Re: Darwinian dynamics unlikely to apply to superintelligence

From: Eliezer S. Yudkowsky (
Date: Sun Jan 04 2004 - 22:01:37 MST

Perry E. Metzger wrote:
> Oh, but it doesn't. Consider what would happen to the human race if
> everyone had only one idea of beauty rather than a continuum. Mating
> would become a difficult problem indeed. We'd all want to live in
> exactly the same place, making real estate there impossibly expensive
> and plenty of good land unused. One can go on from there.
> Put another way, many would never consider removing garbage for a
> living, so the man who does not mind makes a good living doing it for
> those others who will not.
> Being different from a large population in and of itself has survival
> value. It provides you with a niche.
> One of the things that then happens, of course, is that you end up
> with this giant system of non-linear equations, so things constantly
> alter rather than stabilizing. It is rather neat, actually.

The cost of commonness is when, for a type X, dX/dt is sub-linear in X,
meaning that growth is sub-exponential, and that new types can invade from
rarity even if they cannot become predominant. If dX/dt is more than
linear in X, growth is superexponential and you have a first mover
advantage - new types cannot invade from rarity even if they would have a
better carrying capacity when predominant. I've heard these situations
respectively called "survival of anybody" and "survival of the first", in
contrast to "survival of the fittest" when dX/dt is linear in X.

Please note that these cases both depend on particular evolutionary
dynamics. Where would a cost of commonness arise in SIs competing to, I
don't know, take over each other's stars or something? If indeed there
was any war at all, which seems to me doubtful (see next post).

Eliezer S. Yudkowsky                
Research Fellow, Singularity Institute for Artificial Intelligence

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:43 MDT