Embodying Computation at Higher Types

Abstract

Since Turing’s early speculations, progress with the building of intelligent machines has been slow, with obstacles tending to be dealt with in ad hoc rather than theoretical ways. This view of history is reflected in Marvin Minsky’s comment (Boston University, 2003) that “AI has been brain-dead since the 1970s”. The optimism of Turing’s reported “I am building a brain” (Hanslope Park, 1944) has withered before the reality of the chimera that is embodied intelligence. There is a growing sense that the universal Turing machine has given rise to a paradigm (McCulloch and Pitts (1943), Kuhn (1996)) which, as well as being powerful and successful, is misleading in important respects: Firstly, the disembodiment implicit in the Turing machine was unreal even for the early construction of a digital computer, and this has become more apparent with the modern respect given to natural computing in its various guises. And more abstractly, this has made us more aware of the extent to which theory has failed to give due prominence to the role and structure of information in the computational processes that science seeks to characterize. The theme of this article is the extent to which embodiment, information, and its type structure are relevant to the material and mental world, and their computational processes. In particular, it is argued that ‘computation’ in the real world is more general—and consequently harder to control—than is usually imagined.