Archive for the Assignments Category

The relevance today of Turing’s “Computing Machinery and Intelligence”

Posted in Assignments with tags , on September 7, 2008 by alessandrovalente

My study group will be doing a presentation  on Turing’s famouse article on AI. 

My part for the initial research for this presentation is thinking and writing, about how relevant the practical ideas of this article, from the year 1950, still are today. So while im at it, why not blog about it at the same time?

After reading the article and a bit of its context it seems to me that Turing’s aim with this article was more philosophical then a practical one. The main point of the article is raising the question “Can machines think?”. By raising this question and making the analogy between human intelligence and the computing power of a machine, Turing gave us material to challenge traditional believes about what intelligence is and perceived boundaries between  human vs machine and mind vs body. Still the article also contains  practical information, wich probably is the reason why most people know the article. Judging from how well known the concept Turing-test (AKA “Imitation game”) has become. This blog will look into how practically relevant Turing article is considered today. The practical part of Turing’s article I consider to be the piece about his “Imitation game” better known as the Turing-test.

 After a bit of research it became clear that there exists a disagreement about how relevant doing the Turing-test should be considered. At one hand there is the still on going annual Loebner prize competition for chatterbots wich gives away awards and prize money to programs based on how well they pass a Turing-test. The official website states that one of the main reasons for this annual prize is the advancement of AI.

On the other hand the test recieved alot of critique in the ninties from computer scientist and theorists. they feel that the  Turing-test is now useless for testing AI and that its implementation is a burden to the progression of AI developments today. According to them, the Turing test should be removed from the textbooks and put in the historybooks. The main problems they have with using the Turing-test is that:

1. Standard objections

The test is too difficult. Furthermore focusing on the Turing test would make research too bound up with natural language understanding. This means ignoring many other aspects of current AI research such as vision and robotics.

2. Setup of the test itself

Its setup is to poor for  any outcome of the test to be usefull. The outcome of a Turing-test doesn’t only say something about the intelligence of the computer but also about the skill of detecting inhuman like behavior by the interrogator. Therefore its outcome can’t be taken as a measurement of a programs intelligence.  Also the test is an Zero-sum game, this makes it unusefull to measure research progress.

3. Wrong definition of intelligence

The Turing-test focuses too much on an anthropocentric view on intelligence. For instance a dog could never pass the test, but we would not say that dogs do not exhibit a form of cognition. The main goal should be to make machines intelligent in any way possible and as intelligent as possible. Trying to copy human intelligence and behavioure is useless, we have enough of them already anyway.

(Hayes, Patrick & Ford, Kenneth (1995), “Turing Test Considered Harmful“)

Personally I agree with the camp that sees a problem in using Turing’s article as THE way to define and motivate AI research. Tho we should, and this brings me to the start of my evaluation, consider the aim of Turing’s text more of a philosophical nature then of a practical one. His description of the Turing test should be considered an effort to provoke programmers to make the leap in creating an intelligent machine. Not to propose definite rules of how we should define progress in AI. When making the first leap, it makes sense to part from what we know about our own brain and actions as an intelligent being and make it our objective to mimic them. This doesn’t mean however that we shouldn’t ever abandon ourselves from these objectives, to make further progress or take different routes within in the research field. At the same time I don’t see the need to ban the Turing-test all together because it contains an anthropocentric view on AI. Making machines that act like humans can have its uses also, if anything else, it’s fun.