Humanist Discussion Group, Vol. 35, No. 536.
Department of Digital Humanities, University of Cologne
Hosted by DH-Cologne
www.dhhumanist.org
Submit to: humanist@dhhumanist.org
[1] From: Dr. Herbert Wender <drwender@aol.com>
Subject: Re: [Humanist] 35.529: GPT-3 and generated poetry (15)
[2] From: Mark Wolff <wolff.mark.b@gmail.com>
Subject: Re: [Humanist] 35.529: GPT-3 and generated poetry (156)
[3] From: Gioele Barabucci <gioele.barabucci@ntnu.no>
Subject: Re: [Humanist] 35.529: GPT-3 and generated poetry (37)
[4] From: maurizio lana <maurizio.lana@uniupo.it>
Subject: Re: [Humanist] 35.529: GPT-3 and generated poetry (26)
--[1]------------------------------------------------------------------------
Date: 2022-02-15 23:25:19+00:00
From: Dr. Herbert Wender <drwender@aol.com>
Subject: Re: [Humanist] 35.529: GPT-3 and generated poetry
Willard,
if I undertand you correctly the true potential of ELIZA like conversations you
would recognize in the surprise on the human side. Such an account I would see
in analogy to Kant's remark that the Sublime is not a property of the obserbed
but is to be found in the feelings of the subject. Applied on the so-called poem
produced by GPT3 that would mean: Its 'value' is not t be judged without the
possible interpretations of it. In principle the case seems to be like the case
of a very late porm by Hölderlin - "Die Aussicht" - analysed by Jakobson et al.
Cf. Roman Jakobson, Grete Lübbe-Grothues and Susan Kitron (1980): The Language of
Schizophrenia: Holderlin's Speech and Poetry. Poetics Today Vol. 2, No. 1a, Roman
Jakobson:Language and Poetry, pp. 137-144.
Kind regards, Herbert
--[2]------------------------------------------------------------------------
Date: 2022-02-15 21:24:09+00:00
From: Mark Wolff <wolff.mark.b@gmail.com>
Subject: Re: [Humanist] 35.529: GPT-3 and generated poetry
I wonder if this all somehow points back to Francis Bacon, who, in the
Advancement of Learning, wrote:
The invention of speech or argument is not properly an invention, for to
invent is to discover that we know not, and not to recover or resummon that
which we already know: and the use of this invention is no other but out of
the knowledge whereof our mind is already possessed to draw forth or call
before us that which may be pertinent to the purpose which we take into our
consideration. So as to speak truly, it is no invention, but a remembrance
or suggestion (II.xiii.6).
A few years ago I experimented with machine learning tools to build
something that generates centos interactively, using word2vec to define
computational analogies that in turn retrieve verses in large corpora:
https://markwolff.name/wp/digital-humanities-2/invent-your-own-sonnet-using-
analytic-tools-to-synthesize-texts/
GPT-3 is completely different. Instead of relying on highly structured and
curated training data, it is able to work with heterogeneous data in order
to respond to "plain" English queries with something that closely resembles
"plain" English. GPT-3 is not recycling verses to make centos. It is
recycling language based on inferred structures in the training data. Is
the GPT-3 output "new" in the Baconian sense of discovery (or in the more
modern sense of creation)? Or is it merely "a remembrance or suggestion" of
the data it learned from?
With GPT-3, using the same input multiple times produces different outputs.
I asked GPT-3 a second time, "Write a short poem like Tristan Tzara." And
it responded:
I do not know what I am saying
The first time I try
I am drunk with words
And the second time too
I am drunk with words
But I know what I am saying
And the third time I am not drunk
But I am still saying words
And the fourth time I am not drunk
And I am not saying words
But I am still writing them
Maybe GPT-3 is referring to the method prescribed for writing a Dadaist
poem (https://www.writing.upenn.edu/~afilreis/88v/tzara.html) but in that
case it's not remembering text it learned from, it's commenting on that
text in an original way.
Willard would like to see interesting applications with AI that provoke
conversation for humans. We might be there already.
mw
On Tue, Feb 15, 2022 at 2:18 AM Humanist <humanist@dhhumanist.org> wrote:
>
> Humanist Discussion Group, Vol. 35, No. 529.
> Department of Digital Humanities, University of Cologne
> Hosted by DH-Cologne
> www.dhhumanist.org
> Submit to: humanist@dhhumanist.org
>
>
> [1] From: James Rovira <jamesrovira@gmail.com>
> Subject: Re: [Humanist] 35.526: GPT-3 and generated poetry (30)
>
> [2] From: Willard McCarty <willard.mccarty@mccarty.org.uk>
> Subject: GPT-3 (or similar) and generated conversation? (41)
>
>
>
> --[1]------------------------------------------------------------------------
> Date: 2022-02-14 15:02:41+00:00
> From: James Rovira <jamesrovira@gmail.com>
> Subject: Re: [Humanist] 35.526: GPT-3 and generated poetry
>
> I think if there's one kind of poem a computer will always be able to
> generate, it's a cento :).
>
> Jim R
>
> On Mon, Feb 14, 2022 at 1:37 AM Humanist <humanist@dhhumanist.org> wrote:
>
> > I do not know what I am saying
> > The world is spinning round
> > I am dizzy and I cannot see
> > I want to sleep
> > And when I wake up
> > I will see things more clearly
> >
> >
> > IMHO this is a pretty good summary of Tzara's poetry.
> >
> > I'm not sure what to make of this. Thoughts?
> >
> > mw
> > --
> > Mark B. Wolff, Ph.D.
> > Professor of French
> > Chair, Modern Languages
> > One Hartwick Drive
> > Hartwick College
> > Oneonta, NY 13820
> > (607) 431-4615
> >
> > http://markwolff.name/
>
>
> --[2]------------------------------------------------------------------------
> Date: 2022-02-14 07:05:10+00:00
> From: Willard McCarty <willard.mccarty@mccarty.org.uk>
> Subject: GPT-3 (or similar) and generated conversation?
>
> This in response to Mark Wolff's question of yesterday. So far, as far
> as I know, the trials have been one-off performances in which someone
> sends something to GPT-3, something comes back, and we all marvel,
> grumble, comment etc. Has anyone tried anything conversational with such
> an agent? Interchanges with ELIZA and the like depended on close
> imitation of what one might expect another person, in that case someone
> in the manner of a therapist, to say, and naive acceptance by the human
> who initiates. What if the human using GPT-3 were to attempt a
> conversational exchange?
>
> Some years ago a student at Berkeley, Dunbar Aitkens, invented a game
> inspired by Hermann Hesse's Glasperlenspiel, which he and his friends
> called The Glass Plate Game. See <https://glassplategame.com> for the
> current (fossilised?) state of his project. It uses images on cards to
> induce conversation among people in a group. I have not plumbed the
> depth of the game, but I can see how it might serve to call up hidden
> thoughts and develop them. One can imagine a computational --
> combinatorial -- implementation.
>
> For something along the lines of a GPT-3 conversation or development of
> The Glass Plate Game to prove useful, perhaps therapeutic, wouldn't the
> problem be to find the 'sweet spot' between mirrored responses and
> utterly chaotic ones? The "edge of chaos" in complexity theory comes to
> mind. This figure of speech unfortunately suggests something like an
> external border one close to, but the 'space' in which interesting things
> happen is far more interesting than that.
>
> For me the significant point in all this is reached when the interchange
> becomes a conversation that unexpectedly provides insight for the human
> participant(s). The emphasis, it seems to me, needs to be on the
> conversation, not the mechanism, whether this be GPT-3 or a bunch of cards.
>
> Comments?
>
> Yours,
> WM
> --
> Willard McCarty,
> Professor emeritus, King's College London;
> Editor, Interdisciplinary Science Reviews; Humanist
> www.mccarty.org.uk
--[3]------------------------------------------------------------------------
Date: 2022-02-15 13:50:34+00:00
From: Gioele Barabucci <gioele.barabucci@ntnu.no>
Subject: Re: [Humanist] 35.529: GPT-3 and generated poetry
On 15/02/22 08:18, Humanist wrote:
> This in response to Mark Wolff's question of yesterday. So far, as far
> as I know, the trials have been one-off performances in which someone
> sends something to GPT-3, something comes back, and we all marvel,
> grumble, comment etc. Has anyone tried anything conversational with such
> an agent? Interchanges with ELIZA and the like depended on close
> imitation of what one might expect another person, in that case someone
> in the manner of a therapist, to say, and naive acceptance by the human
> who initiates. What if the human using GPT-3 were to attempt a
> conversational exchange?
Such uses of GPT-3 are prohibited by its license.
There have been a couple of attempts at making chatbots with GPT, but
all have been closed for breaking GPT's terms of use.
One such chatbot worth of note, at least for its background story, is
Project December [0] a chatbot that simulated the deceased girlfriend of
its developer, Jason Rohrer.
(While we are at it, let's spend some words on Jason Rohrer. Jason
Rohrer defines himself as as game programmer. He is definitely worth
being called an artist. His work Passage [2] is part of MOMA's permanent
collection [3]. His other game Gravitation [4] should be as well. All
people that I know that played either games have been profoundly moved
by them.)
[0] https://projectdecember.net/
[1] http://hcsoftware.sourceforge.net/jason-rohrer/
[2] http://hcsoftware.sourceforge.net/passage/
[3] https://www.moma.org/collection/works/145533
[4] http://hcsoftware.sourceforge.net/gravitation/
--
Prof. Dr. Gioele Barabucci <gioele.barabucci@ntnu.no>
Associate Professor of Computer Science
NTNU — Norwegian University of Science and Technology
--[4]------------------------------------------------------------------------
Date: 2022-02-15 09:03:33+00:00
From: maurizio lana <maurizio.lana@uniupo.it>
Subject: Re: [Humanist] 35.529: GPT-3 and generated poetry
hi,
I agree with the comment of Jim about the cento.
I would add the we humans have a tremendous capability of
interpreting, in the sense of creating to ourselves reasons for
considering any object or phenomenon interpretable. particularly in
front of the artifacts of a supposedly intelligent agent.
if one of us came across the quoted poem without knowing that it is
a product of GPT-3, I don't think we would pay much attention to it.
the conversation (Willard: "The emphasis, it seems to me, needs to
be on the conversation, not the mechanism, whether this be GPT-3 or
a bunch of cards") is a much more complex situation in which an AI
system could be put to the test - as long as the human interlocutor
does not fall into that said adoration of assuming that every
artefact of a supposedly intelligent agent is an expression of
intelligence
Maurizio
Maurizio Lana
Dipartimento di Studi Umanistici
Università del Piemonte Orientale
piazza Roma 36 - 13100 Vercelli
tel. +39 347 7370925
_______________________________________________
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
