Abstract

Because thought is a process involving representations, philosophers have tried to understand it by invoking metaphors rooted in other known forms of representation. The two most common metaphors for thought are the language metaphor and the picture metaphor. Any viable theory regarding the representational format of thought will need to account for the central characteristics of thought. Foremost amongst these is the capacity to manipulate representations in a truth-preserving manner. It is on the basis of such representational manipulations that we are able to predict the consequences of our actions. In addition, a viable theory will need to account for the representational expressiveness, or productivity, of thought. Early artificial intelligence researchers recognized that artificial languages such as predicate calculus are expressive enough to represent any of a wide variety of domains and provide a way to effect truth preserving representational manipulations. Philosophers in favor of the language/logic metaphor often take note of this fact about artificial languages. Early AI research also revealed, however, that this approach is seriously flawed. The problem with formal deduction techniques, called 'the frame problem', is that they do not support predictions regarding an open-ended set of alterations to a given represented domain. They exhibit representational productivity but not inferential productivity. An alternative to the language/logic metaphor for thought is the physically isomorphic model metaphor. A representation is physically isomorphic with what it represents if it has some of the very same properties as what it represents---as, for instance, do scale models, sculptures and pictures. Not only do the media used for the construction of PIMs often exhibit a high degree of representational productivity, but PIMs themselves exhibit inferential productivity. The most serious objection to the PIM metaphor stems from the recognition that systems like brains and computers fail to outwardly evidence any of the properties of PIMs . However, the same conceptual tools brought to bear in order to give a mechanistic reading to the language/logic metaphor can also be used in order to provide a more mechanistic reading of the PIM metaphor for thought