Re: [agi] A difficulty with AI reflectivity

From: Christian Szegedy (
Date: Fri Oct 22 2004 - 07:50:51 MDT

Ben Goertzel wrote:

>About impracticality -- Schmidhuber (whose Godel machine Eli is discussing)
>is trying to make a theory of AGI by starting from
>infinite-processing-power AGI and "working down" to finite-processing-power
>AGI. I guess this approach can give significant insights, but it can't tell
>you everything you need to know to make a finite-processing-power AGI...
>-- Ben
I liked Schmidthubers paper. It contains several important ideas that could
be important for construction of a practical AI.

However, there are some important issues to be criticized>

Schmidthubers paper is a greatly modified version of Levins (and Hutters)
universal search algorithm. If one understands the original idea, it becomes
immediately clear that although the universal search is a very simple and
provably almost optimal algorithm (at least asymptotically), its
practical value
is simply zero: the unimaginably huge additive constant makes it practically

Schmidthuber tries to cure this shortcoming by using arbitrary proof- and
algorithm-search methods and combine it by provably optimial rewrite
In fact, he retains the same optimality guarantees as Levins/Hutters
but with a hope that the algorithm *may* work practically, with
*suitable plugins*.

However there is an astronomically huge gap between the theoretical
optimality and the practical feasibility. If algorithm/proof-search
a similar method would work, then they could be modified to work in his
"provably optimal" framework. The plugins would work *without*
his framework equally well. This means that the *practical tractability*
would depend on factors that are completely untreated in his exposition.

For me, it appears to be some kind of cheating.
Assume that some people develop by hard work some really good
program-/proof-generation techniques, then we simply puts their
plugins into his framework and gets provable optimality and practical
performance at the same time and can reap the credits for his
"important contribution"

The problem is that the optimality he promises may be completely
impractical and never reached in real-life, while the practical
tractability would not be his achievement at all.

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:49 MDT