Turing's Test, the Loebner Prize and Chatterbots
Written by Mike James   
Thursday, 12 September 2013
Article Index
Turing's Test, the Loebner Prize and Chatterbots
The Loebner prize
Restricted Domains

The Turing Test - it's pure Hollywood. Any time a character in a film wants to impress all they have to mention is the Turing Test. There is even a $100,000 prize, the Loebner Prize, for the first AI to pass the famous test - but this too is mostly Hollywood. With the next Loebner Prize due on September 14th, it's a good time to review the whole situation.


The very first thing to say is that we are nowhere near any computer program passing the Turing Test in any meaningful sense. No computer program has even got close to winning the Loebner Gold medal complete with the cash prize but every year since 1991 a chatterbot has fooled the judges to win the Bronze medal. This is sometimes portrayed in the popular press as being "close - perhaps next year" but getting the Loebner Bronze isn't like coming in third in a running race. The Bronze is altogether different to the lofty goals of the Turing test and even the Gold Loebner medal. 

Lets explore some more. 

Can a machine think?

It is difficult to understand the way that people thought about computers back in the 1950s.

For a start the word was used to describe a human occupation - that of performing arithmetic calculations in banks or for insurance companies. Computers, the machines, now seem such a "normal" part of our lives how could they have been viewed differently then?

But this ignores the revolutionary nature of the machine that caused it to be misunderstood even by the people responsible for creating it. Often it was left to mathematicians and other thinkers to make the sort of predictions that we now see as very reasonable.

 

TURING

Alan Turing (1912-1954)

 

Alan Turing was just such a thinker. It is difficult to know what to call Turing, but mathematician seems to be a reasonable term. But it is important not to think of him as a traditional mathematical thinker, indeed at the time there must have been many mathematicians who thought Turing's work more like a cross between philosophy and engineering - which is not a bad description of what we now call computer science.

Turing is known for three distinct achievements:

  • one major theoretical discovery in computer science - roughly but unfairly summed up by the idea of a Turing machine
  • his contributions to national security and computer design - roughly identified as the Enigma project plus his work on the ACE computer.

And

  • his thinking about what we would now refer to as the possibility of machine intelligence.

While these three areas may seem only tenuously connected they all demonstrate a remarkably sophisticated view of the computer at a time at which it was either denigrated as nothing more than a calculator or raised to the unreasonable heights of the "electronic brain".

Turing's view of the new technology was more measured and reasonable than either of these two extremes and he strove to discover the exact limitations of the computer.

In his work on what was and what was not computable he defined a computable procedure, very reasonably, as one that could be a carried out by a human computer following instructions such as add and subtract, and be able to read or write results on a paper tape (not the punched variety).

A Turing machine is just a mechanical version of the human computer. I say a mechanical version but Turing wasn't overly fussed about how the machine was actually constructed. The Turing machine is more a sort of "aid to thought" or a component in a gedanken experiment than anything that you would actually want to build.

The importance of the Turing machine is that even though it is simple, very simple, it can compute anything that is computable. It is more proper to think of the Turing machine as a definition of computability. So far no one has come up with a computer that isn't equivalent to a Turing machine of some type and so if a Turing machine can't compute something then neither can any other computer.

Using the idea of a Turing machine Turing was able to show all sorts of startling results. For example, the majority of numbers are non-computable in the sense that there is no program that computes them, writing a program to deciding if any other program ever halts is also impossible and so on.

These esoteric sounding results set limits on computation which previously was thought to be without limit.

What has this got to do with the artificial intelligence?

Well it has been argued that the human mind/brain isn't equivalent to a Turing machine and so there are things that it, the mind/brain, can compute that a mere machine cannot. I can't personally see any great sense in the argument that the brain is more than a Turing machine but if this is the case then artificial intelligence isn't something that can be achieved by a mechanism like a computer.

Turing thought that artificial intelligence was very possible and that it would be achieved by the end of the 20th century. Sixty years on I personally am of the opinion that the problem is rather more difficult than anything we have ever tackled and that it is still over the event horizon beyond which prediction becomes vague.

The Turing test

Turing didn't have any idea how artificial intelligence would be achieved but it seemed a reasonable product of the progress in computing. Although he did do some work on how machines could be made to play chess, a few years before Claude Shannon (of information theory fame) did the same thing.  Turing spent time considering the more philosophical side of the problem. In particular what constitutes artificial intelligence.

In answer to this question he invented a simple thought experiment in a general discussion paper, Computing Machinery and Intelligence, Mind, 59, 433 (1950). This is the famous Turing Test. 

Take a computer and a person separate rooms and arrange for a communications device - a teletype in Turing's day - to be available as the sole method of passing information into and out of the rooms. Now if it isn't possible to discover which room contains the person and which contains the machine then there is - operationally, practically, for all intents and purposes etc. - no difference between the contents of the two rooms. Of course this means that if we ascribe the quality of intelligence to the human in one room then it is only fair to ascribe the same quality to the machine, the indistinguishable machine, in the other room.

What Turing actually wrote was:

"I believe that in about fifty years' time it will be possible to programme computers, with a storage capacity of about 109, to make them play the imitation game so well that the average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning."

Notice that 109 storage locations is only 954MBytes and while this looks large it isn't silly. Also notice that Turing mentions a 70% chance of identification within five minutes, so giving a rough quantitative value to the test.

Today in most cases the five minutes and 70% part are ignored and a sort of ideal Turing test of 50% chance and as long as the interrogator needs to make a choice are substituted.

Thus Turing provided an operational and practical test for the presence of intelligence that worked even though there was and still isn't a good definition of intelligence. A clever idea but there are many who would still argue that passing the test proves nothing.

It is all related to your view of the "black box" approach to the world. A behaviourist psychology of human intelligence simply relates the inputs to the outputs i.e. the behaviour of the human, and doesn't ask questions about what happens deep inside the processing box. From this point of view Turing had defined a behaviourist or operational definition of human intelligence. If there was nothing you could do to tell the difference between the computer in the box and the human in the box they were essentially the same. 

The most common objection to the Turing test is made by philosophers who insist that if you do look inside the box and find something mechanical, or even trivial in principle, then no matter what the behavior it can't be the equal of a human intelligence. We think, we feel and we know we exist - a simple mechanism can do none of these things. 

The best known of these objections is Searl's Chinese room which is explained in the following video:

 

This argument is either metaphysical or even religious clap-trap that relies on invoking some sort of magical property that biological systems have or its the core of the issue - depending on your persuasion. 

 

 

Banner

<ASIN:0099116413>

<ASIN:0753822008>

<ASIN:1402096240>



Last Updated ( Friday, 13 September 2013 )