutionary comp.

Artificial life

Genetic Algorithm

Inductive reasoning


Deductive inference

Natural languageprocessing


Affordance

Understanding of meaningConcept learning

Knowledge acquisition

Knowledge discovery


AI ethics

Legal AI

AI lifecycle

Machine LearningEngineering

Fairness of AI

AI reliability

AI evaluation/validationExplainability

Fuzzy Logic

Summarization SerachTransfer learning

Deep Learning

Reinforcement learning

Anomaly detection

Clustering

Embodiment

Multi agent system

Human Interface

Kansei

Onomatopoeia

Cognitive science

Cognitive neuroscience

Knowledge sharingInteraction

Behavioral economics


Expert system

Media synthesis


Game AI

Creativity


HAI

HRI

Human computation

Speech recognition

Segmentation

Gesture recognition

Knowledge graph Theoremproving

Computational neuroscience

Ontology

Knowledge base

Logic programming

Planning

Robotics

Kinematics

Skill science


Semantic web

Singularity

Swarmvintelligence

AGI

GAN

Bayesian inference


Causal inference

Brain science

Computer vision

Touch, tasteand smell recognition

Image recognition

PredictionSignal recognition



11

There are 24 special interest groups (SIGs) in the Japanese Society for Artificial Intelligence (JSAI). The following website provides informa-tion on SIGs.https://www.ai-gakkai.or.jp/sig/sig-list/

For beginners and interdisciplinary researchers who are interested in AI research, active partici-pation in SIGs is desirable for obtaining up-to-date research information and making contact with front-line researchers.

However, it is difficult to link the names of groups, titles of papers, and users’ interests. Therefore, we have created maps of the groups that overlap with AI Map, beta version to encour-age participation.

Map of special interest groups in the Japanese Society for Artificial Intelligence

Financial Informatics

Business Informatics

Knowledge-BasedSystem

Semantic Web and Ontology

FundamentalProblems in

Artificial Intelligence

Web Science

Web Science

Natural Computing

AGI

Language Senseprocessing Engneering

Artificial Intelligence in MedicineKnowledge Sharing Network

Knowledge Sharing Network

Advanced Learning Science and Engineering


Crowd Co-creative Intelligence

Knowledge and Skills Transfer

Skill Science

Skill Science

Skill Science

Data Oriented ConstructiveMining and Simulation

MolecularBioinformatics

Society and Artificial Intelligence

Interactive InformationAccess and Visual Mining

CommonsenseKnowledgeand Emotion


Spoken Language Understanding and Dialogue Processing


Commonsense Knowledgeand Emotion

Crowd Co-creativeIntelligence

Commonsense Knowledge and Emotion

AI Challenges

AI Challenges

Placement difficult

Internalenvironment

Externalenvironment

Robot, AgentHuman, Interaction,

Information


Brain science

Cognitive science

Media

AI elementaltechnologies

Application of AI to society

AI Integration


Evolution, Growth

AI ethicsAI ethics

ReliabilityReliability



Discovery, Search

Creation

Knowledge, Language

Body, Motion

Prediction

Learning

Cognition

Reasoning

MeasurementInformatics

12

However, it is difficult to link the names of groups, titles of papers, and users’ interests. Therefore, we have created maps of the groups that overlap with AI Map, beta version to encourage participation.

The four SIG maps shown here were created based on the results of a questionnaire answered by the leaders of each group. The maps are arranged to ensure maximum visibility.

There were several groups that were difficult to place on a specific map because of their diverse scope, so we have placed them outside of the maps. In addition, some groups are combining or simultaneously handling distant areas on the map. We have shown the spread of the target area by connecting areas with a line or vertically and horizontally.


Web Science

Natural Computing Measurement Informatics

Measurement Informatics

AGI





Skill Science





Crowd

Co-creativeIntelligence





Placement difficult


AI Challenges

AI Challenges


Society andArtificial Intelligence





Artificial Intelligencein Medicine

Web

Natural language

Image

Sound

Agent

Cyber Physical

Society and AI

System

Multimedia

Monomedia

Human


Learning Technology

Application

コモンセンス知識と情動

Skill Science


Placement difficult

13

These maps show the areas where research groups are concentrated, and the similarities between groups.

The JSAI holds joint research meetings of the SIGs once a year. Joining neighboring groups on the map will allow researchers to grasp the latest research trends in related areas quickly.

In addition, the areas in which the groups are concentrated on the map are likely to be hot topics in AI research and will highlight current research trends.







Web Science

Natural Computing


Triggers forBehavior Change

Artificial Intelligence in Medicine






Knowledge andSkills Transfer








AI Challenges


AI Challenges

AGI

3rd layer

2nd layer

1st layer

Application

Method

Basis

Economic

Production

Life

Culture

Knowledge managementand inheritance

Society

Robotics

Social Computing

AgentMachine Learning/Data Mining / Text Mining

User interface

Industry / Infrastructure

Medical and biologyManagement

Industrial application

Environment

EducationTransportationand logistics

Entertainment and art

Natural language processingMultimedia

Deep LearningMetaheuristics

Artificial intelligence basics / theory

Knowledge engineering

Knowledge expression Logic and reasoningSemantic technology

Creativeactivity

Livingassistance

Basic science

Applied science

Social infrastructure Industrial infrastructure

Probability / StatisticsPhilosophy Mathematical science Data Science Computer science Psychology Biology

Peripheral research area

AGI : Artificial General IntelligenceAIMED: Artificial Intelligence in MedicineALST : A d v a n c e d L e a r n i n g S c i e n c e and EngineeringAM : Interactive Information Access and V isua l Min ingBI : Bus iness Informat icsCCI : Crowd Co-creative IntelligenceChallenge: AI ChallengesCKE : Commonsense Knowledge and EmotionDOCMAS: Data Oriented Constructive Mining and SimulationFIN : Financial InformaticsFPAI : Fundamental Problems in Artificial IntelligenceKBS : Knowledge-Based SystemKSN : Knowledge Sharing NetworkKST : Knowledge and Skills TransferLSE : Language Sense processing EngneeringMBI : Molecular BioinformaticsMEI : Measurement InformaticsNAC : Natural ComputingSAI : Society and Artificial IntelligenceSKL : Skill ScienceSLUD : Spoken Language Understanding and Dialogue ProcessingSWO : Semant ic Web and Onto logyTBC : Tr iggers for Behav ior ChangeWebSci : Web Sc ience

List of SIGs

14

ことば工学

Placement difficult







Web Science

Natural Computing

AGI











Skill Science

Skill Science

Skill Science












Commonsense Knowledgeand Emotion Commonsense Knowledge

and Emotion




Artificial Intelligence in Medicine


AI Challenges

Reasoning, Knowledge, Language

Discovery, Search,Creation



Body, Robot, Motion

Learning, Cognition,Prediction

AI Frontier


15

The JSAI, AI Map task force

TSUTSUMI Fujio,Central Research Institute of Electric Power Industry

MORIKAWA Koji, Panasonic Corporation

ICHISE Ryutaro, National Institute of Informatics

UENO Ken, Toshiba Corporation

TSUMOTO Shusaku, Shimane University

FUKUSHIMA Shunichi, Japan Science and Technology Agency

The JSAI board of directors

The members of the Leader, Secretary of JSAI SIGs

Pamphlet production: Apricot Design

https://apricot-design.com/

Contact information on AI Map: info [at] ai-gakkai.or.jp

(Convert [at] to @)

Date of issue: June 6, 2019

Acknowledgements:

Publisher:

Future development and requests

In the map, we only used four viewpoints; however, while we were

creating the map, we were asked whether we could create various other

related maps. Some of the requests are shown below. We would like to

take these requests into consideration when making future updates.

(1) Data-driven interactive map

The JSAI has many high-quality documents, such as papers, journals,

and workshop reports, which contain many keywords. Therefore, a map

automatically created from these documents could be created by

visualizing mined text. Furthermore, converting the map to interactive

Web content that allows users to walk through multiple viewpoints would

allow the spread of AI research to be understood in three-dimensions.

(2) Learning support map for beginners

Beginners also need maps that can be used as guides to deepen their

understanding and gain practical knowledge that can be used for

research. For example, it is useful to know which fundamental academic

fields can contribute to the basis and applications of image recognition. In

addition, journals provide a wealth of knowledge, and there are many

tutorial articles that help beginners. A guide to the month and year of

issue of journals will improve a beginner’s ability to gather information.

(3) Major conference map and conference session map

There are numerous domestic and international conferences related to AI

research. Maps showing their relevance are also useful for efficient

article submission and information gathering. In addition, at annual

conferences, a large number of general and organized sessions are held,

which are wider in scope than the SIGs. Maps of these sessions will also

help to make the events meaningful.

The beta version of AI Map was created as part of the activities of the

Japan Society for Artificial Intelligence. If you are interested in this map,

we would recommend joining JSAI. Members can access useful informa-

tion, such as academic journals with articles on the latest AI research

and applications. In addition, members can contribute to annual

conferences, SIGs, and journals, and get discounts on participation fees

for events such as seminars. The admission process is detailed on the

following homepage (URL).

https://www.ai-gakkai.or.jp/about/membership/

Additional maps Join the Japan Society for Artificial Intelligence

Recruitment of people who would liketo make new maps

Every AI researcher will have a map based on his or her own perspective

and a map of his or her field of study. In addition, for each SIG, you may

be able to create a map showing research trends in each field, or a

tutorial for beginners in the research group. JSAI supports the creation

of new maps. In addition, we intend to recruit researchers who will help

update the next version of AI Map.

map a flow of intelligence activity map b matching ... · perspectives (fig. 2). the four maps are...

Documents