Any progress towards AI at all? was Re: [extropy-chat] Futures Past

[Russell Wallace]

On 10/10/05, Eugen Leitl <eugen at leitl.org> wrote:
>
> On Mon, Oct 10, 2005 at 04:18:11PM +0100, Russell Wallace wrote:
>
> > Oh, you could try that approach, but I don't think it's going to work.
>
> My approach worked at least once (we're soaking in it), yours hasn't so
> far.


Your approach required a nanocomputer the size of a planet running for four
billion years, and even with those resources it failed 57 grillion times for
one success if the Great Silence is anything to go by! (Have you worked with
evolutionary computation at all?)

> (Which is just as well, because you can't put Friendliness into brute
> force
> > evolution.)
>
> It also happened at least once (though it's not very friendly,
> admittedly).
> I don't think Friendliness exists. It is not a well-defined property
> to start with, so you can't gauge conservation over system evolution.
>
> Is it worthwhile to try for a friendly system? Sure, if you can make it
> so.
> But, I'm not holding my breath. It's too strangling a constraint, and
> a seed is already hard as is.


I'm trying to prove you wrong, though admittedly you wouldn't be wise to
hold your breath!

> Well, my prediction is that we _won't_ see a human equivalent by 2015,
> > however hard I try to prove myself wrong :)
>
> What I meant, there won't be hardware equivalents of a human CNS,
> regardless by which metric.


Agreed.

> That depends, but if you think you can talk a manufacturer into building
> > hardware like that, please go for it!
>
> Unfortunately, this is beyond ASIC budget. Way beyond ASIC budget.
> This could easily ruin a major manufacturer, so nobody is going
> to try until prototyping costs fall due to desktop nanoelectronics
> fabs (which won't be there by 2015, not even by 2025).


Agreed there also. (Indeed, short of Friendly AI we're never going to see
desktop molecular factories for legal reasons alone.)

- Russell
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20051010/d5d0c651/attachment.html>