Main Page
From Wikipedia
p9anP6 <a href="http://fggdpkrpftfh.com/">fggdpkrpftfh</a>, [url=http://svpttdbwufwm.com/]svpttdbwufwm[/url], [link=http://bdrtnhrblhks.com/]bdrtnhrblhks[/link], http://uytzegrpbeyq.com/
Contents |
ACT-RBOT : Introduction & History
The ACT-RBOT (agent) project was started at the University of Groningen in an attempt to create a 'Unified Social Cognitive Agent Theory'. Its roots are in the field of Management and Organisation; a PhD project first aimed at simulating behaviour and change of organisations based on a quick to implement stimulus-response (behaviourist approach) agent.
Although an implementation of a social & organisational model could be a surplus for the field of organisational behaviour and social simulation, the main problem of the simulation would be to explain change at the level of the individual. Cristiano Castelfranchi (2001) points out that a significant theoretical problem exists in the field of social sciences; there is a lack of understanding or explanation of unconscious, unplanned forms of cooperation among intentional agents. The second problem is the grounding of social behaviour. Scott Moss (2006) states claims that agent-based models can capture independent validation by "..ensuring that the specification is well verified with respect formal models from cognitive science understanding that those formal models are themselves well validated experimentally and observationally. Presuming that cognitive science is good science...[]...then the verification of agent designs with respect to cognitive science by adopting SOAR and ACT-R specifications supports good social science via social simulation.
Back in 2003, we (Martin Helmhout, René Jorna and Henk Gazendam), were exactly struggling with this problem. The theory of ACT-R seemed to us (still) the best theory with strong empirical grounding. Besides that we also studied SOAR and adopted some of their internal graph structures into ACT-RBOT. Hence, the project should be called ACT-RBOTsoar. In general we speak about ACT-RBOT or RBot, referring to the same architecture.
What has / is ACT-RBOT?
- a production system
- a hybrid architecture (symbolism and connectionism/ similar to ACT-R)
- a social situated agent (support for norms, social constructs)
- support for (symbolic) subsumption architecture
What has / is ACT-RBOT NOT (yet)
- a BDI architecture like JADEX, although we soon will integrate concepts of JADE behaviours by letting RBot inherit the jade agent
- ACT-R: although the agent model follows mainly the concepts of ACT-R, thereby inheriting the foundations of ACT-R, RBot is not tested with cognitive psychological experiments
- SOAR: RBot and SOAR (and ACT-R) are all production systems. SOAR and ACT-R are concentrating on the individual, RBot's intentions are to concentrate on social embeddedness / social constructs
- a quick learning curve. Regretful: in order to understand production systems, we advice to get a general understanding of production systems by exploring the possibilities of ACT-R and SOAR.
Main goals:
- create an agent that brings together connectionism, symbolism, embodied cognition and social cognition. Hence, the goal is to connect (different streams of) cognitive science (AI) and social science
- create a minimal memory model with maximum flexibility
- strive to reach a "Model Social Agent", see "The Nature of the Social Agent" Carley & Newell 1994
References:
- Carley, K. M. & Newell, A. (1994), "The Nature of the Social Agent", Journal of Mathematical Sociology 19 (4): 221-262.
- Castelfranchi, C. (2001), "The theory of social functions: challenges for computational social science and multi-agent learning", Cognitive systems Research 2: 5-38.
- Moss, S., "Cognitive Science and Good Social Science", in Sun, Cognition and Multi-Agent Interaction, MIT Press, pp. 393-400.
