Grounding Symbols through Sensorimotor Integration
چکیده
A prominent robotics professor surprised me at last year's rsj conference: \There isn't really a symbol grounding problem for robotics, is there? I often ask people, `Is symbol grounding a problem for your re-search?' and no one says, `Yes.' " Sensing irony in his voice, I replied, \That's because no one is building systems with a human | or even vertebrate | level of competence. When they try to, they may respond to your question dierently." Harnad (1990) identied the symbol grounding problem as a problem of embodying symbol systems (Newell 1980). Symbol systems are usually comprised of a xed set of elementary symbols and rules for combining them into more complex representations. Symbol systems are designed to manipulate these representations according to their constituent structure (i.e., syntax) in the hope that these transformation make sense semantically. The problem is: How do we causally connect a symbol system's internal symbols to the external objects, events, and relations they are supposed to rep-resent? Solving this problem is not simply a matter of learning some kind of grounded internal representation in a exible and adaptable way. We are already building emergent systems that can do that (e.g., Uchibe et al. 1998), and articial neural networks (Smolensky 1988), reinforcement learning (Sut-ton & Barto 1998), and behavior-based architectures (Brooks 1991) oer three popular approaches. But what the current generation of emergent systems generally lacks are three key aspects of thinking: its systematicity, productivity, and inferential coherence. They lack these aspects because their underlying methods have been unable to deal eectively with constituent structure, though more elaborate implementations should overcome this limitation (see, for example, Chalmers 1993). We tend to think of these aspects of thinking as being typically human. This is probably because they were rst studied in relation to language (Katz & Fodor 1963; Chomsky 1959, 1965). Human thinking is systematic in-sofar as people who can understand one sentence John loves Mary) can, in general, understand structurally similar sentences (Mary loves John); it is productive insofar as people can understand and generate an unbounded number of sentences; and it is inferen-tially coherent insofar as people are capable of forming valid conclusions from valid premises. Nevertheless, Fodor and Pylyshyn (1988) have shown that the same arguments that have been applied to language understanding can be extended to belief in general and even nonverbal perception and cognition in animals. For example , it seems reasonable …