[ExI] Superhuman Poker

[Dave Sill]

On Wed, Jul 17, 2019 at 8:17 PM William Flynn Wallace <foozler83 at gmail.com>
wrote:

>  Can we learn what it learned?
>>
>
> Not really, the program can't tell us why it's so good at what it does for
> the same reason a human genius can't tell us why he's so good at what he
> does, neither of them knows.
>
>  John K Clark
>

Unlike the human genius, the poker program can't talk or understand speech
or converse in any language. It's like an idiot savant. If if could talk,
it'd just say "I just pick the statistically best play".

It's right there in the code, isn't it?  But can't someone tell what
> changes have been made from the original code, which was written by humans?
>

Machine learning isn't self-modifying code. The code never changes. The
learning process builds huge tables of statistics recording the outcomes of
different plays.


>   I am way out of my depth here, but if AIs are going to learn things for
> us, somehow we should be able to tell how it does what it does.  Maybe not
> now but in the future?  Otherwise the AIs have all the secrets!
>

These AIs are learning very narrowly-defined games in very simple domains
and a tiny set of well-defined rules. What they do is impressive to us in
the same way that a calculator is impressive to us at doing arithmetic.
They don't have a strategy beyond making the play that the numbers say is
the best one.

Their benefit to us will not be as teachers but as tools to get jobs done
faster/better/cheaper than we can. We don't need an AI poker program
(unless you want to try to cheat with it), it's primarily a demonstration
of what can be done with machine learning. There are many practical
applications like self-driving cars, computer vision, Youtube video
recommendations, medical diagnosis, ...

-Dave
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20190718/bce5659f/attachment.htm>