Wednesday, 20 May 2009

Autoself

Looking in a latin dictionary I see that "auto" means self. Ah ha I think now I see more clearly the fascination I have had with automatic mechanisms and computers ... it is the dream of the ghost in the machine!

An automatic mechanism , a robot, a computer program that runs itself is a very exciting idea: it alludes to the existence of "self" as a discrete entity. It is the soul made almost tangible in metal and silicon. This was the goal of a lot of my time as a child: making intelligent machines - or working out how to do it at least. I failed, but not without realising that it was impossible. Understanding why it is impossible is more subtle.

The problem lies in this notion of "self". It is impossible to make a self. Even a mother's biology doesn't make a self. Buddha explains very clearly and unequivocally that such an entity does not exist. So what does exist?

The problem is that we like to imagine "self" as discrete and self-contained - separate from the surrounding world and environment. An intelligent machine is not just a machine that responds to its environment - that is simply the processing of inputs into outputs. We imagine that one day we will build a personality whereby the machine will have gained something that is separate from the stream of inputs and separate - most importantly - from "us" the makers.

It is not the machine that we are building but really an extension of ourselves - a negation of ourselves, another person... and that is where the project fails. We can't build an entity that is distinct from us in "personality" and "self" because even we don't possess such a thing! The machine will always be an entity intricately involved with its surroundings like we ourselves are.

If machines ever get personalities it will be because they take part in the "human world" so completely that they can take part in the myths and stories that populate life and in so doing will gain the myth of self, personality and soul.

Using Python, Microsoft Speech and the AIML module I linked the speechbot Alice to the input from the recognition engine and had a crude (an inaccurate) buddy to talk to. The illusion was incredibly powerful having a voice return meaningul responses to my own speech. It is a simple game (altho a bit more complex than football) but one that almost allows me to believe that there is a person there. It is dreams that we speak of when we speak of me and you.

No comments:

US displaying its Imperialist credentials... yet again

Wanted to know the pattern of UN votes over Venezuela and then got into seeing if ChatGPT could see the obvious pattern of Imperialism here....