Wed, Jun 07, 2017 @ 01:00 PM - 02:00 PM
PhD Candidate: Zahra Nazari
Date: Wed, June 7th
Time : 1 PM
Dr. Jonathan Gratch
Dr. Milind Tambe
Dr Peter Kim
Title : Automated Negotiation with Humans
Negotiation is a crucial skill in personal and organizational interactions. In the last two decades, there has been a growing interest to create agents that can autonomously negotiating with other agents. The focus of this thesis, however, is on creating agents that can negotiate with human opponents. Besides improving on artificial social intelligence, such agents could be used for the purpose of training or assisting human negotiators. A central challenge is to handle the complexity of actual human behavior. When compared with idealized game-theoretic models,
human negotiations are far richer, both in terms of the nature of information exchanged and the number of factors that inform their decision-making.
We consider a negotiation task that is simple, yet general enough to drive agent-human research, and
analyze an extensive data set of transcribed human negotiation on such tasks.
Based on human behavior in this task, and the previous research on human negotiations, we propose a new framework to structure the design of agents that negotiate with people. We address two main decision problems inspired by this framework: modeling and influencing the opponent. Three techniques are proposed to model an opponent's preferences and character (e.g. honesty and personality traits) and a misrepresentation technique is then used to influence the opponent and gain better profit. The proposed techniques are then implemented in automatic web-based agents. We ran a number of negotiations between these agents and humans recruited on Amazon Mechanical Turk. The resulting data show that the agents can perform these strategies successfully when negotiating with human counterparts and give us valuable insight about the behavior of humans when negotiating with an agent.
Location: Kaprielian Hall (KAP) - 134
Audiences: Everyone Is Invited
Contact: Lizsl De Leon