Representing the Mind

Researchers in the fields of cognitive neuroscience and Artificial Intelligence are fond describing the brain as a highly sophisticated computational device in which data is encoded and processed. Prior to the invention and development of computers many of the most prominent theories of the "workings" of the brain assumed either that it was minutely mechanical in its functioning (tiny cogs and pulleys, weights and levers) or else – or also - that the physical matter of the brain was simply some form of interface between the brain and the non-physical mind. This “Dualistic” theory, as it came to be known, is most closely associated with the 17^th Century French philosopher Rene Descartes.

Since the 17^th Century, advances in scientific understanding have enabled us to be a great deal more certain that the mind is entirely physical in origin, yet vestiges of Cartesian Dualism continue to linger in many quarters despite the overwhelming evidence that consciousness – the mind - is a product of electrochemical processes occurring in – or largely in - the brain.

Throughout history, many differing conceptions of the origins of thought and intelligent agency have stocked our vocabulary with a rich variety of metaphors borrowed from a range of objects, activities and occupations. We talk of the mechanisms of the mind, of streams of consciousness and information flows, of brain cells lighting up, sparking and firing, of cascades, filters, causal chains and neural triggers.

Such terms are extremely useful and often unavoidable in formulating theories and characterising barely understood processes. But characterisations can also present significant difficulties, especially if taken too literally as conceptual models. When attempting to understand complex and currently unexplained phenomena we are inevitably restricted both by what we know and what we don't know. Our knowledge and the terminology we have to describe it frames our thinking in ways that do not always help to illuminate the issues or processes we seek to understand. The same was true for Descartes, and is no doubt the reason he rejected the mechanical model of mind in favour of a non-physical account. It is perhaps possible then, to see how the versatility and character of the analogies available to us can significantly influence the concepts we are likely to formulate and the conclusions we are likely to draw during our exploration of unfamiliar processes and phenomena.

Since the 1970's Deirdre Gentner’s research has focussed on the use of analogical reasoning in the domains of cognitive science and artificial intelligence in particular. Gentner’s findings show that “analogical access and analogical inference are governed by very different rules.” In other words, when we make comparisons we have a tendency to pick out superficially accessible analogies instead of analogies that might offer greater predictive potential. I think this is precisely what is happening when the functioning of the brain is explained by way of computational analogies. Even more worrying though is the assumption – frequently made by theorists and scientists - that brains necessarily utilise encoded representations. I would argue that the failure to recognise the incoherence of this analogy is having a significantly detrimental influence on the development of both cognitive neuroscience and artificial intelligence.

In their 1995 book “Foundational Issues in Artificial Intelligence and Cognitive Science” Mark Bickhard and Loren Terveen argue that the fields of Cognitive Science and Artificial Intelligence are “in the midst of a programmatic impasse.” They make the vital point that this impasse cannot be overcome by making changes – no matter how insightful - at the “project level.” That is to say; if the foundations aren’t sound, no amount of work above ground is ever likely to resolve the initial errors - in fact it is far more likely to compound them.

Bickhard and Terveen coin the term “Encodingism” to describe the assumption that cognition involves the use of representational encodings. They write:

“It is assumed that the symbol represents a particular thing, and that it - the symbol - somehow informs the system of what that symbol is supposed to represent. This is a fatal assumption in spite of its seeming obviousness.”

The point they are making is that it is by no means straightforward what any symbol is actually intended to symbolise. As C. S. Peirce theorised at the end of the 19^th Century, symbols represent their denoted objects as a result of rules or interpretive habits that are independent of any properties that may or may not be shared between symbols and the objects they represent. Symbols are things that require interpretation. They do not lead immediately and without complication to the things they denote. The rules and habits of symbol use have to be learned. If this is true, then it is extremely doubtful that an intelligence could emerge capable of interpreting symbols without a prior non-symbol-using evolution. And if intelligence is possible without symbol use, then no encoded symbols are required for cognition.