An interesting comment on the problems of artificial intelligence. A
problem long foreseen but never actually exemplified until now is that
an AI that gains it's intelligence by experience (as AlphaGo does) is
going to be unpredictable when confronted with problems outside the
range of it's experience. Of course in a sense people are like that
too; an intelligent bushman will have no predictable response to alarms
in a nuclear powerplant.
Brent
-------- Forwarded Message --------
Actually, the go endgame isn't just "tidying up the boundaries":
I can just invade your territory. If you play right my invasion
will eventually die, it was unjustified, so humans do not do it.
(Except: if it is a pro versus me, he'd likely do an unjustified
invasion anyhow because
he'd guess I'd often be too dumb to respond correctly.)
But my point is, all those invasions and refutations always are there
like a sleeping volcano beneath the surface, strong humans just never
play them and their refutations, since it'd be a waste of time; they
just agree to end the game. It doesn't hurt your score in chinese
rules if you play an unjustified invasion which gets killed, so in
principle you should try it.
What that also means: the part of alphago's training that was based on
strong human games, contained zero examples of that stuff!
So maybe if Lee Sedol tried it, alphago would mess up.
I guarantee you, it has a higher probability of working than resigning.
Also, presumably alphago usually works well, but even its designers really do
not understand why and cannot guarantee it will not do something crazy
next move.
It really is an AI in the sense that nobody understands what it is thinking
and there is no way it could explain it to us. Unlike, say, today's strongest
chess programs, which I feel I pretty much totally understand -- they
have human designed
evaluation functions with some parameters tuned by fairly crude
automated methods,
but the whole evaluation function is something a human understands.
That again, means if Lee Sedol took more of a software tester
attitude, it could pay.
Anyhow, LS has been rocked! There was a little press interview with LS after,
and he looked a bit like he might break into tears. He said he just could not
believe the machine could beat him, and even when LS felt he was
behind, he still figured it was just a matter of time before machine
would find a way to blow it. :)
--
You received this message because you are subscribed to the Google Groups
"Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To post to this group, send email to [email protected].
Visit this group at https://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/d/optout.
