Re: Neurons vs. Transistors (IA vs. AI)

Eliezer S. Yudkowsky (
Thu, 29 Jul 1999 05:08:42 -0500

Paul Hughes wrote:
> "Eliezer S. Yudkowsky" wrote:
> > Paul Hughes wrote:
> > >
> > > **Since no one has actually built or designed a theoretical human-level AI, how can
> > > anyone possibly claim what it takes to build one? This seems completely absurd
> > > to the point of self-contradiction! As so many are fond of saying around here -
> > > extraordinary claims require extraordinary proof.
> >
> > Hello-oo-o?
> Read it, re-read it, has potential, been there, done that. A design is
> just a design. The real question is - will it work? You don't know that,
> you can't prove it, because until it actually *is* working, you have
> no proof. At this point it's nothing more than a self-consistent piece
> of computational and logical poetry. Interesting, original, fun to read,
> but utterly useless until it can actually be applied to *do* something.

We weren't talking about usefulness. You said that since nobody had designed an AI, nobody knows what it'll take to build one. I designed an AI. I have some kind of vague idea of what it'll take to build it. That's all I'm trying to say. My estimate may be wrong, but there's nothing absurd "to the point of self-contradiction" about my making one.

Or are you seriously suggesting that I should try to write and run the entire AI before even *trying* to estimate the hardware required? The quantity and architecture of hardware *available* is a major factor in software design. Yes, we'll have a much better idea of what it takes to run an AI after the first version fails - but I still have to guess, now, with what I know now.

BTW, my estimate is 1e17 ops/sec on CPU-like architectures.

           Eliezer S. Yudkowsky
Running on BeOS           Typing in Dvorak          Programming with Patterns
Voting for Libertarians   Heading for Singularity   There Is A Better Way