Kant, Borges, & AI go to Bollywood: Connections Nobody Asked For
Intro: I am not entirely sure of the point of this post: a somewhat random associative rumination that brings together a German philosopher, a blind Argentinian author, a lovely though not well known, Bollywood movie, and large language models. All I can say is that these connections have been rattling around in my head, rent free for a while, and I needed to get them out. Sometimes that’s reason enough.
Concepts without percepts are empty,
percepts without concepts are blind.
— Immanuel Kant
I came across this quote a few days ago as I was digging into my Dropbox archives. I don’t even remember what I was looking for, but sort of got sucked into double-clicking on random files that had just been sitting there undisturbed for over 2 decades or more. This quote, as it turned out, was in a paper I had written for my comprehensive examination back when I was a graduate student in UIUC.
Finding this quote was serendipitous because it made concrete something I had been thinking about for the past few weeks—about how LLMs differ from human brains.
Kant’s formulation captures something essential about how we make sense of the world. Raw experience needs frameworks to become meaningful; frameworks need grounding in experience to avoid becoming mere abstractions floating in conceptual space.
If we think about this quote in the context of Large Language Models (LLMs) it becomes clear that LLM’s seem to embody the first half of Kant’s statement quite perfectly. Here are these remarkable systems, weaving together words with impressive fluency, yet entirely divorced from the perceptual world that gives those concepts meaning.
Spinning webs of words that can seem profound until you remember a LLM has never felt rejected by a journal or stubbed a toe or laughed out loud at a meme on their phone while ignoring a beautiful Arizona sunset, while walking their dog.
LLMs are, in a sense, pure concept without percept.
The question then becomes, is the opposite possible? What would the other end of the continuum look like?
In considering this I was reminded of a short story I had read back in high school by the blind Argentinian author Jorge Luis Borges, in his book Labyrinths, titled Funes the Memorius. This eponymous story was about a young man, Ireno Funes. As the story goes, Funes suffers a horse-riding accident and hurts his head, as a consequence of which he develops the ability to recall every detail of every moment he has ever experienced. While this might seem like a gift, Funes becomes paralyzed by the overwhelming flood of sensory details and memories. He is unable to abstract or generalize because every moment is so specific, so rich and detailed, and so special. Every instance is completely unique and different from all others. This makes him incapable of generalities, of abstraction; his world is one of intolerably uncountable details. As Borges describes it:
He was, let us not forget, almost incapable of general, platonic ideas. It was not only difficult for him to understand that the generic term dog embraced so many unlike specimens of differing sizes and different forms; he was disturbed by the fact that a dog at three-fourteen (seen in profile) should have the same name as the dog at three-fifteen (seen from the front). His own face in the mirror, his own hands, surprised him on every occasion.
The tragedy of Funes, as Borges describes it, suggests that forgetting and abstraction are critical for human thought and functioning. It argues that it is our ability to form general concepts rather than being trapped by infinite particulars is essential to intelligence and sanity.
Funes was the opposite of an LLM, all percept, no concept.
As it happens, this whole chain of thought was sparked by a recent conversation on creativity and learning that I had with Raaghav Pandya. In our conversation, we were discussing the need to re-learn how to see, to try to perceive rather than simply recognize, to encounter things freshly rather than categorizing them into familiar patterns. It was in this context that he mentioned a relatively unknown Bollywood film, Ankhon Dekhi (which translates as With My Own Eyes or Witnessed Firsthand), hardly expecting me to have heard of it. As it turned out, not only did I know of the movie, but I was also a huge fan. This honestly felt like a small miracle given how under-the-radar this movie is.
This made me go back and rewatch the film on YouTube once we had gotten off the call. And it was with this experience in mind that I stumbled across that old Kant quote from my graduate school paper, that I realized how perfectly Ankhon Dekhi illustrates the very spectrum Kant was describing.
Briefly, in this movie, the main character called Bauji, in an amazing performance by the veteran character actor Sanjay Mishra, decides to believe only what he directly experiences, rejecting the inherited concepts and social abstractions that so often mislead us. (This idea of “witnessing firsthand” explains the title of the movie: Ankhon Dekhi.) Though a futile exercise, it does free him in unanticipated ways (even while making him a minor local celebrity and also driving his family crazy in the process). There’s something beautiful about his refusal to let received wisdom substitute for lived truth, his insistence on grounding knowledge in direct experience. For instance, he refuses to accept the existence of a place called Amsterdam, since he has never been there, or the fact that lions roar, since he had never actually heard a lion roar. The movie is funny, touching and subtly philosophical, even while remaining grounded in the craziness of life in Delhi.
Bauji finds a kind of freedom in his approach, breaking away from the stereotyping that emerges when concepts become unmoored from perception. But the movie also points to the problems that arise if we are to take a stance like Bauji does and the movie speaks to the breakdowns and failures that occur. There is a recognition of the fact that we cannot live purely in the immediate, that some form of abstraction and pattern-making is necessary for navigating this world, for making meaning, and for connecting with others.
In both these examples, Bauji (by choice) and Funes (due to matters out of his control) demonstrate the fundamental limitations of the percept-only approach.
But Ankhon Dekhi has other Kantian surprises in store, a little detail I had forgotten, but one that Raghaav reminded me of. What is truly funny, Raghaav pointed out, is that Ankhon Dekhi has another character, who is in some ways the opposite of Bauji. This character, who makes a short, brief appearance, seems to suffer from some ailment that makes him talk incessantly. Essentially this character speaks non-stop, just jumping from topic to topic, making one random connection after another in an endless chain of associations, unmoored from reality. It hit us both that it was almost like watching an LLM in action, that same quality of ceaseless linkage without the anchor of embodied experience to give it weight and direction. )You can watch a clip of that scene here).
In some powerful way, the director Rajat Kapoor had captured, something essential about the extremes of the Kantian quote in his movie, and had created a human LLM in the process.
Outro: It occurs to me that this whole reflection might be a bit like that bit character from the film – one association leading to another, from Kant to LLMs to Borges to conversations about movies, each connection sparking the next. Though perhaps with a bit more restraint, and hopefully with at least some anchor in the embodied experiences that give these ideas their weight. But maybe there’s something to be learned through this very process, these somewhat random connections forming a web of meaning that has value in and of itself. A conversation with Raaghav about a movie we both love, which somehow brings back Borges and his impossible Funes, which then bumps into those old graduate school notes gathering digital dust in Dropbox. Maybe that’s what minds are, vast archives where ideas sit undisturbed until something makes you double-click, and suddenly they’re part of a new conversation.