Representations and sensorimotor loops in intelligent agents
Massimo Tria ()
Department of Philosophy and Social Sciences, University of Siena
Via Roma 47, 53100 Siena, Italy
This paper examines the role of sensorimotor loops in current analyses of intelligent agents, focussing on Brooks’s “intelligence without representations” approach, and with a threefold aim.
First, commonalities between Brooks’s approach and methodological tenets of earlier cybernetic views are emphasized. These commonalities address two fundamental questions: first, the attempt to explain cognitive behaviours by referring to observable performance alone with no reference to mentalistic terms and concepts and second, by stressing the central role of an organism’s interactions with his own environment.
Second, these commonalities enable one to isolate some epistemological problems afflicting cybernetic accounts of human purposeful behaviours that are inherited by Brooks’s approach. The most relevant problem afflicting cybernetics and Behavior-Based Robotics approaches regards the possibility of modelling so called ‘high-level’ behaviours – like problem solving, abstract reasoning and language – with explanatory tools that seem to be suitable only for reactive behaviours – or in general to behaviours that involve processes of physical adaptation to an environmental niche.
Third, behavioural studies of various insect species suggest that the austere set of explanatory tools adopted by Brooks is insufficient even just to account for “low-level” behaviours performed by simple insect-like organisms. Empirical data gathered in particular from experiments on the desert ant Cataglyphis show how it seems to be necessary invoking theoretical constructs like symbolic representations because of the evident capability of this insect to be independent from external stimulations and local cues in order to perform his own feeding-related behaviours. These results can be interpreted as the suitability and the importance of cognitive psychology and Artificial Intelligence conceptual apparatus in explaining even the so called adaptive behaviours, that are the actual theoretical core of cognitive sciences researches. I will briefly expose the works of Maja Mataric that within a Behavior-Based perspective offers a critical analysis of Brooks’s approach claiming that in fact Behavior-Based models and agents need some sort of internal representing state – in other words, mental/symbolic representation – in order to accomplish their tasks.
The opportunity of suitably amalgamating methodologies underlying traditional AI and behaviour-based systems emerges from the confluence of these various observations.
1