[sl4] Gray goo risks

From: Matt Mahoney (matmahoney@yahoo.com)
Date: Mon Aug 02 2010 - 09:42:15 MDT

In http://www.foresight.org/nano/Ecophagy.html Freitas quantifies the risks of
gray goo. In particular, self replicating robots cannot be much smaller or
faster than living cells due to the limits of physics and chemistry and
available energy. This seems to suggest that a gray goo accident could not
destroy DNA based life (ecophagy) faster than a plague could spread, and could
not evolve into a dangerous form faster than ordinary evolution (e.g. antibiotic

However, in these scenarios, nanorobots compute by moving atoms rather than by
transmitting electrical signals or light, either of which would be about a
million times faster. Freitas does not consider scenarios where nanorobots, each
with a small amount of computing power, organize into supercomputers and
coordinate their actions. Questions:

1. Could nanorobots speed up either reproduction or evolution using electronic
computation, like the way that we speed up evolution using breeding and genetic

2. If so, how much computing power is required?

I have doubts that the risks are significantly increased. First, I don't think
that any species could improve itself faster than evolution by blind search. If
an agent knew which of its children were smarter, then it would already be that
smart. Humans can breed dogs, but dogs can't breed humans.

Second, biological computational efficiency is already near the kT limit of
~10^-20 J per bit operation and already consumes a large fraction of the
available solar energy reaching the Earth. Equivalent molecular scale electronic
computation necessary to simulate evolution a million times faster would run a
million times hotter.

But I am not confident enough in my answers to dismiss the enormous risks.
Surely there are other scenarios I am overlooking?

 -- Matt Mahoney, matmahoney@yahoo.com

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:05 MDT