Shane Legg wrote:
John,
It still seems decidedly odd to me to call AIXI intelligent. For
similar reasons, I wouldn't call a program that generates all
possible strings of characters, sometimes randomly producing a
literary masterpiece, artistic or creative.
I don't understand your argument. Generating all possible strings is
"easy" in theory, the hard part is to decide which of these strings
is a masterpiece and which is not. If you could give me a formal
equation for deciding whether something was a masterpiece of
literature, I would be very impressed.
Shane
Ridiculously easy.
Thus:
Define an algorithm W that generates an infinite set N of literary works
by random concatenation of words (or characters).
Define a second algorithm G that generates an infinite set of complete
universe simulations, U, using the standard laws of our universe, with
each simulation being governed by an initial random seed.
For each literary work n in N, use G to generate a universe u, and
within that universe, inject a copy of the literary work at a random
point in the spacetime of u. Measure the reaction, in terms of critical
acclaim generated by the work in any species who happen to be hanging
around the injection point (critical acclaim = number of times the work
is described in the language of that species as a "literary
masterpiece", summed over the entire history of that universe). Compute
the critical acclaim density (using a measure theory of your choice) as
the injection point is varied randomly in an infinite set of universes.
Compare the critical acclaim density across all n in N, taking only
those that exceed the critical acclaim of a reference work (say, Moby
Dick). Those in the accepted set are all literary masterpieces.
Nothing could be simpler. I hereby define this algorithm to be called
"FIXI" and I propose it as a general solution to the problem of
generating literary masterpieces.
If someone would now give me a very generous research grant, I would be
happy to do some work on possible simpler versions of this algorithm
that might be feasible on real computing machines. No promises, but I'd
be happy to work on it.
Richard Loosemore.
-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?list_id=11983