Cognitio 2011

Abstract: Is artificial intelligence (AI) possible? Via a quick tour of AI and cognitive science history, we will try to identify the turning points that established their bases and favored their evolution. Computationalism teaches us that the biggest problem facing Turing machines (Turing 1936) is not so much the fact that they might not be adequate to solve all problems, but more importantly that they require algorithms and, therefore, are not sufficient to explain intelligence. In other words, where do algorithms come from? And even before that, is intelligence algorithmizable?



The physical symbol systems hypothesis: "A physical symbol system has the necessary and sufficient means for general intelligent action." presented by Newell and Simon (1976) was highly instrumental for the elaboration of a theory of mind based on information processing and modeled on computer operation, but also failed to explain intelligence due to the lack of symbol grounding (Harnad 1990). In other words, where do symbols come from? However, Newell and Simon materialized Turing’s machine by the physical constraints of their systems and introduced the notion of agency (intelligent action) into AI and cognitive science.



More recent approaches (Brooks 1986/1990/1991, Arkin 1998, Pfeifer et Scheier 1999) focused on the autonomy of intelligent agents and highlighted the importance of embodiment and situatedness. On that basis, Steels and Brooks (1995) and

Brooks and Maes (1994) proposed that artificial life might be the only route to the emergence of AI. These concepts of embodiment and situatedness shed some light on the symbol grounding problem since sensors (the elementary ancestors of senses) are capable of semiotic perception long before consciousness interprets the signals into representations and/or symbols. Before reaching the symbolic level, autonomous systems must first be semiotic or, in other words, be able to “discover” signs (i.e. information) in surrounding physical signals.



Situated embodiment is generally seen as being closer to connectionism than to computationalism, but the former does not imply any rejection of the Church-Turing-Post thesis which is fundamental for the latter. The difference is rather at the algorithm level. While representationalism posits that intelligence is directly algorithmizable, connectionism tries to algorithmize the brain. The basic hypothesis becomes neurological or biological, while the psychological one is replaced by the probability of seeing intelligence emerge from a (computational) simulation of neuronal mechanisms.



In parallel, biologists, Maturana and Varela (1980, 1992), suggested that cognition and life resulted from a single process referred to as autopoiesis. Most important from our perspective is the fact that an autopoietic process is a process having as a product the process itself. In other words, life creates and maintains life while cognition, participating in this process, is also self-generating and self-sustaining. Any simulation based on a biological hypothesis must necessarily take this constraint into account. To be biologically plausible, an artificial neuron network shall therefore contribute to its own genesis and development.



Pushing Newell and Simon’s hypothesis somewhat further, we propose that only autopoietic semiotic systems have the necessary and sufficient means for general intelligent action.



On that basis, we will present a (computational) model of dynamic, analog and asynchronous neurons which, when networked, are sufficient to simulate an autopoietic semiotic system. Finally, we will describe the experimental methodology we intend to follow to show that such autonomous networks are capable of intentional and decisional behaviours, at least from the observers’ point of view.



__________________

Arkin, R.C. (1998) “Behaviour-based Robotics”, Cambridge MA: MIT Press.

Brooks, R.A. (1986) A robust layered control system for a mobile robot. IEEE Journal of Robotics and Automation, RA-2, 14-23

Brooks, R.A. (1990) Elephants Don't Play Chess, Robotics and Autonomous Systems 6, 3-15.

Brooks, R.A. (1991) Intelligence without representation, Artificial Intelligence, 47:139-160

Brooks, R.A. and Maes, P. (editors) (1994) “Artificial Life IV: proceedings of the fourth international workshop on the synthesis and simulation of living systems”, Cambridge MA: MIT Press.

Harnad, S. (1990) The Symbol Grounding Problem. Physica D 42: 335-346.

Maturana, H.R. and Varela, F.J. (1980) ”Autopoiesis and Cognition: The Realization of the Living”, Kluwer, Dordrecht, Holland

Maturana, H.R. and Varela, F.J. (1992) “The Tree of Knowledge: the Biological Roots of Human Understanding”. Boston, MA: Shambhala

Newell, A. and Simon, H.A. (1976) Computer Science as Empirical Inquiry: Symbols and Search (Chapitre 4) in Mind Design II J. Haugeland (éditeur) (1997) MIT Press, Mass., USA

Pfeiffer, R. and Scheier, C. (1999) "Understanding Intelligence", Cambridge MA: MIT Press.

Steels, L. and Brooks, R.A. (editors) (1995) “The Artificial Life Route To Artificial Intelligence: Building Embodied, Situated Agents”, Lawrence Erlbaum, Hillsdale, NJ

Turing, A.M. (1937), On Computable Numbers, with an Application to the Entscheidungsproblem, Proceedings of the London Mathematical Society, 2 42: 230–65, (et Turing, A.M. (1938), On Computable Numbers, with an Application to the Entscheidungsproblem: A correction, Proceedings of the London Mathematical Society, 2 43:544–6