Kasparov-Deep Blue, Navier Stokes-AI:
Human versus machine metaphor quickly calcifying
« previous post | next post »
A few weeks ago, I got an interesting email from Jean Charconnet, referencing a 9/10/2026 LLOG post, noting general issues about the logic of analogies and metaphors, and offering to submit a guest post on the topic.
I agreed, and couple of days later he sent me a fascinating document, which starts by citing his doctoral dissertation on analogy, later published in 2003 as « Analogie et logique naturelle », and its roots in earlier work by Jean-Blaise Grize.
I immediately ran aground trying to locate copies of those references, and/or other ways to learn about the intellectual world they come from. And my life over the past couple of weeks has been dominated by the need to wade through through various "AI HR" swamps*, so I haven't been able to get back to the search.
To avoid further delay, I've formatted Jean's guest post as an HTML document, more suitable to its style than LLOG's antique WordPress theme, and here it is. Enjoy!
* To be clear, the local HR representatives have been very helpful in guiding me through the mudholes created by the outsourced software claiming to put AI to work for us.
Kenny Easwaran said,
October 1, 2026 @ 5:58 pm
Really nice discussion of both where the metaphor works and where it doesn't. It's a little bit odd to read this in the voice of Claude, and also notable that after the first paragraph, one particular sentence stood out to me as *not* sounding Claude-like: "That loop is a story for another post." I don't know what in my psychology makes that particular sentence sound non-Claude-like. Especially because some of the other passages only sound Cllaude-like on the basis of very little data – for instance, the clause "The announcement arrived tangled in a priority dispute" primarily only sounds Claude-like to me because there was one time four months ago when I was asking Claude to summarize some ideas I was working on, and it used "tangled" in that same sort of way.
JPL said,
October 1, 2026 @ 8:45 pm
I don't have any extensive knowledge of the discourse conventions of the field of biology, but wrt the term 'genome' I don't see a consistent distinction between reference to the structures that determine concretely the unfolding of a particular individual organism, on the one hand, and on the other the general abstract structure that determines whether or not any given instance belongs to that species or not, e.g., when reference is to "the human genome" as opposed to "the dog genome". Only the former could be compared to a text, while the latter could be compared to the speech community norm for a natural language system, such as "English" vs "Yoruba". The "genome" for a species has objective existence just as the speech community norm does for a natural language system, an existence that persists and predates the instantiation of the system in an individual organism. But the concrete instantiation is a different empirical object from the abstract system it instantiates. The unfolding of a particular system of genetic information always takes place in an ongoing context of a history of particular organisms and the phenomenon of life. My hunch is that biologists will not want to acknowledge the objective existence of the abstract system, but that's what those who are trying to "map the genomes" of species are doing, just as descriptive linguists writing grammars of particular language systems take themselves to be describing real structures over and above those of individual speakers.
Jonathan D said,
October 1, 2026 @ 9:49 pm
I'm not convinced that calcification of this metaphor has been unusually quick. Charconnet mentions the use of the "Deep Blue-Kasparov moment" in the more fully analagous context of Go, but that didn't just happen after the 2016 victory for the algorithm. People were wondering when the moment would arrive well before that, and not only in Go, but about <a href="https://time.com/19866/ping-pong-robots-still-looking-for-their-deep-blue-kasparov-moment/"table tennis robots. My impression is that the comparison is by now completely unremarkable in the context of games.
Making that analogy in non-gamelike situations also did not start with the Buckmaster's comment. It may have been more frequent in the last five years than previously – examples include a claimed DB-K moment in film-making invoking the same machine beating human persective, as well as explicitly broader comparisons in terms of unexpectedness. However, my memory is that the Deep Blue attempt was framed as first steps towards broader "artificial intelligence" endeavours (and that framing was taken up by many commentators) from the start. Buckmaster's expression was taken up so easily because the idea that we are waiting for a similar man v machine moment in all sorts of fields is one that has been pushed around for decades.
JPL said,
October 2, 2026 @ 11:46 pm
WRT the phrase "Deep Blue-Kasparov moment", the process you are trying to describe is essentially the typical process or pathway for the origin and evolutionary development of lexemes, specifically for the category that constitutes its sense (or "meaning"). To understand it fully you have to start with the original sentence, "This is a Deep-Blue-Kasparov moment.", which expresses the logical act of assimilating an intentional object in the world (in this case the Navier- Stokes event) to a category of thought belonging to the language, on the model of a toddler grasping an object and putting it into its mouth in order to understand it. The original coining use only establishes an incipient category for the phrase through the comparison of the Navier-Stokes event to the 1997 event, the Kasparov- machine contest, but the phrase only becomes part of the language when you have a subsequent use of it ("Deep Blue-Kasparov moment") by a different user in reference to a different intentional object, a third one in this case. The modification of the sense of the expression 'Deep Blue-Kasparov moment' that the third use brings about is typical in the historical development of lexemes (in this case not a lexeme, but a phrasal term). I would suggest that the sense of this expression could be described not so much as "man vs machine", but something similar to the sense of another metaphorical phrase, "watershed moment", focusing on the directional change.
Jerry Packard said,
October 3, 2026 @ 9:21 am
Absolutely fascinating post. The one element that grabs me is JPL’s comment “…don't see a consistent distinction between reference to the structures that determine concretely the unfolding of a particular individual organism, on the one hand, and on the other the general abstract structure that determines whether or not any given instance belongs to that species or not, e.g., when reference is to "the human genome…”
To me, the ‘unfolding’ seems equivalent to ‘phenotype’, while the ‘abstract structure’ seems equivalent to ‘genotype’. Thus, the unfolding would seem to be not only phenotype, but epigenetic modification of the genotype that yields the phenotype.
Jonathan Smith said,
October 3, 2026 @ 12:44 pm
Re: the title’s "human vs. machine metaphor," this is rather a trope or a theme or a meme no?
If "Deep Blue – Kasparov moment," OK so Deep Blue > LLM, Kasparov > Buckmaster/Alpöge is metaphorical I guess. But (quite aside from the question of structural correspondence) this is pretty dull domain shift and thus pretty weak metaphor juice. Maybe "Deep Blue – Kasparov" per se is first a trope instance and later the archetypal instance if you will…
FWIW, well ahead of the Navier-Stokes kerfuff, Terence Tao was pointing out in various forums that (massively paraphrasing here) solutions to millennium problems and the like weren’t, upon reflection, very interesting in and of themselves. This is the point at which, in similar instances, industry folks whine about "moving the goal posts"… but what can you do; "AI" approaches here and elsewhere are always going to allow legitimate insights along the lines of solutions to millennium problems aren’t interesting in and of themselves; sorry about that.
Not totally irrelevant: back in the day, IBM went to some lengths to refute the notion that Deep Blue was "AI" – because it was badum just a computer of course.
Nice to see that this essay refers to LLMs as generating "syntactically impeccable, often locally convincing text through pattern completion, with no guarantee anywhere in the pipeline that meaning […] was ever built at all." Between the freakish Altruists (?) at Anthropic (see recent NYT write-up; who’s surprised we are dealing with red-pilled Evangelicals) and the utter bumblefucks in the U.S. government, the sane are increasingly numbered…
Oh yeah, and the Pain Axis people bait-and-switched their title/conclusions while the internet was freaking out about "LLM welfare" hahahah, see v2 vs v1 here. Why not this level of concern for Einstein’s brain cross-sections at the Mütter though? come now
JPL said,
October 3, 2026 @ 6:44 pm
If, as Kenny Easwaran's comment implies, the text of the guest post is "in the voice of Claude", why was this not mentioned anywhere in the OP? Is it a translation from French? And for Kenny, what made you conclude that the text of the guest post was a Claude production? (I did notice some infelicities of expression, but I put it down to non-native speaker use.) It's a puzzle.
Khaby Lame said,
October 4, 2026 @ 5:03 am
Really interesting discussion. The comparison between Kasparov–Deep Blue and modern AI shows how quickly our language for describing machine capabilities can become outdated. I think the bigger challenge is learning how to discuss AI without reducing it to a simple “human vs machine” competition.
ardj said,
October 4, 2026 @ 2:45 pm
@ Jonathan Smith
Much to agree with.
But regret to say that your link to Arxiv returns "Article identifier '2609.16247”' not recognized".
Author ? Title ?
Jonathan Smith said,
October 5, 2026 @ 3:35 pm
@ardj
Oops extra close quotes, let me try again:
link
or search
The Pain Axis: LLMs Represent Self-Directed Harm and Act on It (v2 title)
The Pain Axis: LLMs Represent Self-Directed Harm and Act to Relieve It (v1 title)
Valen Tagliabue, Leonard Dung, Cameron Berg
Olaf Zimmermann said,
October 8, 2026 @ 2:50 pm
Lest we forget:
Deep Blue needed an awful lot of human help to eventually, narrowly, beat Kasparov.