Re: Is an SI possible with todays hardware?
Sun, 21 Feb 1999 17:56:37 EST

In a message dated 2/21/99 2:50:15 PM, wrote:

>Jeff Davis wrote:
> >Would it be possible to construct a self-enhancing AI with
> >today's hardware?
>Let's see how complicated it could possibly be, the human genome is 6
>billion base pairs long, there are 4 bases so each base can represent 2
>bits and there are 8 bits per byte. That comes out to 1.5 gigabytes,
>however 80% of that is junk DNA so the working genome of the entire
>human body is 300 megabytes. A chimpanzee's genome is only about 1%
>different from a human one, so I think I'm being very generous if I say
>that 10% of our working DNA involves intelligence. Thus a seed
>intelligence able to grow in power as it interacts with the environment
>as a newborn baby does can be specified with just 30 megabytes and
>probably much less.

I think you're misinterpreting the analogy. The DNA is not the actual hardware which performs the computations for intelligence. Rather, it is a blueprint for a hardware which can perform the necessary computations and which has a certain but limited pre-programming.

The accurate analogy would be that we could produce a blueprint for an intelligent machine with current technology. Actually making the object the blueprint describes, however, is a different problem. If we can't test the blueprint, of course, we will be hard-pressed to decide whether the blueprint we make is valid.

Incidentally, about 50% of our DNA is involved with the brain, so 10% may be low. Naturally shifting from 30 megs to 150 megs makes no qualitative difference to your argument.