There are a variety of approaches to the question of what aspect of human intelligence to attempt to simulate or replicate with a computer system. The basic division, which is particularly germane for brain scientists, is whether the approach is to reproduce functional equivalence with no regard for the mechanisms and structures of the human brain or whether it is important to reproduce structure as well as function.

A. Functional Equivalence

Within a functional equivalence approach to AI the goal is to reproduce (or supercede) some aspect of human intelligence, such as medical diagnosis or chess playing, by using whatever means are available. It is believed, for example, that the best human chess players examine only a few of the many possible move sequences that may be available at any given point in a game. The success of chess-playing computers has been founded on very fast exploration of thousands of possible move sequences. The result is very high-quality chess games from these machines but based on

Encyclopedia of the Human Brain Volume 1

Copyright 2002, Elsevier Science (USA).

All rights reserved.

mechanisms that would not appear to be used by humans who exhibit equally impressive chess-playing behavior.

The bulk of work in AI has taken this purely functional approach for two good reasons. First, the mechanisms underlying human intelligence tend to be options for debate rather than secure knowledge. At the cognitive level, the mechanisms and structures, such as short-term and long-term memories, are by their very nature difficult to describe in detail with certainty. Also, although details of brain physiology may be asserted unequivocally, the uncertainty lies in their roles in the production of the aspects of human intelligence being reproduced. The second reason for neglect of structure and mechanisms is the suspicion that the human brain may not be an optimum architecture on which to build intelligence. Its basic design evolved long before it was required to support intelligent behavior, and there are many seemingly obvious weaknesses in the human brain when viewed as a computational machine, such as error-prone memory retrieval. Therefore, most of the work in AI must be viewed as attempts to reproduce some aspects of intelligence without regard for the way that the human brain might be thought to be operating to produce human intelligence.

1. Abstract Structure: Cognitive Science

The majority of AI work that has not eschewed structural equivalence has accepted a structural model of the brain at the cognitive level, and much of the AI work in this subarea is also known as cognitive science. Alternatively, most of the AI work in this subarea is aimed at evaluating and exploring cognitive models of human intelligence.

2. Physical Structure: Cognitive Neuroscience

There is a relatively small but growing body of work in AI that does accept constraints that derive from anatomical knowledge of brain structure and function. The impetus for this type of approach to AI has been the surge in interest and activity based on computation with neural networks. A neural network is a computational system that is founded on simple communication between highly interconnected networks of simple processing elements, i.e., a computational basis that appears to reflect some aspects of brain physiology. Conventional programming is also used to construct AI models of this type, but it tends to be restricted to the reproduction of a limited aspect of human intelligence. A system might, for example, attempt to reproduce certain aspects of human learning behavior based on a model of synaptic facilitation at the molecular level.

Anxiety and Depression 101

Anxiety and Depression 101

Everything you ever wanted to know about. We have been discussing depression and anxiety and how different information that is out on the market only seems to target one particular cure for these two common conditions that seem to walk hand in hand.

Get My Free Ebook

Post a comment