What is Machine Intelligence

Machine intelligence (also know as Artificial intelligence) is the attempt to get the machine to do things that if a human did them would be classed as intelligent, which simply put is to try and make our computers and machines just a little less stupid.

Russian Artificial Intelligence (AI) researchers say that someone or something is intelligent if it gives intelligent answers to intelligent questions. While there are other definitions, this one does get around the problem that western AI have of arguing about what is intelligent and what isn't.

It also means that you can avoid the western split between social AI which is about how good the man-machine interface is, and technical AI which is more about things like does your house design program tell you if you are putting a door where you can't open it.

Machines and Intelligence

Machines have a significant impact on intelligence in a few ways. First you have machine intelligence, where you endeavor to encode (often) high-level human reasoning into a computer. The best examples of this at the moment are expert systems. The more technophobic people out there worry about if this will lead to ultra intelligent machines that will take over.

Then you have the really worrying case, where as the computers start thinking, the people stop. You can pretty much get around this by only letting people use computer technology to do things that they already know how to do.

The other time you come across machines and intelligence is in the case of Intelligence Amplification (IA) where you and the computer work together to do things that neither of you could do alone. The simple case of this is when you have your database linked to your spreadsheet, which then uses graphs and other statistical methods to help you visualize how the data is changing over time. The complex cases include haptic displays which use force-feedback to help chemists better design medicines, and micro-worlds in computer languages like prolog and lisp which let children learn quantum mechanics in a simple way.

The Turing Test

In the 1930's, the genius Alan Turing suggested a procedure that has since become known as the Turing test. This basically refers to playing the imitation game. So how does it work?

You put a computer and a person on the end of communications links that hide the identity of the person on the other end of the link. Now an impartial observer must pose questions to the computer and the person, not knowing which is which. The questioner must try and decide which is the computer and which is the person. If the questioner is unable to distinguish between the quality of the computer's replies and those of the person, then the computer is considered to have intelligence.

The problem with this is that it begs the question that it sets out to answer by assuming that intelligence is there, and that is how you detect it. For this reason John Searle came up with his chinese room problem, which has its own problems.

Another problem is that it does not tell you how intelligent the thing on the other end of the link is. This has lead to a call for using a Turing measure.

The Turing measure

This works the same way as the Turing test, but instead of saying are you convinced it's intelligent, you find the intelligence of the person doing the test, and the thing on the other end is viewed as being as intelligent as the average person that you manage to fool. Of course the people taking the test have to have their IQ measured using Intelligence tests.

So where does it leave us?

On the whole, all non human intelligence seems very strange, but it is possible to learn to live with it.

Part of the Intelligence Project