The form of our society has been transformed into the mixture of human beings and intelligent machines superior to humans in some respects. As the society is transformed, social information systems, rules, and mechanisms also have to be changed. It is, however, not clear how we should change them.

The core of research and development in CDL (Collaboration Design Laboratory) is multi-agent systems (MAS) which is a part of artificial intelligence researches. MAS consists of software agents, each of which is like a micro AI program modeling an entity such as a human/robot. MAS is a technical field that is useful for modeling large-scale complex systems such as human societies and for designing highly complex information systems.

In order to create information systems, rules, mechanisms, and services to maintain our societies and urban systems, we have worked on multi-agent social simulation (MASS). Towards the purpose, we have engaged in a variety of research activities, web mining&field sensing for obtaining social data, modeling human behaviors based on machine learning technologies, developing computation techniques for large-scale MMAS, etc.

ICT/AI compensate for our abilities to realize a convenient and affluent society. For example, MMAS is able to help us to estimate the complex behaviours of human society for designing improved social rules/systems, city structures, and new services. However, since human society involves diverse values, we ourselves need to think about what our society should be. At present, computers do not calculate our well-being while respecting morality and ethics. Thus, we have also worked on how to support discussion or dialogue considering the variety of sense of values. This research activity is necessary to make clues and knowledge for designing/improving society, obtained by MASS, acceptable to stakeholders.

[Overview of Research Activities at CDL]

As shown in the figure above, we are trying to realize the cyclic social system design processes consisting of social big data gathering, analysis&modeling using social big data, MASS-based social system design&evaluation, and implementation of the designed systems in the real world.

Traffic Simulation and Policy Making Support

The traffic flow in the city can be reproduced by multi-agent simulations. In order to conduct agent-based traffic simulations, various types of vehicle agents such as general vehicles, taxis, and buses are implemented, and thousands of vehicle agents are simultaneously run to generate virtual traffic flows. Based on simulated traffic flow, we have conducted simulations of energy sharing in a supposed environment where electric vehicles and photovoltaic power generation are widely used. In addition, we are working to build a detailed simulation environment for Kusatsu City, where the university is located, and to enable verification of traffic policies and transportation systems. For example, MAS can be used to support the preliminary verification of the effects of introducing self-driving vehicles and the optimal design of new transportation systems such as ride-sharing.

  • Kingetsu, H., Hattori, H.: Investigation of the Potential of Multi-Agent Traffic Simulations to Find Good Arrangement of Taxies. Transactions of the Japanese Society for Artificial Intelligence, Vol. 34, No. 3, pp. C-IA2_1-9, 2019 (in Japanese) [J-STAGE]
  • Noda, I., Murase, Y., Ito, N., Izumi, K., Hattori, H., Kamada, T., Mizuta, H., Takeuchi, M.: Project CASSIA – Framework for Exhaustive and Large-Scale Social Simulation -, Advanced Software Technologies for Post-Peta Scale Computing, Springer, PP. 271–299, 2019
  • Kingetsu, H., Hattori, H.: Modeling Individual Strategies of Taxis using Probe-Data
    and its Application to Traffic Simulations. IPSJ Journal, Vol. 59, No. 5, 2018 (in Japanese)
  • Hattori, H., Jumi, S., Nakajima, Y.: Exploring Potential for Social System Design Using Multi-Agent Simulations. IEICE TRANSACTIONS on Information and Systems, Vol. J100-D, No. 2, pp.180-193, 2017 (in Japanese)
Drone-based Field Sensing

It is difficult to obtain enough volume of traffic data on how people and vehicles are moving in the real world. We are thus trying to generate the data of the city dynamics by recording moving images from the sky using a drone and technologies such as image analysis and deep learning. The obtained data can be used to construct computational models for mobile entities, assimilate data in simulations, and verify the validity of models and simulations.

Discussion Support System to Incorporate Various Sense of Values

While access to vast amounts of information has become possible, it has become difficult to grasp, interpret, and understand new and changing things from a broad perspective, taking into account diverse values. We have engaged to develop a prototype system, AIR-VAS, which extracts words that appear in people’s discussions, visualizes their relationships, presents different viewpoints, and helps people to understand and think about new things. This work has been conducted by the collaboration with AIR members.

  • Yoshizoe, M., Hattori, H.,Ema, A.,Osawa, H.,Kanzaki, S. Values Awareness Support: Visualization of the Discussion. Journal of science and technology studies, No. 16, pp. 120-132, 2018 (in Japanese)


コラボデザイン研究室では,人工知能の一分野であるマルチエージェントシステム(multi-agent system)をコアの技術として,研究・開発を行っています.マルチエージェントシステムとは,たとえば人間やロボットのように考え,行動する主体をエージェント(agent)とよぶソフトウェア(小さなAIプログラム,と思ってもらえば良いです)としてモデル化・実装し,それらエージェントの相互作用を研究する領域です.相互に作用するエージェントの集合体であるマルチエージェントシステムは,個々に独立した行動主体の集積から成る組織,たとえば人間社会のような大規模複雑系をモデル化したり,また高度に複雑なシステムを設計する際に有用と言われる技術分野でもあります.

当研究室では,社会や都市を支える情報システム,制度,しくみやサービスなどを創り出す事も目的に,まず,マルチーエージェントに基づくシミュレーション,すなわちマルチエージェントシミュレーション(multi-agent simulation:MAS)による社会システム・制度のデザインに関する研究を行っています.そのために,Web(情報空間)でのマイニングやフィールド(物理空間)でのセンシングによる社会データの獲得と,機械学習等による個人の行動分析とモデル化を行うための研究を行っています.また,社会という大規模複雑系を扱うために,シミュレーションの大規模実行やその応用の研究開発を行っています.

参考:服部宏充,栗原聡.エージェント研究におけるシミュレーション.人工知能学会誌,Vol. 28, No. 3, pp. 412-417, 2013 [AI書庫]


参考:服部宏充.人々の良きパートナーとなる人工知能へ(<特集>編集委員会企画-社会とAIの羅針盤2015-).人工知能学会誌,Vol. 30, No. 1, pp. 31, 2015 [AI書庫]






  • 金月寛彰,服部宏充.マルチエージェントシミュレーションによるタクシー営業戦略の改善シナリオの提案.人工知能学会論文誌,Vol. 34,No. 3,pp. C-IA2_1-92, 2019 [J-STAGE]
  • Noda, I., Murase, Y., Ito, N., Izumi, K., Hattori, H., Kamada, T., Mizuta, H., Takeuchi, M.: Project CASSIA – Framework for Exhaustive and Large-Scale Social Simulation -, Advanced Software Technologies for Post-Peta Scale Computing, Springer, PP. 271–299, 2019
  • 金月寛彰,服部宏充.プローブデータを用いたタクシーの個別営業戦略のモデル化と交通シミュレーションへの適用.情報処理学会論文誌,Vol. 59,No. 5,2018
  • 服部宏充,十見俊輔,中島悠.大規模マルチエージェントシミュレーションに基づく社会システムデザインの可能性.電子情報通信学会論文誌D,Vol. J100-D,No. 2,pp. 180-193,2017




  • 吉添衛,服部宏充,江間有沙,大澤博隆,神崎 宣次. 多様な価値観への気づき支援 – 議論の可視化と考察. 科学技術社会論研究, No. 16, pp. 120-132, 2018


  • 西朋里, 小川祐樹, 高史明, 高野雅典, 森下壮一郎, 服部宏充:ネットテレビのニュース番組に投稿される視聴者コメントの道徳性に基づく分析 ,人工知能学会論文誌 36(1) WI2-E_1 – 9 2021

ソーシャルメディア上には,人々の様々な意見や感情が日々多数投稿されています.なかには,ニュースや組織・企業に対する意見や感情も含まれており,これらの情報はオフラインでの出来事に対する人々の意見を写す情報であるとともに,それを閲覧する人々にも影響を与えています.本研究では,ソーシャルメディアに投稿される書き込みと株価指標との関連を探り,株式市場のリスク・リターン予測するといった研究をしています.(共同研究:Yahoo! JAPAN研究所,東京都市大学,奈良先端科学技術大学院大学)

  • 佐々木皓大, 諏訪博彦, 小川祐樹, 梅原英一, 山下達雄, 坪内孝太, 安本慶一:ヤフーファイナンス掲示板の投稿を用いた株投資リスク低減のための日経VI上昇予測,第40回社会システムと知能合同研究会(SIG-SAI)/ 社会システムと情報技術研究ウィーク(RST ’21) 2021
  • Kodai Sasaki, Hirohiko Suwa, Yuki Ogawa, Eiichi Umehara, Tatsuo Yamashita, and Kota Tsubouch.: The 53rd Hawaii International Conference on System Sciences, (HICSS-53) 2020


  • 鶴山優季子,諏訪博彦,小川祐樹,荒川豊,安本慶一:W2VXMを用いた飲食店向け賃料推定モデルの構築,社会システムと情報技術研究ウィーク(WSSIT20) 2020
  • 河村一輝, 諏訪博彦, 小川祐樹, 荒川豊, 安本慶一, 太田敏澄:飲食店向け不動産営業を支援する申込み顧客推薦モデルの提案,人工知能学会論文誌 32(1) 1 – 10 2017


  • Kobayashi, T., Ogawa, Y., Suzuki, T., Yamamoto H.: News audience fragmentation in the Japanese Twittersphere, Asian Journal of Communication 29(3) 274 – 290 2019
  • 林浩輝,梅原英一,小川祐樹:否決された大阪都構想のTwitter投稿における世論形成理論成立の考察,社会情報学 8(3) 165 – 175 2020
  • 小川祐樹,山本仁志,宮田加久子:Twitterにおける意見の多数派認知とパーソナルネットワークの同質性が発言に与える影響-原子力発電を争点とした Twitter上での沈黙の螺旋理論の検証-,人工知能学会論文誌 29(5) 483-492 2014