RE: Seed AI milestones (was: Microsoft aflare)

From: Ben Goertzel (
Date: Wed Feb 27 2002 - 10:49:55 MST

> > General intelligence is not property of a simple system. Far from it.
> > As a result I predict that human software engineers coding an AI
> > explicitly (i.e. using not stochastic/noisy/evolutionary methods) are
> > going to fall short of the goal.
> As far as I can tell, stochastic/noisy/evolutionary methods buy you
> relatively little.

I have worked with evolutionary computing methods a great deal.

It is very clearly not feasible to evolve an intelligent system using GP
type methods, or any moderately close variant thereof, any time in the near
future. The computational intractability is obvious from the performance of
these techniques on simpler problems.

The argument against designing an intelligent system in the near future is
much less clear, in my view.

I think that evolutionary computing has a role in AI design, which is as

1) It is one among many important cognitive techniques, to be used for
generating new ideas and procedures inside a digital mind. Some AI's may
explicitly use EC methods (e.g. Novamente does), others (such as Pei Wang's
NARS system) don't use them explicitly, but may be viewed as using them

2) It is a valuable tool for *automated parameter tuning* of a complex AI
system, if one has quantitative performance metrics to which to tune the

-- Ben G

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT