Published papers

Online services make users' communication activities and content public. This online information has contributed to accelerating the creation cycle of user-generated content. Moreover, these services also allow researchers to utilize these online texts as a public source for easily analyzing human activities, also referred to as social sensing studies. However, we need to realize that there exists a controversial problem of privacy especially in the sensitive areas of creation, even though the content is public. This study tries to create new guidelines for online study using the case of the flaming of a study of female fan-fiction, which attempted to extract and filter sexual expressions using online fan-fiction novels as source. Researchers from the fields of both engineering and humanities, including law and ethics, discussed the violations in this case, and extracted ethical, legal, and social issues according to their specific areas of expertise.

When we meet a person for the first time, impression of the person is positively correlated with the similarity of the personality. In this paper, the authors researches on whether such a correlation appears even between human and game agent. The author uses AI plays cooperative card game Hanabi as the game agent. The author implemented two kinds of Hanabi agent which has difference on tendency to take risky plays. Participants of an experiment played Hanabi with both agents and evaluate them. As a result, there are some positive correlations between humans' points of view of risk and failure and their evaluation in similar agent. This suggests that we need to take difference of users' personality into account when we design cooperative AI.

人狼ゲームを対象とした研究は人間との高度なコミュニケーションをとるエージェントの実現などへの応用が期待できる。しかし、人狼ゲームを対象とした非言語コミュニケーションの分析はこれまでほとんどされてこなかった。本研究では、人狼ゲームをプレイする人間のジェスチャーとゲーム結果との関連性を調査する。ジェスチャーに関するいくつかの指標を設け、それを定式化することで、ゲーム中に見られるジェスチャーの定量的評価を試みた。これらの指標のいくつかで、人狼ゲーム中でのプレイヤーの役割が測定出来ることがわかった。例えば、ゲームの勝敗には占い師と狂人がそれぞれどの程度リーダーシップを取っているかに影響を受け、それをプレイヤーの両腕の開き具合で測ることが出来る。

AI agent for cooperation with human has to follow human thought. There are some research using cooperating game 'Hanabi' from various aspects such as self-estimation, psychology, and communication theory. In this research, we developed an agent's strategy for following human thought for guessing human understanding of AI's strategy to utilize the length of thinking time of the human player and changing the estimation reliability. As a result, the author found that this agent estimated more frequently to the good player of Hanabi than the conventional agent and there is a positive correlation between some evaluation to this agent and score. However, the agent that change estimation reliability depending on opponent's thinking time didn't affect human's impression compared to the conventional agent and there was no significant difference in the score and the success rate of the estimation by changing the reliability of the estimation according to the thinking time.

This paper introduced analysis of agents' information sharing with evolutionary computation. We focus on a method that human adds a hidden meaning to our word for sharing information only with cooperators in subgroup. In this paper, we defined 3 player version of werewolf game, and run a 10000 generations evolutionary computation. Then, we classified villager's strategies with label of "rational", "irrational" and "ambiguous". We suggested that when villager used an "irrational" strategy and won the game, seer and villager used the above information sharing method. As the result, the above method appeared in both case of wolf fixed and wolf evolved. In the case of wolf fixed, seer added hidden meaning to a fixed word in 99.2% of times to share information only with the villager. In the case of wolf evolved, in 26.4% of times, seer succeed to share information only with the villager by the above method and seer started to change word with hidden meaning for keeping avoid sharing information with the wolf.

The various possibilities and risks of artificial intelligence are currently being discussed as an urgent issue for rapid AI development for real-world applications. Even the media and researchers often cite science fiction as a vision of the future. But because of the technological inaccuracies of entertainment, it's not always appropriate to bring these AI in science fiction to serious discussion in the future. On the other hand, some science fiction themes play an active role in reminding people of new visions. To understand the mixed effects of science fiction in the field of AI, the authors analyzed how AI was described in science fiction with science fiction critics and writers. First, we selected 115 AI stories in science fiction based on three criteria: intellectual diversity, social dimensions, and expansion of human intelligence. Eleven factors representing AI characteristics were analyzed using cluster analysis and principal component analysis. The results suggest that there are four characteristic AI clusters in SF. They are people, machines, helpers, infrastructure, and mapped to two-dimensional space. They are intelligence and humanity.

The cooperative game Hanabi is a suitable interactive subject for analyzing cooperative behavior based on implicit intention communication. In recent years, research has been conducted to analyze people's cooperative behavior by comparing the results of cooperating with several Hanabi agents that behave differently. However, these methods have problems such as large costs of experiment with human subjects and the inability to directly observe human's internal psychological states. One solution to these problems is to use models created on cognitive architectures, which are used to reproduce human behavior, and conduct simulations. In this study, we simulated the Hanabi game using model, developed on the cognitive architecture ACT-R, with and without risky actions. As a result, we found that the time spent per game changed similarly to that of the human Hanabi game. It was also suggested that having risky actions as options reduced the learning efficiency to improve scores and increased time required for action. In addition, memory retrieval was found to be a factor that increased the time required for action decisions in Hanabi play.