Humanist Discussion Group

Humanist Archives: Nov. 29, 2022, 8:15 a.m. Humanist 36.270 - a robot wrote this, and does other things too

				
              Humanist Discussion Group, Vol. 36, No. 270.
        Department of Digital Humanities, University of Cologne
                      Hosted by DH-Cologne
                       www.dhhumanist.org
                Submit to: humanist@dhhumanist.org


    [1]    From: Tim Smithers <tim.smithers@cantab.net>
           Subject: Re: [Humanist] 36.263: a robot wrote this (159)

    [2]    From: AEOLIAN Project <Aeolian@lboro.ac.uk>
           Subject: New articles on AI and digital archives (41)


--[1]------------------------------------------------------------------------
        Date: 2022-11-28 11:46:48+00:00
        From: Tim Smithers <tim.smithers@cantab.net>
        Subject: Re: [Humanist] 36.263: a robot wrote this

Dear Thomas and Willard,

Thank you, Thomas, for explaining the weaknesses of the GPT-3
outputs on feuilleton writing.  As a partial response to your
concerns, Willard, I would say it's comments and explanations
like Thomas' that we need more of, to counter the more noisy,
and often more plentiful, claims of how impressive GPT-3, and
its ilk, are.

Thomas, when you wrote originality and creativity are
important maxims for feuilleton writing, I didn't know what
feuilleton writing was.  The only place I can remember seeing
this term, I think, is in The Glass Bead Game, by Hermann
Hesse, but I don't remember what, if anything, it said about
it.

So, following an example of what you can do with GPT-3, I
thought to ask GPT-3 about this.  Here's the input, and the
first five alternative outputs I got.

 Input: Explain what feuilleton writing is

 Output 1

  Feuilleton writing is a type of writing that is usually
  found in newspapers and magazines.  It is often light and
  entertaining, and can be about any subject.

 Output 2

  Feuilleton writing is a style of writing that is light and
  often satirical.  It is often found in newspapers and
  magazines, and it is often used to discuss current events.

 Output 3

  Feuilleton writing is a type of journalism that is
  light-hearted and often literary in nature.  It is named
  after the French word for "little leaf," which is a small
  section of a newspaper that is devoted to this type of
  writing.

 Output 4

  Feuilleton writing is a type of writing that is usually
  found in newspapers and magazines.  It is usually light and
  entertaining in nature, and often deals with topics such as
  fashion, celebrities, and other popular culture.

 Output 5

  A feuilleton is a type of writing, typically found in
  magazines and newspapers, that is light, entertaining, and
  often trivial.

Once again, I think we can't say these are wrong, but none of
them, I would say, help much in understanding what feuilleton
writing is.  The differences between the repeated outputs also
don't add anything, it seems to me.  They look more like poor
attempts to avoid accusations of Plagiarism.

GPT-3, we are told, was 'trained' on massive amounts of
(mostly) English text scrapped off the web.  Almost certainly,
some (tiny mount) of these words came from the Wikipedi page
on Feuilleton here <https://en.wikipedia.org/wiki/Feuilleton>.
Once again, I'm not qualified to judge how good this Wikipedia
page is, but it looks a whole lot better than anything GPT-3
seems able to cough up.  So, after immense amounts of
computation, and plenty of CO2 emissions, we have a system
that squashes out all the human intelligence behind the texts
it's been 'trained' on, and gives us, in return for all this
effort, as much as you want of bland next to useless text.

To me, this doesn't look like intelligent behaviour, on the
part of GPT-3, nor on the part of the people who think GPT3
displays any kind of intelligence.  This is, I would say,
re-using the title of Drew McDermott 1976 paper, an example of
"Artificial Intelligence Meets Natural Stupidity,"
<https://dl.acm.org/doi/10.1145/1045339.1045340>.

Which brings me back to your question, Willard, "...  why so
much effort is spent in imitating human behaviour ... and why
is it that we seem to be impressed by its successes?  ..."  An
answer, it seems to me, is that work in AI has done more to
show us Humans how stupid we can be than it has helped us
understand what intelligent behaviour can be, ours or that of
"the machine".  Perhaps it's this apparent delight in seeing
our own sillinesses replicated in the artificial that gets in
the way of some serious attempts to understand the "true
nature of the machine"?  AI has, I would say, always had it's
over enthusiastic claims, but it wasn't always as silly, not
to say dishonest, as it seems to have become today.

More humans pointing this out would help.  This is where the
Humanities come in, and the Arts too, I think.

Best regards,

Tim



> On 22 Nov 2022, at 09:59, Humanist <humanist@dhhumanist.org> wrote:
>
>
>              Humanist Discussion Group, Vol. 36, No. 263.
>        Department of Digital Humanities, University of Cologne
>                      Hosted by DH-Cologne
>                       www.dhhumanist.org
>                Submit to: humanist@dhhumanist.org
>
>
>    [1]    From: Thomas Gloning <thomas.gloning@germanistik.uni-giessen.de>
>           Subject: a robot wrote this / theatre criticism (235)
>
>    [2]    From: Willard McCarty <willard.mccarty@mccarty.org.uk>
>           Subject: the bar set low (27)
>
>
> --[1]------------------------------------------------------------------------
>        Date: 2022-11-21 20:27:07+00:00
>        From: Thomas Gloning <thomas.gloning@germanistik.uni-giessen.de>
>        Subject: a robot wrote this / theatre criticism
>
> Dear Tim
>
> many thanks for putting GPT-3 to the test. The system demonstrates that
> it is able to address some specific topics and to use relevant
> expression patterns. What I did not find are the more elaborate forms of
> originality and creativity so characteristic for feuilleton writing.
>
> In a realistic perspective, there are two more problems.
>
> First, in order to write a piece of 'real' theatre criticism one has to
> go to the theatre and actually watch the play.
>
> Second, in order to give a sound evaluation of the play in question one
> needs broad knowledge and experience with the standards of critical
> evaluation.
>
> While it seems possible to generate standard reports on football games
> from the information gained from data bases, this method does not seem
> viable in respect of forms of criticism.
>
>> The play "Every Monday" by German playwright Bertolt Brecht
>
> That was funny!
>
> All best wishes and thanks again
>
> Thomas
>
> P.S.: For those with a command of German:
>
> Gloning, Thomas: "Man schlürft Schauspielkunst ...". Spielarten der
> Theaterkritik. In: Hagestedt, Lutz (Hg.): Literatur als Lust.
> Begegnungen zwischen Poesie und Wissenschaft. Festschrift für Thomas
> Anz. München: belleville 2008, 59-86.


--[2]------------------------------------------------------------------------
        Date: 2022-11-28 09:21:56+00:00
        From: AEOLIAN Project <Aeolian@lboro.ac.uk>
        Subject: New articles on AI and digital archives

Dear all,

Interested in digital archives and new technologies such as AI?

These new articles might be of interest:

“Applying AI to Digital Archives: Trust, Collaboration and Shared Professional
Ethics” co-authored with Arran Rees. Digital Scholarship in the Humanities
(published November 2022) *OPEN ACCESS*

https://doi.org/10.1093/llc/fqac073

“How Can We Make Born-Digital and Digitised Archives More Accessible?
Identifying Obstacles and Solutions.” Archival Science 22 (2022): 417-36. *OPEN
ACCESS*

https://doi.org/10.1007/s10502-022-09390-7

“Unlocking digital archives: cross-disciplinary perspectives on AI and born-
digital data” co-authored with Annalina Caputo. AI & Society 37 (2022): 823-35.
*OPEN ACCESS*

https://doi.org/10.1007/s00146-021-01367-x


Best wishes,
Lise
---
Dr Lise Jaillant | Senior Lecturer (Associate Professor)
School of Social Sciences and Humanities | Loughborough University, UK

UK PI for three AHRC-funded projects:


  1.  LUSTRE<https://lustre-network.net/> (Unlocking our Digital Past with AI),
in partnership with the UK Cabinet Office
  2.  EyCon<https://eycon.hypotheses.org/> (Visual AI and Early Conflict
Photography) in partnership with French researchers
  3.  AEOLIAN<https://www.aeolian-network.net/> (UK/ US: AI for Cultural
Organizations)



_______________________________________________
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