Re: Creative Machines

Eliezer Yudkowsky (
Sat, 25 Jan 1997 18:59:11 -0600

> From: Damien Broderick <>
> > A simple example (too smug, too hasty, I got it wrong): if ABC changes to
> > ABD, what will MRRJJJ become? What does XYZ turn into? A smart program
> > named Copycat has provided a range of solutions. Answers: since ABC maps to
> > 123, ABD equals 124; hence JJJ becomes JJJJ, yielding MRRJJJJ (not, say,
> > MRRJJD or MRRKKK, let alone MRRKKKK). And XYZ: XYZZ? XYY? XYA? Maybe.
> > But clever Copycat saw that A maps onto Z. Running backwards, we get WYZ.
> > Elegant. Is that artificial mind creative, or what?
[Hal Finney:]
> Douglas Hofstadter's group is doing this work, and his book from last year
> goes into it in detail.
> I am skeptical about how well his results will generalize, although it is
> an interesting class of problems. I can guarantee you that Hofstadter,
> a very smart man, saw the answers to the problems above immediately.
> In fact he and his students created the problems knowing they had these
> clever answers. So then the task they faced was how to write a program
> which would get the same answers that they got. That they were able to
> do so is very impressive. But the question remains whether the data
> structures they came up with are of general enough applicability that
> they can shed light on other problems of AI.

Speakin' as someone who looked at the source code, the data structures
*look* general, but FARG needs a 'Cat that can *learn* symbols and
group-types instead of havin' a preprogrammed "Slipnet" and
"successor-groups" before the method used is really general.

--       Eliezer S. Yudkowsky

Disclaimer:  Unless otherwise specified, I'm not telling you
everything I think I know.