Pablo Medina

Persuasion-Aware Systems for Dialogue-Based Negotiation

Project Summary

Goal-oriented software systems interacting with people through dialogue are becoming ubiquitous. Today, an increasing number of chat-bots and automated personal assistants are acting as intermediaries between users and service providers — or become the service itself–, adding value to society via personalisation, enhancement, time and cost efficiency, and accessibility. Applying dialogue systems to the negotiation domain brings new commercial opportunities such as delegated negotiation or skills training, but is also of academic interest. In the context of negotiation, recent advances in Natural Language Processing have supported the development of neural end-to-end dialogue systems that learn to negotiate from human conversations.

Humans use a wide variety of strategies when trying to reach agreements, including persuasion.  In negotiations, persuasive tactics attempt to influence the opponent to modify their beliefs, actions, and attitudes. Despite the generalised presence of persuasive language in politics, commercial activities, and the media, its controversial nature makes it subject to ethical debate, especially when applied to autonomous technologies. Therefore, it is crucial to understand the ethical limitations of persuasive systems and develop methods to interpret and control the learned strategies. The main goal of this project is to explore how the automated classification and generation of persuasive language can improve the effectiveness and realism of data-driven dialogue negotiation systems while providing interpretability and control. Given the controversial nature of persuasion, this research will include an ethical analysis of linguistic persuasion strategies to help integrate human values and moral norms into AI systems for negotiation.

Research Interests

Experience-based decisions under uncertainty.

Cognitive models of judgment and decision making.

Reinforcement Learning.

Bounded and environmental rationality.

Background

BSc Economics at University of Almeria.

MSc Finance at Carlos III University of Madrid.

MSc Economics (Behavioral and Experimental Economics) at University of Granada.

Supervisors

Prof Özgür Şimşek

Dr Harish Tayyar Madabushi

Dr Subhasish Chowdhury