It's been a long time since I read Hofstader's book "Fluid Concepts and Creative Analogies" that this is presumably based on, so I'll be interested to watch the video as a refresher, later. :) He also wrote a follow-on book, "Surfaces and Essences: Analogy as the Fuel and Fire of Thinking", published after this video was made, that I've never read.
From what I recall, to Hofstafer analogy making isn't some higher level cognitive process, certainly not a language based one, but basically is THE cognitive process all the way from perception on up, and is the mechanism by which we form object categories in the first place.
As always Hofstader's ideas are interesting, but I can't say I agree with them. It seems that the key evolutionary benefit, and function, of a brain is prediction, which is the superpower that moves us from being stuck in the present to being able to "see" (predict) the future, and therefore from being merely reactive to being able to proactively plan and predict future outcomes (what will the sabre-tooth do, where is the water supply?) based on our experience.
Given the never-same-twice nature of sensory perception, before you can predict you need to be able to generalize/categorize, which I think Hofstader would regarded as analogy making (how is this thing I'm seeing similar to what I've previously seen?), although it seems the actual mechanism involved is embeddings or embedding-like representations where similar inputs have similar representations, and what might more simply be considered as associative recall provides the generalization from view/instance to identity/category.
So, is it really analogies all the way up, or are our perception and cognitive processes better regarded as generalization and prediction, which seem not only seem to have direct and obvious neural realizations, but also match the evolutionary needs that we would expect to exist?
> to Hofstafer analogy making isn't some higher level cognitive process, certainly not a language based one, but basically is THE cognitive process
This is so funny to me, because as many people know, sharing an analogy with another person is the fastest way to LOSE an argument with someone, or otherwise spiral it into an unproductive place.
I think it’s Scott Adams who used to say analogies work well for explaining. They work terribly for persuasion.
> sharing an analogy with another person is the fastest way to LOSE an argument with someone
This is true if the arguer is hostile, but as I've gotten older, if I get the sense that someone is entering an argument with the primary goal of "winning", I'll try to avoid that framing or just look for an offramp entirely.
Sure, the other party might think they "won", but they were going to think that anyway. For those more inclined to feel that the point of talking to each other is to learn from each other, I'll continue to use analogies and other things reasonable people understand.
> This is true if the arguer is hostile, but as I've gotten older, if I get the sense that someone is entering an argument with the primary goal of "winning", I'll try to avoid that framing or just look for an offramp entirely.
Absolutely. The best kinds of arguments are those where you both have the shared goal of reaching consensus, and treat reaching consensus as a collaborative activity of finding the correct answer, even though you disagree with the starting point. There are a few techniques for this, which work well when operating in good faith, and can backfire when dealing with hostile counterparties. Most things work badly when arguing with hostile counterparties.
I think that's just because every analogy gives an entire "second front" of ideas for a hostile recipient to find a "flaw", when they ignore the intended boundary between the stuff that does/doesn't matter to the analogy.
Ex:
Explainer: "Getting a spleen means cutting open the patient and taking it out. It's just like how I'm going to unzip this section of the patient-shaped doll, and remove this little purple bean. In both cases a hole is necessary in a similar location."
Hostile listener: "Nonsense! I can just buy beans at the store! So just buy a spleen! No hole!"
Agreed, and while subjective experience can be misleading (being a Hofstadian "strange loop" is a strange thing) you'd think that if our entire perceptual and cognitive processes were analogy based, then this would filter through to our everyday experience. You'd expect us to see analogies, not generalities, everywhere... looking at your car engine and seeing a hologram of your failing football team floating over it, perhaps!
It's notable that when we do use analogies there are really two use cases
1) Trying to use them to communicate an alternate or simpler POV
2) To make a joke, by pointing out a ridiculous comparison
The joke case is interesting, because this seems to be the basis of half of humor - breaking expectations (the other half perhaps being taboo breaking?), which really isn't how you'd want an adaptive cognitive system to work. You need to see the predictive regularities in the world, not see absurd similarities that make you laugh.
Constructing explicit analogies for persuasion seems to be a bit of a different thing.
I’m not sure if Hofstadter puts it this way, but to me even the core aspects of your sentences in this post have roots in analogies. What does it mean to lose an argument or to spiral it to a different place? There is no place, there is no lost item, but we talk about these abstract ideas in ways that largely depend upon understanding things like physical objects and space and movement.
There are certainly a lot of analogies to be found in language - it would be interesting to know how much of this is cultural (e.g. business use of sports analogies) vs universal across cultures. Your "spiraling" example reminds me of one I recently came across as a roman coin collector - the origin of the word cataracts, and also portcullis (latin: cataracta) as being analogical to a waterfall (whose greek name gives us cataract), and the notion of something coming crashing down (even if only an eye covering).
Perhaps the common occurrence of analogies in language is because language is distinct from the things it is referencing, so some degree of abstraction is unavoidable?
Ironically, I think he was right! In fact some of his initial experiments, like copycat, were about predicting patterns. You can see next-token-prediction from there. But I think he always held out for an algorithmic/logical method rather than a purely statistical one.
If he had accepted the "Bitter Lesson", I think he would have been at the forefront of LLMs.
Maybe you haven't noticed that the "Bitter Lesson" had itself a "Bitter Lesson" - that scaling pure data and compute did not lead to AGI: diminishing training returns, GPT-5 disappointment, even openAI stating it was the last 'pure scale' model.
The path forward all big llm providers ("ai" labs) have gone is neuro-symbolic (even though they publicly would never labeled it as such to not admit critics like Gary Marcus were right - even though all their actions actually point in that direction).
I think me just means neural network models RLed to Chain of Thought reasoning? The thinking tokens are the symbolic bit.
Smolensky's latest paper posted here the other day has some thoughts on how modern neural networks might beconsidered neurosymbolic, or rather "gradient symbolic processing," from another perspective entirely.
I wouldn't say the bitter lesson has given out! If you haven't noticed, these things keep getting bigger and bigger.
When my daughter was 6 months old, she once pointed at a towel and said "P". Apparently she recognize Winnie the Pooh from a book she saw earlier.
If you think about it, this requires recognizing something common in things which are very different: e.g. book and towel have different texture and color, and you somehow need to separate what's depicted from the background. So that's pretty much innate, core brain function. I mean, any animal with vision can recognize an object from the background - otherwise vision is useless. But for humans (and some animals) this translates to depiction of object on a flat surface very easily.
So, yeah, analogies-all-the-way-down seems plausible. Even object-vs-background and depiction-vs-paper is itself an analogy.
Well, recognizing (recalling) Pooh on a towel as being the same as a picture of Pooh in book can also just be regarded as generalization, same as recognizing that someone looking left is the same person as them looking right, etc.
Our cortex (all of it - visual, auditory, association. motor..) is remarkably simple and regular - it's six layers of neurons with a specific circuit design, that appears to make it a predictor, with top down predictions meeting bottom up sensory feedback. As noted, it makes a ton of sense that evolution would have equipped us with a predictor (and the more powerful the better). Prediction directly supports reasoning and planning (multi-step what-if prediction).
So, my question for Hofstader, or you, would be given that the brain seems to be literally built for prediction, then what is to be gained by regarding this as analogy making instead? Is there any evidence for this? Even as a philosophical viewpoint, doesn't that make for very convoluted explanations of simple things like reasoning and language? (I explained prediction-based reasoning above in one sentence - the analogy-based explanation takes a whole dense book!).
One could certainly use the language of analogies to describe perception (Pooh's ears on the towel are analogous to his ears in the book, etc), but what is the benefit of doing so, rather than the simpler view of this just being generalization (which is specifically what the animal needs), and where is the evidence in the brain?
Analogy-making points to more specific mechanism: ability to identify features within representations, and modulating associative lookup using those features. I.e. it's not as simple as D(x, y) i.e. distance between embeddings, but something like D(f(x), f(y)) where f projects representation to a specific feature space.
The evidence can be found e.g. in LLM interpretability - people were able to identify features as something concrete. Also in our cognition - we can tell _how_ two things are similar, or find Y similar to Y in context of F.
"Prediction" by itself seems like a black box: if we are trying to predict sensory input, that gives no explanation to our ability to focus on specific details, or explains why some prediction failures are more likely, etc. OTOH if we say that cortex might be 'disassembling' sensory signal into high-level features and is trying to predict those (features themselves might be identified as "something useful for prediction", i.e. something which explains a lot of variance), it's much easier to connect low-level "prediction" to our high-level cognition.
Generalization is necessary as a way to sense the world for a bunch of reasons:
- The never-same-twice nature of sensory perception
- The need to recognize objects from different directions, distances, partially obscured, etc
- Because the nature of the world is that predictive categories are equally or more important than individual identities, especially to basic survival, and for basic organisms
Rather than "generalization" being unspecific, I'd say that as far as perception goes there is a very concrete realization of generalization in our brain - embeddings and associative recall by partial embeddings (see half a face, recall the whole face).
Analogies don't seem to offer any benefit over prediction in terms of what to focus on - there is always a need to separate foreground (object) from background, and to learn what is instance-specific vs generalizable. Is is irrelevant or coincidence that sabre-tooths are near the water hole, or is this part of a predictive pattern (or analogy of some sort?!) that should usefully be learnt?
When we are not sure about what we are looking at, we often then switch from the gestalt "looks familiar" to looking for specific points of confirmation. Is that car in the parking lot really ours? Does it have our air-freshener hanging from the mirror?
Needing to focus on details can also be part of generalization and exception making - when we recognize that two situations that seem similar are in fact different, and we are trying to isolate what the differences are.
I think that the strongest argument for our brains being built for prediction, not analogy making, are that:
1) Prediction is exactly what is needed to survive and thrive - to take advantage of the external world, and not be killed by it. We directly need to predict. We do not need to make analogies, unless analogies can also be used as a way to predict.
2) Our brain appears to be built for prediction, and prediction gives rise to the very obvious learning signal of prediction failure. One of the lessons of LLMs is confirmation of the power of prediction as all you need to drive learning.
The overall way our perception seems to work is indeed as much by prediction as sensory input - in a way prediction really comes first, with sensory inputs narrowing/correcting the prediction (and when you turn it off then we predict/dream without the constraint of sensory reality). The environment you are in already leads you to predict what might be there, and memory of what specifically was there previously will shape those predictions. Sensory feedback corrects/narrows those predictions, and then - if necessary - we look for specific memorized features to confirm what we are seeing/sensing.
Considerations of evolution, and what would be the sensory and cognitive "needs" of organisms of ever increasing complexity, all the way up to ourselves, can also point to what brain capabilities are older and more fundamental, and would have been there to be refined, adapted and built upon later. Even the most simple organism needs to never-same-twice generalize, but will be stuck being reactive until evolution has either equipped it with "reactive predictions" such as the fly being startled by the fast-moving visual input (predictive of being swatted by the horse's tail), or eventually developing a predictive brain like our own, especially useful to an animal that is a generalist living in an environment that is more difficult to predict.
Yes, in the text he explicitly argues for categorization and analogizing being two sides of the same coin and functionally equivalent. For instance - what is an ‘embedding’? In itself its an analogy. Lakoff explored similar ground.
> processes better regarded as generalization and prediction, which seem not only seem to have direct and obvious neural realizations
If the workings of those circuits are obvious to you, I'd really like to learn. Do you mean the level of analysis at https://transformer-circuits.pub/ ? (That looks like good work but not a deep understanding.)
The predictive nature of our cortex seems even more direct that that, although perhaps obvious is overstating it. To read about the 6-layer circuit of our cortex, google for "cortical microcircuit". Basically the repeating structure of our cortex (across all regions) is this vertical connectivity pattern ("circuit") between the 6 layers, and also the connections coming into and out of the cortex at the top and bottom of those layers. There are also horizontal connections within layers of the cortex, but these are short-distance/localized likely for coordinating activity with a local region (cortical column).
Our cortex is connected in a loop with our thalamus (cf recent talk of looped transformers!), which can be separated into thalamic core and matrix. Basically our senses feed into the thalamic core, and cortical outputs (predictions) feed into the thalamic matrix. The thalamic connections/loop seem to act a bit like a patch panel connecting different regions of the cortex.
The cortical microcircuit is a real physical circuit - a pattern of synaptic connections between layers and inputs/outputs. Those Transformer circuits are more functional than physical. We know how a Transformer is physically connected, with attention heads etc, and this "circuit" analysis then describes how it functions, e.g. with attention heads in adjacent transformer layers effectively combining to form "induction heads" that copy data across embeddings.
From what I recall, to Hofstafer analogy making isn't some higher level cognitive process, certainly not a language based one, but basically is THE cognitive process all the way from perception on up, and is the mechanism by which we form object categories in the first place.
As always Hofstader's ideas are interesting, but I can't say I agree with them. It seems that the key evolutionary benefit, and function, of a brain is prediction, which is the superpower that moves us from being stuck in the present to being able to "see" (predict) the future, and therefore from being merely reactive to being able to proactively plan and predict future outcomes (what will the sabre-tooth do, where is the water supply?) based on our experience.
Given the never-same-twice nature of sensory perception, before you can predict you need to be able to generalize/categorize, which I think Hofstader would regarded as analogy making (how is this thing I'm seeing similar to what I've previously seen?), although it seems the actual mechanism involved is embeddings or embedding-like representations where similar inputs have similar representations, and what might more simply be considered as associative recall provides the generalization from view/instance to identity/category.
So, is it really analogies all the way up, or are our perception and cognitive processes better regarded as generalization and prediction, which seem not only seem to have direct and obvious neural realizations, but also match the evolutionary needs that we would expect to exist?