Japanese
Japanese

Research

OPAL-RING
Takeshi ITO Laboratory

Realizing Human-Friendly, Enriching Computers Through Research into Human Thought and Learning

As of August, 2015

Faculty/Department Department of Communication Engineering and Informatics
Graduate School of Informatics and Engineering
Takeshi ITO
Members Takeshi Ito, Assistant Professor
Affiliations Cognitive Science Society, International Computer Games Association, Information Processing Society of Japan, Japanese Cognitive Science Society, Japanese Society for Artificial Intelligence
Website http://minerva.cs.uec.ac.jp/~itolab-web/wiki.cgi
PDF

Keyword

Cognitive science, Artificial intelligence (AI), Games, Shogi, Go, Curling, Man-machine interfaces, Sports, Biological data measurements, Human error, Human-like

Measuring human mental states and developing friendly artificial intelligence

The Ito laboratory hopes to work with companies on joint research projects investigating human thought, learning, decisions, and other behaviors (see Figure 1). This research focuses on two main areas.

Figure 1: Image of the laboratory’s joint research areas. The two main research topics are biometrics and artificial intelligence.

The first research area measures, using biometric equipment, physiological conditions and cognitive states closely related to human mental states. Some of the measurement tools used are eye cameras (devices that track eye movement), diaphoremeters (sweat sensors), sphygmometers, and optical topography equipment (devices that track blood flow variations in the brain).
The second area is joint research looking at intelligent systems (artificial intelligence systems). The vision is to develop friendly systems that act human and that are enjoyable and interesting.
The enabler of this kind of joint research is the Ito laboratory’s work on research into advanced human cognitive activities. For example, the laboratory is attempting to clarify such questions as how do human solve difficult problems, what is human intuitive knowledge, and what thought processes do human follow in playing games (see Figure 2).

Figure 2: The laboratory researches advanced human cognitive activities. Advanced cognitive activities include problem solving, learning, and emotional sensitivity.

To answer these questions, the laboratory measures human cognitive processes and builds artificial intelligence systems. And in consideration of the interaction between humans and machines, another aim is to develop systems that enrich people’s lives.

Different thought processes used in shogi and Go

Much of the research on human thought and learning examines the process of solving problems while following certain constraints. Games are good models of the actions taken to solve problems while following certain constraints - rules, in other words.
Shogi and Go are two popular board games that have been played throughout Japan for centuries. The Ito laboratory has researched shogi for many years and has determined the cognitive processes characteristic of shogi players. For example, the more experienced the player, the fewer the positions the player examines. Unlike a machine, these players intuitively select their move without randomly studying many moves.
The laboratory has recently started studying Go players as well using eye-tracking and other kinds of devices. The results show that expert players look at and consider a wider range of positions than beginners (see Figure 3). The finding that cognitive processes varied according to the game (problem) was of great interest.

Figure 3: Differences in eye movements by level of experience with the board game Go. The left image is a beginner, the center image is an intermediate player, and the right image is an expert player.

The laboratory also studied, in partnership with RIKEN, the process by which beginners remember and become proficient at minishogi, a variant of shogi played on a 5 x 5 board with fewer pieces. The research found that when a certain part of the cerebrum (the caudate in the basal ganglia) is larger, learners gave a higher percentage of correct answers to intuitive problems at the outset of the learning process. The research also discovered that people who learn well initially thought for longer periods of time, had less eye movement, and recognized moves faster.

Making computers act human

The Ito laboratory has also researched artificial intelligence (AI) for many years using various thought games. The laboratory’s research examines the thought processes of people who have mastered games and attempts to create AI to play the games on computers.
Shogi computers have gotten much more powerful in recent years, and the defeat of a professional shogi player by a computer at a public match made news. In the world of video games, smartphone games, and computer games, shogi, Go, Othello, mahjong, and many other game titles are very commonplace.
The problem here is the strength of game AI. If a software title runs game AI that is too strong, it is likely to be very boring for human players, since they can never defeat the computer.
For this reason, the Ito laboratory started researching AI that has natural, human-like weaknesses. There are two basic approaches to implementing weakness. One is to change the data that the computer learns from. For example, in shogi, have the computer learn from game records of weaker players with an amateur rank of under 5 kyu instead of learning from the game records of professional players.
The other approach is to have the computer deliberately make mistakes. If the computer makes a wrong move in a complex position that is difficult to work out, the conditions of the game will change significantly. The advantage may swing to the human player. With this approach, researchers are trying to bring out human nature in computers by incorporating models of human error in AI.
The laboratory is also working on developing AI that emulates a specific playing style for a game. For this, the research is trying to identify the elements that make up a human’s playing style (personality). Taking shogi for example, researchers are attempting to enumerate the elements related to a certain shogi playing style and then identify in game records the elements making up distinctive characteristics and the distinctive characteristics themselves. For first-person shooter games, the laboratory is analyzing the play logs of advanced players.

Aiding sports tactics with AI

The Ito laboratory’s next probable major research topic is the application of AI technology, accumulated through thought games, to sports. In curling, soccer, baseball, football, and other sports, constructing tactics to win (or not to lose) is important. The laboratory’s researchers are thinking of systems to assist in forming tactics.
Research is currently underway on a tactical-support system for curling (see Figure 4). Plans are to expand the sports covered by the system over time.
[Interview and article by Akira Fukuda (PR Center)]

Figure 4: Proposal of a tactical-support system for curling. The system selects the best probable shot based on the location of the stones.