Humanist Discussion Group, Vol. 39, No. 11.
        Department of Digital Humanities, University of Cologne
                      Hosted by DH-Cologne
                       www.dhhumanist.org
                Submit to: humanist@dhhumanist.org




        Date: 2025-05-13 10:23:52+00:00
        From: Manfred Thaller <manfred.thaller@uni-koeln.de>
        Subject: Re: [Humanist] 39.6: repetition vs intelligence

Dear Gabriel, Dear Willard,

Allow me a few tangential comments upon your recent exchange:

>          Date: 2025-05-08 07:52:11+00:00
>          From: Gabriel Egan<mail@gabrielegan.com>
>          Subject: Re: [Humanist] 39.4: repetition vs intelligence?
>
> Willard asks:
>
>   > . . . to what extent, in what ways, do the strategies of
>   > the so-called Large Language Models produce results that
>   > only echo back to us current linguistic behaviour (parole),
>   > in effect saying nothing new, however useful, however news
>   > to the questioner?
>
> I should say that the extent to, and ways in, which they
> do this is about the same as the extent to, and ways in,
> which people do.
>
>   > . . . being truly creative, is exceedingly rare. But
>   > isn't that exactly what we want of intelligence
>
> If we set the bar for intelligence that high, most
> of our fellow human beings -- and I think the likes
> of me too -- fall below the threshold. That is a
> politically dangerous way to define intelligence.
> Even Ayn Rand set the bar lower than that.
>
> Regards
>
> Gabriel
>

Observation / Comment 1:

It seems to me that two concepts are mixed up here, which are connected,
but not identical.

If I drive across a bridge, I very much hope that its design was checked
by a structural engineer, who was totally repetitive in the application
of well established formulae. And I would certainly grant him the
property of intelligence.

If architects for the 91st time apply Palladian proportions to construct
a villa, they are presumably intelligent as well. Nevertheless, many
people might roll their eyes while reluctantly trying to appreciate this
91st repetition. Though others might appreciate it very much not to be
forced out of their comfort zone of well established tastes.

The difference I am trying to point out is obviously the one between
intelligence and creativity. Both of which I would not consider binary
properties (being either absent or present), but expressed by degrees on
a scale. (Cf. the last 567 publications of behavioral biologists
claiming "some" intelligence for their animals of observation.)

Both tasks - structural complications and designing buildings out of
rule sets - can and have been solved by algorithms, by the way, which
until very recently would NOT have been called AI, though they might today.

Observation / Comment 2:

Coming back to: "Though others might appreciate it very much not to be
forced out of their comfort zone of well established tastes."

Artists who are too creative usually find it rather hard to get
established. Remember the saga of the impressionists, and remember that
the Vienna "Secession" was named rebelliously for a reason.

I've chosen an example from the pictorial arts, because they present us
with an interesting observation when it comes to the replacement of a
human skill by mechanical devices. The eighteenth century lady and her
gentleman most certainly had a very fine and developed taste when
choosing a painter for their portraits. Nevertheless they would not be
very convinced by a painter who was so creative, that their
acquaintances could not recognize their likeness. Much like my
enthusiasm for a creative structural engineer building bridges crashing
beneath me would be severely restrained.

Now we all have heard and read many times, that a painted portrait much
better catches the spirit of a person than a photograph. Some of us even
believe it. Nevertheless competing with a camera becomes boring after
some time - so the pictorial arts have (almost) completely dropped the
idea of visually representing reality. ["Almost": I will watch with
interest, what happens to the remaining /Trompe-l'œil /artists under the
influence of generative AI.] Indeed, not trying to represent reality
within art has lead quite a few pictorial artists into happenings and
performances ... where they become almost indistinguishable from former
actors and stage directors who once tried to give a plausible
presentation of a human acting in a situation drawn from real or
imagined life, formerly known as "drama". "Giving a plausible
presentation of a human acting in a situation drawn from real or
imagined life" simply works much better with cinematic means.

I'll watch with great interest what happens to "creative writing" in the
next decade or so. When the next two thousand coming of age novels are
written by LLMs, what will their current authors do? Founding another
subgenre of happenings and performances?

Observation / Comment 3:

While my previous observation / comment has shown me as a blockheaded
Philistine who never sufficiently appreciated true art, I have a much
closer relationship to academia.

Coming back to the difference between intelligence / competence vs.
creativity: When talking about research politics, particularly about
university funding, I always claimed that the point of the university
was <emph>not</emph> to teach people the answers we already knew, but
how to go about answering questions which we have not yet discovered.
Being retired and not having to think about politics anymore, I may
finally confess, that I wonder, whether we would not need a small number
of institutions of a new type, focusing on this task. While existing
universities may finally and unrestrictedly embrace the task of
propagating fluctuating canons. LLMs seem to me perfectly suited to the
later task.

Observation / Comment 4:

Gabriel comments "... echo back to us current linguistic behavior ..." with
> I should say that the extent to, and ways in, which they
> do this is about the same as the extent to, and ways in,
> which people do.

Yes, indeed. At least in the antediluvian period of university teaching
I was exposed to, you were not expected to pass a course in academic
writing, but you were expected to read academic literature, preferably a
lot, and by that act learn to write as the people had written whom you
read. Allegedly only from a stylistics point of view; but in reality
more or less intentionally doing what charitably might be described as
"learning by example also contentwise", less charitably picking up
semi-consciously all sorts of silent assumptions in the chain of arguments.

If as a historian you read systematically 19th century texts of
completely different types - historiography, non-fiction books,
journals, newspapers - you soon notice, that there is a underflow which
they have in common, quite independent of their intellectual level. What
they have in common is a flurry of nationalist truisms (in all countries
I encountered) which are so ubiquitous, that none of the authors have
probably even noticed all of them. And only extremely few on rare
occasions woke up to the fact, that their carefully built argumentative
chains were meaningless unless you endorsed that underflow. So all of
these texts - I repeat, quite independently from their intellectual level
- consist of two parts: an extremely tiny component, trying to convey
something [sometimes even something original] and a flood of linguistic
noise which is just there, because it's always there.

  I find it extremely easy to discover the same structure in
contemporary texts, albeit the truisms are not nationalist anymore, but
obviously as unreflected (even if I sympathize more with them). To
recognize yourself is always the most difficult: so it took me sometime
to understand, that the same holds true for the texts having arisen
during my own academic socialization, mainly in the seventies and eighties.

And I definitely think, that the "linguistic noise which is just there,
because it's always there" can perfectly be produced by "Large Language
Models produc[ing] results that only echo back to us current linguistic
behaviour (parole)". Obviously for Human understanding this embedding
noise is necessary, maybe slowing down perception to a sufficiently low
speed that can be handled comfortably. Short provocation: Would a stage
be possible, in which historians could just formulate tiny bits of new
knowledge to be embedded by Articificial "Intelligence" into
sufficiently much contemporary babble to make it digestible periodically
for succeeding generations? (In historiography I have at least some
suspicions; in other disciplines I have not.) Well ... probably not. If
the LLMs are trained on texts produced by LLMs they become stale. The
memetic equivalent of genetic inbreeding, one might say. In our
provocative line this simply means that generations would not need
retranslations because they get totally stagnant intellectually.
Hopefully they would eventually get bored out of stagnation, though.

Observation / Comment 5: Willard, your paper "The residue of uniqueness"
(https://nbn-resolving.org/urn:nbn:de:0168-ssoar-378398) I usually
summarized for my students (possibly against your intentions) as
"Willard assumes, that the true Humanities question is what remains, if
you subtract anything that can be entrusted to a computer." Do you see
any relation to observation / comment 4 above? 

Yours truly, Manfred


_______________________________________________
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