Autobiographic Agents - ECIRCUS - EU Framework VI Project - Education through Characters with emotional-Intelligence and Role-playing Capabilities that Understand Social interaction

In the eCIRCUS project, we aim to model the psychological concept of autobiographic memory computationally and integrate it into our synthetic agent architectures. With this memory included, agents are not only capable of recognising and ranking significant events which originate in the agents’ own experiences, but can also remember, recall and learn from these experiences. Thus we expect that the agents’ believability can be increased and the interactivity of the software we develop in the eCIRCUS project can be more fulfilling for the user.

In Psychology, autobiographic memory is a specific kind of episodic memory that contains significant and meaningful personal experiences for a human being. Theoretically, an autobiographic agent is an embodied agent which dynamically reconstructs its individual history (autobiography) during its life-time [1]. This individual history helps autobiographic agents to develop individualised social relationships and to form communications, which are characteristic of social intelligence, and it may also lead to more appealing and human-like engaging interactions, making them more pleasant and acceptable to humans.

Different types of computational memory architectures for Artificial Life autobiographic agents have been developed and experimentally evaluated in our previous research work, for an overview, see [2]. These architectures include typical human memory modules which are commonly acknowledged in Psychology: short-term, long-term, and positively and negatively categorised memories. Agents embedded with these computational autobiographic memories outperform Purely Reactive agents that do not remember past experiences in surviving in both static and dynamic environments. Furthermore, it is evident that enabling narrative story-telling in Long-term autobiographic memory control architecture as an additional communication feature helps agents to be more adaptive in coping with different environmental dynamics.

In the paradigm of developing synthetic agent architectures, we proposed that 1) knowledge representations in the computational autobiographic memory can be based on general episodes that agents have experienced and 2) goal structure, emotion, and attention processes, support and are influenced by, autobiographic knowledge [3]. Autobiographic knowledge may also support long-term development and learning in synthetic agents as they gain new experience from acting in each new situation. Therefore, character-based narrative story-telling systems can benefit from agents with autobiographic memory.

In addition to the basic characteristics of human autobiographic memory (e.g. event reconstructions and influences on emotion in reasoning), as already implemented in the current FearNot!v2 agent architecture, we are going to further utilise the capacity of this memory to enable agents to form their unique “self” through organising abstract knowledge in long-term memory based on major goal attainments in the past. The relationship between autobiographic memory and general event representations in manipulating current goal structures can be implemented in the ORIENT system for agents’ to generate appropriate planning and coping strategies [4].

References:

[1] Dautenhahn, K. (1996) Embodiment in animals and artifacts, AAAI FS Embodied Cognition and Action, AAAI Press, pp. 27–32. Technical report FS-96-02.

[2] Ho, W. C. (2005) Computational memory architectures for autobiographic and narrative virtual agents, PhD Thesis, University of Hertfordshire.

[3] Ho, W. C. and Watson, S (2006) Autobiographic knowledge for believable virtual characters, Intelligent Virtual Agents 2006 (IVA 2006), Springer LNAI, pp.383 – 394.

[4] Ho, W. C., Watson, S. and Dautenhahn, K. (2006), 6.1.1: Knowledge Representation in Computational Autobiographic Memory for Believable and Intelligent Agents. eCIRCUS Technical report.

Created on 10/23/2006 11:44 AM by ecirweb
Updated on 10/23/2006 11:46 AM by ecirweb