AI

Rob Harris (rob@hbinternet.co.uk)
Wed, 24 Nov 1999 17:10:26 -0000

>Well, I think the Asimovian (?) laws should be applied to any AI,
intrinsically,
>within some kind of non-alternative hard-wired framework. That is,
similarly to
>Java, any motive AI should operate within a "sandbox", subject to the
protection
>of humanity.

I've posted this same point a million times, but I'm going to do it again anyway - cos it's not getting through. When you build a system to perform a certain task, you have to tell it what to do - not what NOT to do. There is nothing that does everything and has to be constrained down to a set task. It doesn't work that way, but I can see exactly why people see the situation in this way. Of course, us humans work in exactly that way, what with us having "free will" and all. Just strange that all us free people always choose exactly the same will (must survive, must gather resources, must have sex, must climb the social hierarchy etc....). The point is that creating an artificial "lifeform", or an evolving program that has "motivations" to survive and so on, will serve no purpose to us whatsoever. "oh, great" the creator will say, I've got a evolving program here on my machine with emergent motivations that mean it will attempt to "survive". And it does. Until I pull the plug.
Some talk of "seed AI" becoming a self aware nemesis of humanity. Crap. You see, the idea of "seed AI" is analogous with the evolution of life itself. In order for that first chain of genetic material to replicate, it had to "find" (literally bump into) the complementary nucleotides to make a new chain. In time and fluke, a structure that optimised this process arose, but at a price. The price is that the next generation must find not only the nucleotides that will make up the new chain, but the components of the optimisation structure as well. And so the quest for resources is born. This is the mechanism by which there has been evolution of life, because the optimised bits of genetic material replicated faster, and so outnumbered the old structures and so on until you have hardcore complexities such as humans. So, how exactly are you going to bring about this situation in "cyberspace"? How can resources ever be made to mean anything in this context? The fact is that you have to hand make motivations by directly specifying the situations in which the old generation can successfully reproduce (a fitness function). So, basically, you have to create a program using perhaps genetic algorithms, and have a specific task in mind, although there is no reason why this solution would be superior to a straight program for any specified task.
What is comes down to is this: if there were to ever be a malevolent program threatening the human race, it is no more "intelligent" or "sentient" than Microsoft Word, and it was created by a complete idiot who could have far easier made a big bomb to blow us all up, than an AI super-hacker programmed to get it's finger on that red button.

Rob.