excogitate‎ > ‎

Turing test

posted 16 Feb 2012, 15:21 by John Brown   [ updated 18 Apr 2012, 22:04 ]

The now classic Turing test challenges a machine to masquerade as a human intelligence using natural language to interact with a human judge. This provocative test compares the capabilities of an 'aspiring' machine to the only readily agreed example of intelligence (a human).

It may be tempting to take humans as the definition of intelligence, rather than the only readily agreed example of intelligence. If so, then machine intelligence is doomed to inferiority, because any difference what-so-ever in the intelligence portrayed by the machine is, by definition, a defect. By this definition, the machine could only ever approach, but never surpass, human intelligence. How can anything be more human than human?

To make matters even worse (for the machine), the Turing test does not recognise the highly variable nature of human intelligence. If a machine were to masquerade convincingly to one particular judge, then would the machine also be convincing to another judge of different gender, culture, education and language? If humans are taken as the definition of intelligence, then it is a shimmering unobtainable goal.

This may be comforting to those with a homocentric prejudice for the seat of intelligence, but I do not consider it to be a useful or interesting perspective. I do not see the point in attempting to artificially simulate a human, especially when construction of the genuine article by traditional methods has so much going for it.

Those machines that significantly extend our limited human abilities, rather than merely offering a labor saving alternative, are the more interesting machines to me.

The Turing test has highlighted the often homocentric prejudices surrounding intelligence and the slippery nature of the very notion of intelligence. It is no surprise that attempts to pin down the notion of intelligence with natural language have not arrived at a satisfying result since Whittgenstein demonstrated quite clearly that words cannot be precisely defined. Probably more importantly, Whittgenstein also demonstrated that we don't need a precise definition of words in order to use them successfully in natural language.

A precise mathematical definition of intelligence is required before an Intelligent device can be Engineered. The absence of even the possibility of a precise definition for intelligence in natural language is an important pointer to the nature of human intelligence.

When the definition of intelligence is isolated from a homocentric base, then there can be no expectation that an intelligent device will be recognisably human in nature. Furthermore, a successful intelligent device may well exceed some aspects of human intelligence or be intelligent in areas that humans are bereft. Alien Intelligence is a more productive term than Artificial Intelligence.



"No one has built a machine or designed a program that 
comes close to this capability [passing the Turing test]. 
Whether this is a problem in principle, as some philosophers insist, 
or simply a problem in practice, as artificial intelligence proponents insist, 
will be debated for a long time. The debate, however, is sterile 
- it produces no insights into how to proceed to build more intelligent machines, 
something that even the most critical philosophers think is possible.
Pagels (1988) The Dreams of Reason

"... the improvement of computers will advance the 
construction of actual artifi­cial intelligences. 
But these will not resemble human intelligence 
any more than wheels resemble human legs.
Pagels (1988) The Dreams of Reason