Re: poly: Are von Neumann machines inevitable, or even likely?

From: Perry E. Metzger <>
Date: Sun Dec 21 1997 - 20:28:26 PST

Tim May writes:
> The "someone will try it" argument is fairly persuasive...up to a point. I
> don't believe von Neumann machines are nearly as easy to construct (or
> model) as many do. I suspect that by the time they are buildable, other
> forces and considerations will have emerged.

The nature of technology is such that it is hard to predict such
things with any certainty. Only time will tell the answer. However, I
think one can make a few reasonable observations.

We already have very good ideas about how to *model* such systems. Von
Neumann came up with very good theory on such things, and we've
learned a lot about how natural self reproducing systems (which are,
after all, also von Neumann machines) work. We are currently missing
a couple of things from the mix: artificial intelligence and universal
assemblers. However, I have no cause to believe that either will be
expensive in resources once they come around, or at least not long
after they come around.

> (Perhaps morosely, I think it *likelier* that a man-made biological virus
> will destroy all human life long before a realistic von Neumann probe is
> ever built. The knowledge to build such a virus (or whatever) is almost
> here now, where nanotech and replicators are many, many decades off.)

I don't think, actually, anone could create such a virus with current
technology. We know how to kill perhaps 80% or 90% of the population,
but killing 100% is substantially harder, and possibly impossible
given what we know now. Even the deadliest diseases we know of are not
"slate wipers".

> Thus, I believe that in, say, 300 years, when the technology/knowledge to
> build von Neumann probes exists, the AIs (or _the_ AI) will be deciding. I
> could be wrong, but I'm skeptical that rogue launchers will be common.

I tend to agree -- what I disagree with is the notion that there will
be few AIs. There are billions of humans -- why should we assume fewer
AIs, or that they would be different from us in terms of willfullness
or whimfullness? If humans are like that, why not AIs? I'm not sure it
would be easy to build AIs that don't share many of the
characteristics of evolved systems in this regard.

> (Though a rebuttal of this is that we are looking at cosmological times,
> and something is sure to "eventually leak out," even if such AIs controlled
> things. My rebuttal to this would be that even such leaking out, or rogue,
> launches could be countered by more advanced probes to destroy them.)

If it doesn't work on earth with living systems, I find it hard to
believe it would work in space. Evolution is evolution. You'll send
out killer probles and the earlier probes will be out there, evolving
to try to stop from being killed. Even if you get 99% of them, what of
the 1% that remains?

> >Dunno. Bussard ramjet style probes would probably be fairly trivial to
> >construct -- or especially to self-contstruct.
> A la the "Tau Zero" ships...but they could still only carry a tiny subset
> of the capabilities of the AI.

They just need human level intelligence or higher in order to function
effectively. More is a frill.

> I certainly do agree, by the way, that the absence of any evidence of such
> probes gives us some hints as to scarce intelligence may be.

In that case, I suspect we are mostly just disagreeing about details.

Received on Mon Dec 22 04:21:30 1997

This archive was generated by hypermail 2.1.8 : Tue Mar 07 2006 - 14:45:29 PST