Artificial Intelligence Solution

Geoffrey Hamilton

May 9, 2006

Skeptic Magazine's article on artificial intelligence purports to prove A.I. is impossible. The subtitle says it all: the futile quest for artificial intelligence. Author Peter Kassan makes many technical points showing the statistical improbability of a machine ever reaching the complexity of a human brain - as if that is the only legitimate benchmark. However, Kassan's first point, and also his over arching critique, that the how is not possible until the what is discovered, also makes his entire subsequent technical probability arguments irrelevant. Also it makes his apocalyptic words ring like so many expert's famous last words - as in, the 'bumble bees cannot fly' logic.

Right now the what is answered by gamegene theory. In 2001, I began, in the following unfinished piece, to articulate the A.I. solution inherent in my 1996 deduction of the game gene. I include it here by way of showing any development I may have made and to show Kassan's point is at least five years out of date.

Artificial Intelligence and the Missing Link: Game Gene Theory

April 5, 2001

There is now a solution to the unrealized ambitions of scientists who wish to create artificial intelligence (A.I.). This solution helps determine what constitutes actual life-like intelligence and to use that determination to create life-like intelligence from scratch. This solution is derived from the Game Gene Theory.

The problem with A.I., as understood by many scientists in the field, is to decide how to build a machine with A.I.. However, the precursor should be how to decide what intelligence is, and further back, why would an A.I. want to exist? because if A.I. is going to be more than a computer it must have its own thoughts, wishes and hopes. It must be primarily independent in action. Scientists are getting lost along the way.

From what I understand of their problem scientists are stuck on how to allow for learning and separately how to create consciousness. Both are necessary at the same time for their game to succeed - the first because, according to these scientists, it mimics the development we see in life forms and the second ( despite it being imagined by them to be of unique significance to human beings) all life has consciousness and so must A.I..

The answer to their problem is too simple for these people: it is this - recognize that all life is playing games, even when learning, and that all life must have a consciousness which is defined only as a sense of self.

To make a sense of self a technology is perhaps something like this. A massive CPU will have a sensor attached to itself which informs itself of the space it takes in the world, so to speak. This space need only be like a minor premise that says that its being is more important than any other fact. Other senses would be established to influence the CPU, including sensors which actually allows harm and gain to be inflicted on the CPU in the form of less or more ram, for example.

All these sensors would formally mimic the physical and emotional influences found in actual life forms. The reward/punishment system must be outside of the direct control of the CPU and tied indirectly to the reactions of the cpu to the games it plays. If it perceives a success the extra ram is given. An internally perceived reversal implies the adverse award. Arbitrary harm and praise too needs to be awarded at times to keep it all unpredictable for the A.I. so that only games are what works.

The main operating system program must play games based on only a few starter goals. Then those goals are developed and influenced by the memory of 'emotional' and 'physical' reactions to previous games.

No regular short cuts to rewards can be given to the cpu. For example, with human life forms, a person can take a narcotic which rewards the individual irrespective of the games' progress.

There must be a method for making arbitrary choices. That method could involve a time limit or a general rule to cut off self debate......
Unfinished

Today (May 9, 2006) I have little to add to these basic concepts because I generally don't care if A.I. succeeds. That is, except to say, no one, not even an A.I. robot, needs to exist, but if you're going to force one to live you may as well get it right.

Peter Kassan and Skeptic Magazine may have ideas similar to what I've laid our here but it's not due a wish to spare some future A.I. a little suffering, or even to get it right, they want math and conventional science to win the day so they can continue their real work of assaulting all those who don't conform to conventional beliefs.

GRH

All essays

Home