Possible BIS (Hons) or Masters Research project, 2008

Empirical study of Interest Based Negotiation in a Human-Computer Environment.

Context: Argumentation is receiving increasing interest in artificial intelligence as a convenient metaphor for modelling and automating reasoning and interaction among autonomous software agents. Computational models of argumentation rely on rules for generating and evaluating arguments based on abstract relationships (e.g. characterising defeat or preference) among arguments. Such "argument evaluation rules" have been mainly developed for the purpose of obtaining desirable formal or computational properties, based largely on authors' intuitions. A long-standing question is whether these rules are plausible models of how humans actually evaluate arguments. Understanding the cognitive plausibility of argument evaluation rules is crucial if we were to build software agents able of interacting persuasively with humans through argument.

Project background: Interest based negotiation is a novel model of bilateral negotiation extending simple monotonic concession protocols by allowing the negotiators to exchange information about their underlying interests and possible alternatives to achieve them during the negotiation. This project is about inquiring the properties of this novel negotiation technique and the associated protocols in the case of human-computer negotiations. The research would involve conducting experiments with human subjects interacting with a software system which generates relevant scenarios. The project is related to the Australian Research Council funded project on "Interest Based Negotiation" http://www.dis.unimelb.edu.au/staff/lizs/ibn/index.htm Colored Trails http://www.eecs.harvard.edu/ai/ct/ is a research test-bed for use as a community resource for the study, modeling, and development of decision-making strategies appropriate for multi-agent human and computer task settings.

Project description: This research project involves:

• getting familiar with the formal theory of interest-based negotiation
• getting familiar with the colored trails framework
• modelling some IBN negotiation scenario in colored trails (including developing the appropriate artificial agents' strategies)
• conducting some human-computer (and possibly some human-human and computer-computer) simulations of negotiation
• analysing the data gathered through the experimentation

Essential background: - a "mid" level of programming ability (to enable you to do modification/tuning of software written by others)
Desirable background: some prior study in mathematics, philosophy, or psychology.

Primary supervisor: Liz Sonenberg, http://www.dis.unimelb.edu.au/staff/lizs
Co-supervision would be provided by Dr Iyad Rahwan, British University in Dubai http://homepages.inf.ed.ac.uk/irahwan/ (Iyad is a PhD graduate from Melbourne University).

Related work:

• Work by Kobi Gal, Harvard University, on modelling reciprocity in human negotiation - they came up with a model of recipricity in games and designed an agent that learns the parameters of this model, showing that it actually generates good predictions of actual human reciprocal behaviour http://www.eecs.harvard.edu/~gal/repeated.pdf
• Work by a group at Bar Ilan University, Israel including a formal model, analysis, and experiments with humans and agents that use a combination of game-theoretic and heuristic strategies http://www.cs.biu.ac.il/~sarit/Articles/linetal-nego-AIJ07.pdf