Re: Donaldson, Tegmark and AGI

From: Russell Wallace (
Date: Fri Aug 11 2006 - 20:42:38 MDT

On 8/12/06, Robin Hanson <> wrote:
> What is the evidence for these two key claims, that Cyc is less
> capable than Eurisko, or that Google is more capable in Cyc's
> strongest areas?

Eurisko: I don't know what's available online, back when I studied it, it
was when paper was where it was at. But it did things like winning the
"Traveller Trillion Credit Squadron" war game tournament against human
experts and producing some interesting results in 3D circuit design. It
wouldn't have become Skynet anytime soon, but at least it _did_ things.

Cyc: well it doesn't do anything. Doesn't necessarily mean it's worthless,
it might be useful as a corpus to some project at some stage, but by itself
it doesn't do anything much even in the area of giving declarative answers
to declarative questions.

Google: well it rocks hard. Real life intelligence amplification. Seriously,
I've been able to reclaim neurons that used to be devoted to storing facts.
This wins. Imagine how we'd feel if an AGI project did something like that.

The task of creating AI does seem very hard, but the task of
> uploading human intelligence seems doable within a foreseeable time
> frame, even if that is at least a few decades away. Why isn't four
> to eight decades soon enough?

Maybe it is. I think that'll be harder than it looks, but like I said, I
haven't studied that end of things as intensively; I don't feel ready to
defend a claim that it won't work out in whatever time window is available.
Let he who thinks it will, by all means prove me wrong!

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:57 MDT