Home About Subscribe Search Member Area

Humanist Discussion Group

< Back to Volume 34

Humanist Archives: Aug. 7, 2020, 7:41 a.m. Humanist 34.216 - on GPT-3

                  Humanist Discussion Group, Vol. 34, No. 216.
            Department of Digital Humanities, King's College London
                   Hosted by King's Digital Lab
                Submit to: humanist@dhhumanist.org

        Date: 2020-08-06 15:40:03+00:00
        From: Brigitte Rath 
        Subject: AW: [Humanist] 34.214: on GPT-3

Many thanks, Tim, Jim, Bill, Mark and Willard, for all your stimulating

Jim Rovira points out that textual meaning is produced in the act of reading, in
the interplay of text with something else. I'd like to take this observation as
a starting point to think a bit more about GPT-3's production of text as a
specific test case for semiotics.

We know that GPT-3's whole universe is encoded text. The only input it ever got
was encoded text, and it is its only output. Its model consists of weighted
relations between markers (let's call them "words*") which are all of the same
kind. This seems noteworthy to me: as Bill Benzon mentions, GPT-3 manipulates
words* without semantics or reference. Within GPT-3, words* only ever connect to
other words*, they cannot connect to concepts or objects. GPT-3's model of
language is entirely relational and entirely homogeneous.

Ferdinand de Saussure's famous structuralist model of language is also
relational, but it is heterogeneous: all signifiers -- "sound images" for
Saussure -- are of the same kind, and within the homogeneous set of signifiers
-- all of them "sound images" --, each signifier is defined precisely by being
different from all others. The same holds for all signifieds, concepts for
Saussure. A sign is formed when a signifier is connected to a signified -- a
sound image to a concept -- and thus when *categorically different* units, each
defined differentially within its own homogeneous system, are brought together.
Signification arises out of a *heterogeneous* system.

For me that leads to two different thoughts:

(1) What happens when a large machine learning algorithm is fed two or more
_different_ sets of inputs with the model tasked to build not one (as GPT-3),
but two or more homogeneous relational systems which are categorically different
from each other, and to connect them together, creating relationships between
heterogeneous units and thus a structure of signification?

(2) Willard McCarty asks about the "perspectives or inclinations with which we
begin forming the questions we ask," and names mimesis and "nothing-more-ism." I
am, first and foremost, a reader, and I'd describe my own inclination as looking
for a different stream of input to connect with the ones I already process, with
the expectation that these additional heterogeneous connections create new signs
which may allow for new meanings. GPT-3 may provide one for me.

I am still fascinated by GPT-3, and thus both curious about and very grateful
for others' thoughts.


Von: Humanist [humanist@dhhumanist.org]
Gesendet: Mittwoch, 05. August 2020 10:00
An: publish-liv@humanist.kdl.kcl.ac.uk
Betreff: [Humanist] 34.214: on GPT-3

                  Humanist Discussion Group, Vol. 34, No. 214.
            Department of Digital Humanities, King's College London
                   Hosted by King's Digital Lab
                Submit to: humanist@dhhumanist.org

        Date: 2020-08-04 07:35:52+00:00
        From: Willard McCarty 
        Subject: questions about questions

In one of my favourite novels, Zia Haider Rahman's In the Light
of What We Know, the narrator, questioning his own thoughts, muses
that "the only answers each of us hears are to the questions we are
capable of asking.” In the present context, this suggests to me that
Brigitte Rath's "What kinds of questions does GPT-3 raise for you?"
(Humanist 34.203) is perhaps more worth pursuing in an open
discussion than we may have realised.

So, to respond to her: mine are about the perspectives or inclinations
with which we begin forming the questions we ask.

The first is the orientation to mimesis, or rather, the limitation of
perspective to only that. If something not of our biological kind can
successfully do what we do, so that it passes the 'test' that Turing
never named a test, rather a 'game', then it raises a question, or a
bundle of them. This is something the Japanese roboticist Masahiro Mori
understood (he of the "uncanny valley" fame). Imitation is only the
first step; that which an imitation of ourselves fails to do is, as
Jerome McGann once wrote, "the hem of a quantum garment", a.k.a.
a game-changer.

The second is the kind of response that Ananda Coomaraswamy once
identified with a particular Sanskrit term he translated as
"nothing-more-ism". (I'd be very grateful to know what this Sanskrit
term is and where Coomaraswamy discussed it, or better, if there's
other commentary on it.) In other words, the response that sums to
"this is nothing more than X", where X is a matter pointless to pursue.
Ok, 'intelligence', like 'consciousness', is a term that seems impossible
to get a handle on. But that does not mean that the effort to do so is
worthless. Anything but: consider, for example, these two very different
responses to those two elusive words:

1. On 'intelligence': G.E.R. Lloyd, Intelligence and intelligibility
(OUP, 2020);
2. On consciousness: Roger Penrose, "Physics and the mind", in Penrose,
Shimony, Cartwright and Hawking, The Large, the Small and the Human Mind
(CUP 1997); Lex Fridman's recent interview with Penrose, "Consciousness
is not a computation", https://www.youtube.com/watch?v=hXgqik6HXc0



Willard McCarty (www.mccarty.org.uk/),
Professor emeritus, Department of Digital Humanities, King's College
London; Editor, Interdisciplinary Science Reviews
(www.tandfonline.com/loi/yisr20) and Humanist (www.dhhumanist.org)

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

Editor: Willard McCarty (King's College London, U.K.; Western Sydney University, Australia)
Software designer: Malgosia Askanas (Mind-Crafts)

This site is maintained under a service level agreement by King's Digital Lab.