I continue to think about the tendency of LLMs to confabulate, that is, to make stuff up that is simply not true of the world. As I have remarked here and there, I tend to think that 1) confabulation is inherent in the architecture, and 2) that this “confabulation” is the default mode of human language. We just make things up.
However, we must live with one another and that requires cooperation. Effective communication requires agreement. It turns out that the external world is a convenient locus for that agreement. We agree that THAT tree over is a pine, that THAT apple is ripe, that THAT bird is a cardinal, that the stew is too salty, the earth is round and that the moon travels around the earth every 28 days. Some of these agreements may come easily, others are more difficult in the making.evolu
However true that may be, it does seem a bit odd to think of external reality as a vehicle for grounding agreement on language use. And, if I thought about it a bit, I could probably come up with some account of why that doesn’t seem quite right. But I’m stalking a different beast at the moment.
Consider this observation that Weston La Barre made in 1972 in The Ghost Dance: The Origins of Religion (p. 60):
... the Australian Bushman themselves equate dream-time with the myth-time that is mysteriously brought back in ritual; myth is as timeless as the unconscious mind. It is the delectability of dreams that makes them desirable, and it is their desirability that (along with lowered critical threshold) that gives them their intense “reality” and conviction. The fact that he dreams first force on man the need to epistemologize.
Let’s pretend we’ve just gotten up from a dream. What do we make of it? Or perhaps, how do we make anything of it?
We tell it to others, and they tell us their dreams. We talk and arrive at some agreement about what transpires in those dreams, about the nature of those events and their relationship to the waking world. I am thus reminded of Bob Dylan’s line in “Talkin’ World War III Blues”: “I’ll let you be in my dreams of I can be in yours.”
Is that what expressive culture is about, religious myth, ritual, folktales, drama, epics, novels, not to mention the visual and plastic arts as well? Are they vehicles for allowing us to reach agreement about certain of our, shall we say, confabulations?
This brings up a post from February, What is it about depicting sex in writing? From shared knowledge to common knowledge. Using three novels as examples (Pride and Prejudice, Tess of the D’Urbervilles, and Lady Chatterly’s Lover) I suggested that as we move from one to the other in historical order, we’re moving from shared knowledge of sexual matters to common knowledge. Those are terms from game theory. We talk of shared knowledge when everyone knows something, but aren’t aware that the others know that same thing. Once everyone comes to know that everyone knows that thing, whatever it is, that knowledge has become common knowledge.
I have thus begun to wonder: Is that how the arts function over time? I recently discussed the matter with Claude 3.7. Here’s its response:
It's not just that artistic styles change, but that art gradually makes explicit what was once implicit across many domains of human experience.
This pattern extends beyond sexuality. Consider how literature has treated:
- Mental illness (from metaphorical to explicit)
- Social class tensions (from coded to direct)
- Religious doubt (from allegorical to overt)
- Political criticism (from veiled to unmistakable)
In each case, what begins as shared knowledge eventually becomes common knowledge through artistic expression. This does suggest a kind of progress - not necessarily aesthetic progress, but progress in cultural honesty and transparency.
It is by no means obvious that this is the case. But I’m certainly sympathetic.
What is undeniable is that in the last 15 years or so digital humanists are found empirical evidence of unidirectional trends measured attributed of literary texts over long periods of time. Perhaps the most interesting example is in Matthew Jocker’s Macroanalysis, where he shows a unidirectional trend in a corpus of 3000 Anglophone novels from the 19th century. I discuss this in a number of posts. This working paper might be the place to start: On the direction of literary history: How should we interpret that 3300 node graph in Macroanalysis? There’s another working paper: On the Direction of 19th Century Poetic Style, Underwood and Sellers 2015. You might also look at this blog post from 2016, From Telling to Showing, by the Numbers, which is also about 19th century novels.
More later.
