Dr. Eric M. Kok
About me
I was a PhD student with the Intelligent Systems group at the
Utrecht University in The Netherlands. Here I worked on the AIDA project
'Agents Interacting in Dialogues with Argumentation', which was supervised by
Prof. Dr. Mr. Henry Prakken, Dr. Gerard Vreeswijk and
Prof. Dr. John-Jules Meyer. This project focused on testing
the applicability of argumentation in multi-agent interactions. Do the apparent benefits hold in a practical
implementation and how and on which points can effectiveness be measured? I successfully defended my thesis in May 2013 to receive the degree of doctor.
Research papers and presentations
-
Baidd Testbed
The software testbed for experimentation with arguing agents BAIDD: BDI Agents Interacting in Deliberation Dialogues is available as open-source software at my erickok/baidd BitBucket repo. The code is released under the GNU GPL v3 copyleft license. The included AspicInference was released under a custom ASPIC license.
-
PhD Thesis
Exploring the practical benefits of argumentation in multi-agent deliberation
On the 21st of May I successfully defended my PhD thesis and got awarded the scientific grade of doctor.
full thesis PDF (printer version) - Dutch summary -
AAMAS 2012
Testing the Benfits of Structured Argumentation in Multi-Agent Deliberation Dialogues
Eric M. Kok, John-Jules Ch. Meyer, Henry Prakken and Gerard A.W. Vreeswijk
To improve communication and shared decision making in multi-agent systems it is often proposed to allow for argumentation in inter-agent dialogues. Throughout the years many frameworks and protocols have been developed and the theoretical reachability of ideal and intuitive outcomes has often been proved formally. However, since not all properties can be studied formally at least three works have experimentally explored the benefits of argumentation in dialogues. (Karunatillake et al. 2009, Pasquier et al. 2010 and Black and Bentley 2011) On the other hand, none of these studies have captured the expressivity of formal models of argument based inference. They particularly lack a language in which arguments with internal structure can be used to cover realistic argumentation dialogues.
extended abstract paper (conference version) - poster -
ArgMAS 2012
Testing the Benfits of Structured Argumentation in Multi-Agent Deliberation Dialogues
Eric M. Kok, John-Jules Ch. Meyer, Henry Prakken and Gerard A.W. Vreeswijk
Work on argumentation-based dialogue systems often assumes that the adoption of argumentation leads to improved dialogue efficiency and effectiveness. Several studies have taken an experimental approach to prove these alleged benefits, but none has yet supported the expressiveness of a structured argumentation logic. This paper shows how the use of argumentation in deliberation style dialogues can be tested while supporting goal-based agents that use the ASPIC framework for structured argumentation. It is experimentally shown that employing an arguing strategy increases the effectiveness over a non-argumentative strategy.
full paper (conference version) - presentation -
LAF 2012
A workshop presentation Practical and Epistemic Reasoning in Argumentation Dialogues: From Theory to Practice
presentation -
EUMAS 2011
A Methodology for the Generation of Multi-Agent Argumentation Dialogue Scenarios
Eric M. Kok, John-Jules Ch. Meyer, Herre van Oostendorp, Henry Prakken and Gerard A.W. Vreeswijk
Increasingly research into the uses of argumentation in multi-agent dialogues takes an experimental approach. Such studies explore how agents can successfully employ argumentation besides the best and worst case situations of formal analysis. While a vital part in these experiments is influenced by the scenarios from which dialogues are generated, there is very little research on how these can be generated in a meaningful way, respecting the characteristics of the underlying dialogue problem. This paper proposes, by means of an example system for deliberation dialogues, a methodology for the construction and evaluation of a scenario generation process. It is shown how scenarios can accommodate argumentation with structured arguments and how it is tested whether the generated scenarios are interesting for experimentation.
full paper (revised proceedings version) - presentation -
LAF 2011
A workshop presentation On Testing the Use of Argumentation in Deliberation Dialogues
presentation -
ArgMAS 2010
A Formal Argumentation Framework for Deliberation Dialogues
Eric M. Kok, John-Jules Ch. Meyer, Henry Prakken and Gerard A.W. Vreeswijk
Agents engage in deliberation dialogues to collectively decide on a course of action. To solve conflicts of opinion that arise, they can question claims and supply arguments. Existing models fail to capture the interplay between the provided arguments as well as successively selecting a winner from the proposals. This paper introduces a general framework for agent deliberation dialogues that uses an explicit reply structure to produce coherent dialogues, guides in outcome selection and provide pointers for agent strategies.
full paper (proceedings version) - presentation -
Master thesis
Adaptive reinforcement learning agents in RTS games
Eric M. Kok
Supervised by Joost Westra and Frank Dignum
Computer game AI nowadays still relies heavily on scripted behaviour. The resulting computer players are therefore predictable and non-adaptive to the opponent. An approach to create challenging, learning computer players, called Dynamic Scripting, has been proposed by Pieter Spronck. Based on this work I developed reinforcement learning agents for a real-time strategy game called Bos Wars. These agents, implemented using the 2apl platform, can autonomously reason on goals and world beliefs. Several approaches to acquire a winning strategy have been tested, including Dynamic Scripting and Monte Carlo methods. To better cope with opponents that switch strategies, implicit and explicit adaptation is tried out. Due to the integration of agent technology and reinforcement learning, agents have proven to be able to quickly and consistently learn to outperform the scripted fixed and strategy switching players. It is shown that incorporating opponent statistics into the learning process of a Monte Carlo agent gives the best learning results.
full master thesis
