Humanist Discussion Group, Vol. 35, No. 385.
        Department of Digital Humanities, University of Cologne
                      Hosted by DH-Cologne
                       www.dhhumanist.org
                Submit to: humanist@dhhumanist.org




        Date: 2021-12-03 09:27:39+00:00
        From: Tim Smithers <tim.smithers@cantab.net>
        Subject: Re: [Humanist] 35.373: convergence and divergence?

Dear Willard,

I see some unsupported assumptions in your reasoning.

You say

    "...  notably with Deep Mind's AlphaGo Zero, significant
     divergences have emerged between human and artificial
     intelligences, e.g. moves in Go that no one in the last
     few millennia of the game have thought of before."

How do you know no human player has thought of and made these
"unknown moves" before in Chess or Go?  Just because the
people studying the play of AlphaZero (in Chess) and AlphaGo
(in Go) -- the Chess and Go experts -- have not seen these
moves does not mean they have not been thought of and made by
someone in the many millions of games of Chess and Go that
have been played, but not recorded.  We don't have a good
record of how humans play Chess and Go.  Assuming, as I think
you, and others, do, that the only interesting or good moves
in Chess and Go are those made by the experts whose games we
do record, is, I would say unfair on human Chess and Go
players.  And, Chess and Go masters down the ages have still
only played a tiny fraction of all the games of Chess and Go
we would score as the best games.  Not having been played by a
human yet does not mean humans cannot, or will not, think of
these moves and make them.

More.  Learning, studying, practicing in humans tends to build
and strengthen particular thinking mechanisms.  And, since,
unlike for AlphaZero and AlphaGo, human Chess and Go players
learn from other humans, and not from playing against
themselves, how those we learn from think influences how we
learn to think.  It therefore seems hardly surprising that
AlphaZero and AlphaGo have "discovered" moves in Chess and Go
not seen before in expert human play (so far).  Why would we
think this difference wouldn't happen?  But, difference is not
necessarily a sign of divergence.

More.  We have seen in the (recent) age of computer assisted
human Chess playing moves not previously recorded in human
Chess competitions.

To me, all this means is that if you play Chess and Go
differently you'll probably see moves you've not seen before.
Are these signs of divergence?  No, I don't think so.  They
are just differences due to the different ways Chess and Go
are now played.

I think we'll see the same kind of thing happen when we have
robot cricket players: differences, sure; divergence, no.

Best regards,

Tim

PS: The real divergence is, of course, that humans think and
machines don't.  Some humans just think machines think -:)

PPS: Notice, our thinking seems to get in the way of us
playing Chess or Go against ourselves in a way that lets us
think we are playing a different (human) Chess or Go player
(or computer Chess or Go program).  Perhaps the fact the
AlphaZero and AlphaGo appear to have no trouble doing this is
evidence that they don't think -:)



> On 29 Nov 2021, at 09:03, Humanist <humanist@dhhumanist.org> wrote:
>
>                  Humanist Discussion Group, Vol. 35, No. 373.
>        Department of Digital Humanities, University of Cologne
>                               Hosted by DH-Cologne
>                       www.dhhumanist.org
>                Submit to: humanist@dhhumanist.org
>
>
>
>
>        Date: 2021-11-29 07:55:43+00:00
>        From: Willard McCarty <willard.mccarty@mccarty.org.uk>
>        Subject: convergence and divergence
>
> Here's a problem. I'd like to know if anyone has given thought to it.
>
> The steady work in computer science and engineering, including robotics
> and AI, to make computing systems more like ourselves has not yet, and
> likely never will, equal the promotional rhetoric with which we are
> bombarded, but its incremental success can hardly be challenged. At the
> same time, notably with Deep Mind's AlphaGo Zero, significant
> divergences have emerged between human and artificial intelligences,
> e.g. moves in Go that no one in the last few millennia of the game have
> thought of before. Actually, divergence of human and artificial
> intelligences began with the digital machine and have antecedents going
> way back, made less obvious than should be by regarding them as failures
> to be intelligent, or as intelligent, smart etc.
>
> The interesting problem for us, it seems to me, lies in the interplay
> between convergence and divergence. How can it be used in beneficial ways?
>
> Comments?
>
> Yours,
> WM
> --
> Willard McCarty,
> Professor emeritus, King's College London;
> Editor, Interdisciplinary Science Reviews;  Humanist
> www.mccarty.org.uk



_______________________________________________
Unsubscribe at: http://dhhumanist.org/Restricted
List posts to: humanist@dhhumanist.org
List info and archives at at: http://dhhumanist.org
Listmember interface at: http://dhhumanist.org/Restricted/
Subscribe at: http://dhhumanist.org/membership_form.php