[ExI] From Arms Race to Joint Venture

Zero Powers zero.powers at gmail.com
Thu Oct 18 18:32:01 UTC 2018


On Thu, Oct 18, 2018 at 8:22 AM William Flynn Wallace <foozler83 at gmail.com>
wrote:

> You should assume that I know little about AI (as if that weren't already
> obvious).  But it seems that it excels at things that involve tremendous
> counting:  chess: as many moves ahead as possible, all countermoves
> considered.  No human can do this past several moves.  Piece of cake for AI.
>

That used to be the case, but not anymore. The system that recently beat
the best human Go player did a qualitatively different kind of thinking
than the system used to beat Garry Kasparov at chess 20 years ago. Deep
Blue might have been called a super counter. But AlphaGo is much more than
that. Deep learning is the difference. It excels in things which used to be
our forte; things like pattern recognition and differentiation. It can
compete against itself millions of times per day, learn from mistakes and
recursively fine tune it's tactics and strategies.

>
> Now here is what I would like to see an AI do:  feed it several studies in
> some area like physics, or even psychology, and ask it to come up with
> reasonable hypotheses (as judged by humans, of course) about what
> experiments to do next.  No counting at all - just abstract reasoning.
> When it can do this it can take over science, right?
>

Would you settle for analyzing diagnostic imaging and making accurate
diagnoses?

https://www.telegraph.co.uk/technology/2018/10/15/googles-ai-can-spot-breast-cancer-better-humans/

*0**0*
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20181018/2c6004a4/attachment.html>


More information about the extropy-chat mailing list