[Computer-go] Notes from the Asilomar Conference on Beneficial AI

Adrian.B.Robert at gmail.com Adrian.B.Robert at gmail.com
Fri Feb 10 03:56:39 PST 2017


Richard J Lorentz <lorentz at csun.edu> writes:

> Thanks for the interesting link. Indeed, some good reading there.
>
> One quote that I've seen various versions of a number of times now: "
> More interesting for the rest of us, AlphaGo is playing moves and
> styles that all human masters had dismissed as stupid centuries ago."

Related to this, the idea was mentioned that maybe AlphaGo is
beating humans partially because the strategies employed have
gotten trapped in a local minimum due to historical factors.

It might not necessarily be historical factors, but simply that
the techniques that AlphaGo has been finding are just too hard
for humans to learn or use effectively.  For example, in
mathematics, different techniques have been developed for both
symbolic and numerical manipulation for use in computer programs,
but humans don't use these because they are tedious or difficult
to error check, place a heavy load on memory, or don't provide
intuitive insight.  Likewise, it could be that certain strategies
work well in Go but require keeping track of details and playing
out more precisely than human abilities typically allow.

It is nice to hope that we could learn something about Go from
AlphaGo, but we may learn little more than what mathematicians
learn when a computer-assisted proof consisting of several
hundred pages is generated for a conjecture like Fermat's last
theorem.



More information about the Computer-go mailing list