The only thing stunning about this breakthrough is that anyoneTom Mazanec wrote: > AI Breakthrough: One step closer to Singularity
> https://gizmodo.com/stunning-ai-breakth ... 1819650084
thinks its stunning.
The first computer checker champion was an IBM computer in 1960.
The algorithm was to let the computer play itself until it had
created a database of the winning moves in any position.
Chess was too complicated to yield to that approach, but for
decades chess playing algorithms were augmented by "best move"
data bases of the best moves in endgame positions. So the database
has been integrated into the logic of more sophisticated play
algorithms.
In other words, this database algorithm is the simplest algorithm for
computerized game play.
Google has apparently done the same thing -- found a way to create a
limited database of "best move" positions, and integrate the database
into their existing alphago algorithm. It's not at all surprising
that the database provides a significant advantage. And it's been
done before, so it's certainly not a stunning breakthrough.
Furthermore, it's pretty much useless in real world situations.
If a computerized general has to make the next decision in
pursuing a war, then there will be no meaningful database
lookup.
I'm still most impressed with IBM's Watson's performance in Jeopardy.
Watson had to learn by reading millions of pages on the internet, and
use very sophisticated natural language processing algorithms to learn
to answer Jeopardy questions. That's much closer to how AI will have
to act in the real world than playing a game where rigid rules are
defined in advance.