Humanist Discussion Group, Vol. 35, No. 655.
Department of Digital Humanities, University of Cologne
Hosted by DH-Cologne
www.dhhumanist.org
Submit to: humanist@dhhumanist.org
Date: 2022-04-16 11:40:22+00:00
From: Dino Buzzetti <dino.buzzetti@gmail.com>
Subject: Re: [Humanist] 35.653: the slippery cliff
Dear Willard,
Thank you for your letter that poses a question which is absolutely
crucial after the so-called connectionist turn in Artificial Intelligence.
What kind of reliability can we assign to deep learning results ?
Only two brief observations that come suddenly to my mind. The first
is about the field «expertness». In my opinion it is not a matter depending
on a particular domain, but a more general epistemological question,
namely on whether we are approaching a problem from a systemic point
of view, as cybernetics does, or not. In other words the problem is more
of a logical one than a bound one to a specific subject matter. You have to
decide whether you are dealing with a deterministic or a nondeterministic
context.
So, what criteria do you have to assume to check the soundness of
a nondeterministic inference ? In a recent article on the connectionist
turn in AI, Dominique Cardon et al.—"Neurons spike back. The invention
of inductive machines and the artificial intelligence controversy",
Réseaux, 2018/5 (No 211), p. 173-220. URL: https://www.cairn-int.info/journal-
reseaux-2018-5-page-173.htm — refer simply to induction. In my
opinion this shoud be better specified, for more properly we have
to do with *abduction*, dealt with more precisely by Ch. S. Peirce
as a special case of induction.
Briefly, you decide that an observed fact is a case of an assumed
hypothesis on the ground of previously observed cases. And this is
how supervised learning proceeds. This is only a sufficient condition
for the the observed cases and the assumed hypothesis, but it is
not a sufficient condition for all possible cases in general. The more
cases you observe, the more probable is the ascertainment of their
dependence on the assumed hypothesis. New observed facts can
change, from a systemic point of you, the whole set of their mutual
interconnections and falsify the assumed hypothesis.
I would think that this procedure is not a case of «slippery cliff», but
that it still remains within the bounds of a not so dangerously treadable
«slope».
Yours, -dino
.
On Sat, 16 Apr 2022 at 09:05, Humanist <humanist@dhhumanist.org> wrote:
>
> Humanist Discussion Group, Vol. 35, No. 653.
> Department of Digital Humanities, University of Cologne
> Hosted by DH-Cologne
> www.dhhumanist.org
> Submit to: humanist@dhhumanist.org
>
>
>
>
> Date: 2022-04-16 06:57:56+00:00
> From: Willard McCarty <willard.mccarty@mccarty.org.uk>
> Subject: the slippery cliff
>
> In 1950, the seventh Macy Conference on Cybernetics: Circular Causal and
> Feedback Mechanisms in Biological and Social Systems (the title of the
> Proceedings) met in New York. Among those present were Warren McCulloch,
> Heinz von Foerster, Gregory Bateson, Julian Bigelow, Heinrich Klüver,
> Margaret Mead, Walter Pitts, John von Neumann, J.C.R. Licklider, Claude
> Shannon... you get the idea. The format was one in which presenters
> started delivering their papers, then attendees interrupted and
> discussion ensued. Lively, you might say. University of Chicago
> physiologist Ralph W. Gerard's introductory remark to his paper* sums up
> the experience: "this particular group is the most provocative one with
> which I am associated. I owe more new ideas and viewpoints to the
> meetings we have had over the past few years than to any other similar
> experience". The sort of thing that a scholar lives for.
>
> What Gerard went on to say thus has particular weight, and is the reason
> I detain you to consider:
>
> > It seems to me, in looking back over the history of this group, that
> > we started our discussions and sessions in the »as if« spirit.
> > Everyone was delighted to express any idea that came into his mind,
> > whether it seemed silly or certain or merely a stimulating guess that
> > would affect someone else. We explored possibilities for all sorts of
> > »ifs.« Then, rather sharply it seemed to me, we began to talk in an
> > »is« idiom. We were saying much the same things, but now saying them
> > as if they were so...
>
> > Since this group has been the focus and
> > fountainhead of thinking along these lines, we surely have a very
> > real responsibility, both internally and externally. Internally,
> > since we bring expertness in such varied fields, no one can be sure
> > another’s statements are facts or guesses unless the speaker is
> > meticulous in labeling suggestions as such. Externally, our
> > responsibility is even greater, since our statements and writings –
> > which may extend beyond an immediate area of competence – should not
> > give a spurious certainty to a credulous audience...
>
> I call this point of sudden transition between 'as-if' and 'is' the
> slippery cliff (too fast to be a slope). Does this not require near
> constant vigilance? Those familiar with the literature on modelling will
> recognise it immediately. I would hope that observers of AI would as well.
>
> Comments?
>
> Yours,
> WM
>
> *Ralph W. Gerard, "Some of the problems concerning digital notions in
> the central nervous system", in Heinz von Foerster, ed. Cybernetics:
> Circular causal and feedback mechanisms in biological and social
> systems. Transactions of the Seventh [Macy] Conference, March 23-24,
> 1950, New York, N.Y. New York: Joseph Macy, Jr. Foundation, 1951.
> Reproduced in Claus Pias, ed. Cybernetics: The Macy Conferences
> 1946-1953. Zürich/Berlin: diaphanes, 2003/2016.
>
> --
> Willard McCarty,
> Professor emeritus, King's College London;
> Editor, Interdisciplinary Science Reviews; Humanist
> www.mccarty.org.uk
--
Dino Buzzetti
Formerly: Department of Philosophy, University of Bologna
Currently: Fondazione per le Scienze Religiose Giovanni XXIII, Bologna
http://web.dfc.unibo.it/buzzetti
*https://www.fscire.it/equipe/ricercatori/card-1619018859
<https://www.fscire.it/equipe/ricercatori/card-1619018859>*
_______________________________________________
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
