Flexible human-agent dialogues for social training based on BDI models

From Master Projects
Jump to: navigation, search


About Flexible human-agent dialogues for social training based on BDI models

  • This project has been fulfilled.
  • This project fits in the following Bachelor programs: {{#arraymap:|, |xXx|bachelorproject within::xXx|,}}
  • This project fits in the following masterareas: {{#arraymap:Knowledge Technology and Intelligent Internet Applications, Computational Intelligence and Selforganisation, Cognitive Science, Human Ambience, Technical Artificial Intelligence, AI and Communication, Internet and Web Technology, Multimedia, Software Engineering, Information and Communication Technology, Computer Science and Communication, Information Sciences|, |xXx|project within::xXx|,}}


Description

|free text=}} Various professional educations offer their employees dedicated training in social skills, like eliciting customers’ needs and preferences through conversation, negotiating about conflicting interests and objectives with contractors, or dealing with aggressive behavior of others. Currently, social skills training often takes the form of role play sessions, but as role playing sessions with real actors are often costly, time-consuming, and hard to organize, training based on serious games is an interesting alternative.


Serious games for social skills training typically make use of dialogue systems that enable a player to engage in conversations with a virtual character. The traditional method to implement such a dialogue system is based on conversation trees: tree structures in which all possible developments of a conversation are specified exhaustively. Although this method is easy to use, it is not very flexible. In particular, conversation trees do not facilitate realistic behavior of the virtual conversation partner without prescribing in advance how the character should react for all possible world situations.


An alternative to the use of conversation trees is to make use of BDI agents as the underlying mechanism for the dialogue system. By using a BDI model, the mental state of the virtual interlocutor can be maintained during a conversation. Next, this mental state can serve as input for the agent’s (verbal) responses. With the help of this method, one can develop a larger variety of training scenarios with less effort.


The goal of this project is to develop a prototype of a training system for social skills, with an emphasis on aggression de-escalation in the public transport domain. As a basis for the prototype, the Multi-Agent Director System (MADS) will be used, a multi-agent system enabling developers to create personalized training scenarios. To develop the prototype, first a number of scenarios will be established within the domain of aggression in public transport. Next, these scenarios will be implemented within the MADS architecture, where the behavior of the virtual conversation partner will be modeled based on BDI models, possibly extended with relevant mental states, such as emotions. Finally, the functionality of the prototype will be evaluated by means of a series of user studies in which participants are asked to interact with the system and report on their experience.


We are looking for a motivated student that satisfies the following requirements:

• Enrolled in an MSc program in Computer Science, Artificial Intelligence, or a related area

• Good programming skills (experience with Java is a pre)

• Knowledge about intelligent agents and the BDI paradigm

• Affinity with the social sciences, in particular the topic of interpersonal communication

• Creative attitude


This project is a collaboration between Delft University of Technology (Interactive Intelligence Group – ii.tudelft.nl) and Vrije Universiteit Amsterdam (Behavioural Informatics Group - http://behaviouralinformatics.nl/), and takes place within the context of the STRESS project (stress.few.vu.nl).


Supervisors:

• Dr. Tibor Bosse (VU) - t.bosse@vu.nl

• Dr. Marieke Peeters (TU Delft) - m.m.m.peeters@tudelft.nl