Multi-Agent Systems Group

In 2007, the Netherlands Organisation for Scientific Research (NWO) awarded its first 41 "TopTalent" grants. Applications were open to all research master's students in their final year of study, who were given the opportunity to write their own Ph.D. project proposals. One of these grants went to Elske van der Vaart. She was sponsored by Charlotte Hemelrijk (Theoretical Biology) and Rineke Verbrugge, who now are her promotors.

In recent years, members of the crow family have been shown to use tools, deceive others, and solve trap-tube problems that baffle chimpanzees. But how smart are they, really? Is it insight, instinct, or associative learning? In this project, we will simulate all three explanations in different 'virtual corvids', and expose them to the same experiments as their biological counterparts. We will focus specifically on their abilities to reason about the minds of others, and examine which accounts of corvid mentality best explain the data.

Nowadays, computer programs based on artificial intelligence work together in teams called multi-agent systems. In the future humans will cooperate with these programs, for example in healthcare, international negotiations, and rescue missions.

Software agents, the artificial members of the team, often reason based on formal logics. Usually they are capable of an arbitrary amount of recursion: A cannot only think about what B knows, but also about what B thinks that A knows that B wants... and so onwards, for every order of social cognition. However, humans lose track of such reasoning after only two or three levels.

If software agents work together with human teammates, it is very important that they take into account the limits of social cognition of their human counterparts. Otherwise an international negotiation, for example, fails, even when it has potential for a win-win solution. In a time-critical rescue mission, a software agent may depend on a human teammate's action that never occurs.

In the Vici research program, we aim to design new formal logics, which accurately model human limitations in higher-order social reasoning. We gather new information about children's development and adults' limitations in applying higher-order reasoning. For this, we use an unprecedented close-knit combination of empirical research and formal modeling. Logical and computer models provide predictions for innovative experiments with children and adults. These experimental results in turn lead to improved formal models and cognitive theory.

Whereas first-order social cognition has been the subject of intense investigation over the last decades, higher-order social cognition is far less well-understood. This research program makes two main contributions. Firstly, it provides a better understanding of higher-order social reasoning, for the benefit of cognitive scientists, logicians and computer scientists. Secondly, it designs realistic logics, ready for implementation in systems supporting higher-order social cognition in mixed human-computer teams.

Modelling intelligent and rational interaction in multi-agent systems has been one of the main issues in Artificial Intelligence that gained momentum in the last decade of the past century. There are lots of detailed studies of agents' knowledge, beliefs, preferences, and also of their long-term powers for influencing the outcomes of games. But we do not have a good theory yet of what may be the most crucial ingredient here: the plans or strategies that information-processing agents have for achieving these goals. In other words, their know-how in addition to their know-that. This is the main topic of this research program. The project will delve into the following topics:

•Sophisticated logical theories of strategies, influenced by the epistemic states of the agents in a dynamic setting.
•Associated language to programmatically model strategy choices in interaction.
•Computational models of strategies in multi-agent interaction systems so as to capture the complex ways of human interaction.
•User-friendly software to model real-life interactive systems.

NWO Vici Project:

Cognitive Systems in Interaction: Logical and Computational Models of Higher-Order Social Cognition

(Current)

Strategies in Multi-Agent Systems: From Implicit to Implementable

(Current)

Crime investigation is a difficult and laborious process that is prone to error and the costs of mistakes can be high. This project will develop software support to tackle two problems that often occur in complex criminal cases: Lack of overview of a mass of evidence by initial investigators and lack of transparency of case files for subsequent investigators, prosecutors and fact finders. Making Sense of Evidence is funded by a Netherlands Organisation for Scientific Research (NWO) grant in the ToKeN (Toegankelijkheid en Kennisontsluiting in Nederland) program; both Floris Bex and Bart Verheij participate in this project.

A demonstrator prototype will be developed of software with which crime investigators can visualize and analyse their reasoning about a case. Such software will support investigators in seeing patterns, discovering new relationships or inconsistencies, and identifying missing evidence. It will also enable subsequent investigators, prosecutors and fact finders to gain a better understanding of a case. The software will be built as an extension to evidence data modelling software currently used in six Dutch police regions. The effects of the software on an investigator's understanding of a case and the quality of file transfer will be measured in detailed user experiments.

The Netherlands Institute for Advanced Study in the Humanities and Social Sciences (NIAS) aims to stimulate research in the humanities and social sciences and to promote interdisciplinary co-operation in an international setting. In 2006 - 2007, Rineke Verbrugge, together with Jan van Eijck of the Centre for Mathematics and Computer Science in Amsterdam, has led a theme group at NIAS, Games, Action, and Social Software.

The purpose of the project has been to bring together a group of creative researchers from philosophy, logic, computer science and economics, in order to investigate the logical, computational and strategic aspects of social mechanisms. The aim has been to arrive at an interdisciplinary perspective, and help create an active community with a definite agenda. This aim has been achieved and the project has led to at least four books and numerous journal and conference publications.

The ANITA Project stands for Administrative Normative Information Transaction Agents. The use of electronic databases meets some important challenges: How to handle the shortage of information (not being able to find legally relevant data that should be available) as well as abundance of information (which may violate privacy rights)? These issues are of great consequence in the domain of police and judicial intelligence. A key tool for reasoning about the quality of communication and proliferation of information is the agent paradigm.

In this research, we aim to design and build an agent framework that enhances the application of norms. Although this framework is intended to be generic, a proof of concept system will be built for the police and judicial intelligence domain and will be tested in practice. Past graduate student Wouter Teepe was one of the initiators of the project. His promotor was Lambert Schomaker, and his daily supervisor and co-promotor was Rineke Verbrugge.

Theory of Mind in Corvids: Insights from Modeling (2007 - 2011)

Making

Sense

of

Evidence

(Current)