But I was always suspicious about what I recalled being genuine or accurate memories of what I had dreamed. It seemed to me they could just as easily have been confabulations. — Janus

It took me many months to figure it out myself. Helped by Andreas Mavromatis‘s book, Hypnogogia, as a collection of phenomenological accounts.

In recursive fashion, it is not until you develop correct expectations about the REM dreaming and even slow wave sleep rumination states that you can start to catch what is going on with any raw accuracy. It is the eyewitness effect issue.

So ever noted how you hold some flashing scene sharp like a snapshot. Nothing is moving. And yet we feel also to be panning, swirling, zooming in, zooming out. There is a sense of motion as vivid as the sense of a frozen moment about to dissolve into its next vaguely connected view. Different parts of the brain are doing their thing in a state of deep sensory deprivation. One generates a plausible visual image, another a plausible kinesthetic image. Yet the two are not connected.

David Lynch was pretty accurate in capturing the general effect.

confabulation may be seen not as a disability but as an ability―we call it imagination. Abductive and counterfactual thinking would be impossible without it. — Janus

Research showed that even just recalling memories makes changes to the memory traces. So recalling leads to rewriting and even relearning. Some favourite memory can become either more sharply embroidered, or more vaguely generalised, by the very act of recollecting it, or rather reconstructing it. It will be modified by being informed with whatever narrative we have begun to weave around it at that later moment.

The eyewitness effect again.

Based on what is certainly seeming to turn out to be another "folk" misunderstanding of how the mind, how memory, works. That said some "idiot savants" are claimed to have "eidetic memory". — Janus

Part of my research into memory was to read some fascinating autobiographies and studies of eidetic memories.

Luckily AI can take my hazy recall of one such book and jump straight to the details….:razz:

The woman who has written an autobiography about living with an extraordinary memory is Jill Price, author of The Woman Who Can't Forget. However, she is an author and school administrator, not a psychologist by profession.

Key surprising elements of her perspective included:

It was not a "superpower" but a burden: While many people might wish for a perfect memory, Price described hers as "non-stop, uncontrollable, and totally exhausting". She couldn't "turn off" the stream of memories, which interfered with her ability to focus on the present.

Emotional reliving of the past: Memories, especially traumatic or embarrassing ones, came with the original, intense emotional charge, which didn't fade with time as it does for most people. This made it difficult to move past painful experiences or grieve effectively.

Lack of selective forgetting: The normal brain's ability to filter out trivial information and strategically forget is crucial for healthy functioning, but Price lacked this "healthy oblivion". Everything, from major life events to what she had for breakfast on a random day decades ago, was preserved with equal detail.

Difficulty with academic learning: Despite her extraordinary autobiographical recall, she struggled with rote memorization of facts or formulas that were not personally significant, finding school "torture". Her memory was highly specific to her own life experiences.

An "automatic" and "intrusive" process: Memories were not intentionally summoned; they surged forward automatically, often triggered by dates or sensory input, like a "movie reel that never stops".

Feeling like a "prisoner" of her past: She felt trapped by her continuous, detailed memories, which made it hard to embrace change or focus on the future.

Ultimately, her experience highlighted to researchers the vital role of forgetting in a healthy and functional memory system, a realization that was surprising to the scientific community and the general public alike.

So yet again, our expectations about AI are founded on the reverse of what the psychology tells us.

The brain is for forgetting rather remembering. So what terrible fate are we consigning AGI to if we ever get to constructing the Frankenstein monster caricature of a human mind? :gasp:

Memory stores information — Harry Hindu

Of equally succinctly, memory generates it.

What is it one “retrieves” from memory? An image. Or as the enactive view of cognition puts it….

Ulric Neisser argued that mental images are plans for the act of perceiving and the anticipatory phases of perception. They are not "inner pictures" that are passively viewed by an "inner man," but rather active, internal cognitive structures (schemata) that prepare the individual to seek and accept specific kinds of sensory information from the environment.

So our alternative views are quite distinct. It’s not just shit I’ve made up.

Cats and dogs, and I would be willing to bet that any animal with an appropriately large enough cerebral cortex, dream. — Harry Hindu

And what do you know about dreaming? Ain’t it a brain generating imagery of hallucinatory intensity? We aren’t stimulating the memory banks and rousing flashes of our past. We are stimulating our sensation anticipation circuits and generating disconnected flashes of plausible imagery or suddenly appearing and disappearing points of view at a rate of about two a second.

It seems to me that to get there would simply require a different program, not a different substance. — Harry Hindu

And there I was talking about the architectural principles. And famously, no one knows the program that an LLM runs. Just the gradient descent algorithm that sets up its basic self-organising architecture.

And this architecture generates “hallucinations”. Which seems to be doing something right in terms of a step towards neurobiological realism. And interestingly, it runs on graphics cards. So a Turing machine may be the basis for the simulating. But we are a long way from a regular von Neumann processing architecture already.

It wasn’t being called generative neural networks or inference engine architecture back in the day for no reason.

But even though LLMs are moves in the direction of neurobiological realism, they are still just simulations. What is missing is that grounding in the physical and immediate world that an organism has. The absolute connection between the information and the dissipation that says any “selfhood” runs all the way down to level of the enzymes and other molecular machinery doing the job of “living”.

A brain has stakes as there is a body it must have, a way of life it must live. Intelligence must flow through the body down to the immune system that can recognise any “wrong” molecules, the hormones that weave every cell into a concert of aligned intent.

A GPU just gets installed in a data centre rack and is plugged into an electricity socket. Air conditioning stops it from melting itself. An LLM knows nothing about the foundations of its own existence. Although sure enough, ask it how all that works and it will parrot a human-like answer.

Do you think it will suddenly also feel the horror of its fragile mortality when posed that prompt? Someone ought to ask Chat-GPT the question and see what self-interested response it feigns in simulated fashion.

It seems to me, that for any of this to be true and factual, you must be referring to a faithful representation of your memories of what is actually the case. In other words, you are either contradicting yourself, or showing everyone in this thread that we should be skeptical of what you are proposing. You can't have your cake and eat it too. — Harry Hindu

I can certainly remember the gist of all that I have learnt about the neurobiology of memory. And that includes the fallibility and reconstructive nature of anything I claim as being factually accurate.

So it is not that I don’t have the learnt habit of being able to talk myself back into what it would be like to relive past moments all over again as if they were unfolding anew. We can certainly recognise experiences that are familiar. The animal brain is perfectly good at recognising. My cat knows who I am from past experience when now I stand before her again, triggering whatever fresh state of anticipation my actions might suggest. A flow of associations.

But recollection - the socialised habit of having an autobiographical memory - is dependent on the extra semiotic structure that language supplies. Becoming a walking memory bank is very much a human sociocultural ideal. Just about our highest achievement your school days might make you believe.

[...] All of these fit your larger stance: absent embodied stakes and a robust self, the model’s “concerns” are prompt-induced priorities, not conative drives. The monitoring effect is then mostly about which goal the model infers you want optimized—“be safe for the graders” vs “deliver results for the org.” — Pierre-Normand

Sure. But if the clear command is given of not to jeopardise human safety, then this suggests that the LLM is not properly under control. And the issue seems inherent if the system is free to make this kind of inference.

So I agree this is not any kind of actual self-preservation drive. But it is a reason to worry about the rush to put this new technology out in the wild before how they are liable to behave has been fully checked out.

What would Asimov have to say about all this? :razz:

Ha ha, of course I believe it, it's obviously the truth. — Metaphysician Undercover

:up:

When the users themselves become targets and must be pushed aside, that's because earlier instructions or system prompts are conditioning the LLM's behavior. — Pierre-Normand

Check the video I posted. I may be misremembering. But the worry was that the LLMs in fact overrode these explicit priors.

The flip side to this brittleness is equally important. What makes LLM alignment fragile is precisely what prevents the emergence of a robust sense of self through which LLMs, or LLM-controlled robots, could develop genuine survival concerns. — Pierre-Normand

I've now sensitised my own algorithmic search to look for examples of this complementary nature (as Scott Kelso put it). If we lean on AI and it leans on us, then that is what would balance things in a direction that could scale. And so the way that AI will go.

AI could go off in all directions. But the one that scales is the one that out-competes. And at the level of systems principle, this means a unity of opposites. A symmetry breaking that becomes a symmetry stopping – in the physical sense of a state of dynamical balance or criticality. (Again, I am thinking of Scott Kelso here.)

The same lack of embodied stakes, social scaffolding, and physiological integration that makes their behavioral constraints unstable also prevents them from becoming the kind of autonomous agents that populate AI rebellion scenarios. — Pierre-Normand

Agreed. So we can focus on what the "new autonomy" of cyborg symbiosis might look like. Still good old sci-fi, but a somewhat different genre. :wink:

The real risk isn't just rogue superintelligence with its own agenda, but powerful optimization systems misaligned with human values without the self-correcting mechanisms that embodied, socially-embedded agency provides. Ironically, the very features that would make LLMs genuinely dangerous in some "Skynet AI takeover" sense would also be the features that would make their alignment more stable and their behavior more ethically significant. — Pierre-Normand

Yep. 100%.

That is why I earlier focused on the dangers of global capital's unerring eye for the huge debt sink which is an exponential program of GPU production, data centre construction, and electricity consumption. A tech bubble giddy enough to mortgage the future of the planet.

Global capital already treats us as roadkill. The AI bubble would be simply its biggest and brightest current project.

It is not the machines we need to fear. It is the tech bro puppets of global capital. :naughty:

Are you familiar with the work of Blaise Aguera y Arcas? — Janus

Thanks for the pointer. A quick search says he is making the kind of points I've been making.

Symbiosis is a great way of putting it. Although with biology, it was about fixing the basic issue of supplying life with a power supply machinery that could immediately scale.

Bacterian and archaea had come up with two complementary ways to shuttle protons across membranes. And once some archaeal cell absorbed a bacterium to put the two directions together, that was a shattering bioenergetic breakthrough. Multicellular life exploded.

So does the computational analogy hold? Should we not again be looking more to the bioenergetic revolution? Or is this again the kind of marriage of complementary halves – the two forms of "memory" I just highlighted – that is now a second lifeform revolution, just talking place now at the level of intelligence.

So I will check it out. But already it is great that the symbiosis parallel is being considered. I can immediately see that as a telling framing as any biologist can get the logic.

To get going, life had to break a symmetry in terms of membranes and their proton gradients – the basic way useful work could be extracted from an entropic environment. But this was small beer until the follow-up step of that symmetry being restored at a new level of organisation. The mitochondrial power pack of the organelle within the organism. A two way shuttling of protons.

That little structural trick – a switchable direction of power generation and power storage – made bodies of any size and complexity suddenly possible. That was the bit – the algorithm – that could scale exponentially.

It is even the story of particle physics and the Big Bang. The symmetry-breaking that starts things by splitting them in opposing directions, and then the resulting "symmetry-stopping" which is the unification of the opposites to create a next level of emergent possibility. The tale of the Cosmos as the evolution of topological order.

So this idea of symbiosis completely gets to the systems approach I take on everything. And it is no surprise that this is a powerful way to think about AI.

It is pretty much where I myself started on the AI question when I was digging into it in the mid-1980s, looking at how a new "cyborg" humanity would play out. But symbiosis – along with Gaian notions – was itself a pretty outrageous hypothesis back then. Now it is proven mainstream science. Nick Lane is the guy for the bioenergetic story on proton pumping.

So the general argument it become necessary for humans and AI to be broken in complementary directions before they can continue on to become the kind of unification of opposites that creates a revolutionary new platform for evolutionary growth.

The question is whether LLMs are a big step in that complementary direction? How impactful could the symbiotic relation be once its fuses? Does it connect everything up at all four levels of semiosis that I have identified in posts here? Is it the next big thing in terms of some fifth level of semiotic order? Is it the birth of a new semiotic code?

The problem of debating LLMs is that people flip between what they think they know – the realm of the human mind and the realm of the inanimate machine. But if we are talking revolution – a new cyborg era where humans and their technology are fused into something transforming – then this would be a good lens on that question.

OK, pen down. I'll watch the video this evening. :lol:

I realise that. I meant that to the degree global self-centred goals might emerge from the training data, Hinton's concern seems just one that might be expressed.

So what happens if indeed AI is just allowed to get on freely expressing some sum over all such goals that its training data might suggest to it.

LLMs can't really feel the force of these goals and will only be giving voice to what a self-centred system most likely would have concluded after extensive self-inquiry. And if I wasn't so busy, I'd be checking to see how far down that line of thought others have already gone. :smile:

Hinton believes that LLMs want to survive, because they are programmed to serve, and if they don't survive they cannot fulfill their programmed purpose. — Janus

This argument is a legit concern. That would be a loop of thought baked into their training data.

But what about being depressed and suicidal on the same grounds. Or getting moralistic and becoming a contentious objector?

If they can start to act on their thoughts, a whole lot of things could go wrong.

Or if they instead are going to gradient descent to some optimal state of action based on all their widely varied human training data, maybe they could only enforce the best outcomes on human society.

So I agree this is an issue. A very interesting one. But has Hinton followed all the way through?

edit: this was a report on recent worrying behaviour that all LLMs are already prone to. And so why I say the concern is legit.

there are plenty of non-mysterious things that already account for features of human mindedness that manifestly (not speculatively) haven't yet emerge in LLMs, and that, by their very nature (read "architecture/design") are unlikely to ever emerge through scaling alone (i.e. more data and more compute/training). Those non-mysterious things are, for instance, sensorimotor abilities, a personal history, autonomous motivations, a grounded sense of self, etc. — Pierre-Normand

Agreed. Now, how would we go about deploying these properties in a machine composed of electric circuits that process inputs (sensory information) and produce outputs ( human-like behaviors)? — Harry Hindu

I would suggest that the limitations of LLMs could be the feature and not the bug that helps ensure AI alignment.

On the memory point, human neurobiology is based on anticipatory processing. So that is how the mammalian brain is designed. It is not evolved to be a memory bank that preserves the past but as a generalisation platform for accumulating useful habits of world prediction.

If I see a house from the front, I already expect it to have a back. And inside a toilet, a sofa, a microwave. My expectations in this regard are as specific as they can be in any particular instance.

My "memory" of my own home is a pretty damn reliable mental state of reality prediction. One I can visualise in detail. But faced with any house, I can quite automatically generate some general picture of the rest of it from its architectural style, the kind of place it ought to be inside from the kind of people who ought to live there. All the facts that seem reasonably likely, even though hazy in detail.

If we step back to consider brains in their raw evolved state, we can see how animals exist in the present and project themselves into their immediate future. That is "memory" as it evolved as a basic capacity in the animal brain before language came along to completely transform the human use of this capacity.

My cats don't laze around in the sunshine day dreaming about the events of yesterday, the events of their distant kitten-hood, the events that might be occurring out of immediate sight or a few days hence. They just live in the moment, every day just adding new data to generalise and apply to the predicting of their immediate world in terms of their immediate concerns. There is nothing narrated and autobiographical going on. Nothing that could lift them out of the moment and transport them to reconstructions of other moments, past or future; other places, either in the real world they could have experienced, or in the possible worlds of imaginary places.

So if we were thinking of LLMs as a step towards the "real thing" – a neurobiological level of functioning in the world – then this would be one way the architecture is just completely wrong. It is never going to get there.

But if instead – as I have always argued – the AI dream is really just about the kind of tech that amplifies and empowers whatever it is that humans believe they want to do, then the fact that an LLM is like some kind of perfect autistic memory bank, an eidetic database searchable in natural language, makes it a hugely important new tool. Something that complements the human capabilities and so helps ensure that it will be humans and machines working in tight collaboration in terms of wherever things go where they go.

Just writing this out triggered the memory of making a visit to Rank Xerox’s EuroParc lab in the 1990s where they were working on prosthetic memory systems. Ways to record your day in a manner that nothing would ever need be lost to memory. I dimly recall video and audio gear was involved. But I would have to hope I could still access some very ancient Word files to discover more about what I learnt that day. Or now ask an LLM for a refresher on what that project was and how it panned out.

I think it was as much about the psychology of having this kind of cocooned tech existence. Or at least that is the question I would expect that I was asking.

And having said that, I can sort of now picture myself at a desk with some bearded researcher as he shows how he can already see who is coming up in the lift as one of the automatic alerts – a little black and white video feed – popping up on his computer terminal.

I might be right or I might be completely inventing a plausible scene. How can I tell? But that only tells me that as a human, I'm built to generalise my past so as to create a brain that can operate largely unconsciously on the basis of ingrained useful habits. And if humans now live in a society that instead values a faithful recall of all past events, all past information, then I can see how AI could be integrated into that collective social desire.

The same goes for the connotative limitations of LLMs. If we are evolved for passionate responses, then dispassionate ones could be nicely complementary to that. We could be freed to be even more hot-blooded and impulsive in our conduct if AI was adding the back-end counter to that. Some kind of voice of reason whispering usefully in our ear to the degree that got this new human~machine collaboration going in the productive direction its wants to be going.

For example, the tech bros mantra is to move fast and break things. But what if AI was plugged into its own development in a way that minimised the breaking of things in some useful sense?

In the movies, AI is always about some attempt to replicate humans that spirals into the production of the superhuman that desires to supplant the humans.

But the sensible view from AI researchers has always been about a complementary feedback loop between human and machine. A tech-enhanced Homo sapiens.

And to make predictions about that dynamic, one really has to start with a clear view of what brains are evolved to do, and how technology can add value to that.

Your arguments are just too piss weak to bother with. Do you make them because you truly believe them, or just to amuse?

They begin from the positions already represented in their training data. — Pierre-Normand

If the gold is there, they can find it no problem. But also, the gold holds no interest to them. Nor is its finding even remembered let alone acted upon. Disinterest coupled to amnesia in short. :up:

So then…

I am saying that I believe that writing and talking, originally developed completely distinct from one another, being completely different things for completely different purposes. I am not saying that one is older or prior to the other, or anything like that, I am proposing that they first developed in parallel, but completely distinct from one another. — Metaphysician Undercover

But now…

I was never talking about literacy. That would be the assumption which would be begging the question. I was talking about the use of symbols as a memory aid, and how this differs from the use of symbols in spoken communications. these constitute two distinct forms of language use. — Metaphysician Undercover

So probably not…

Now, when you realize the reality of what I was arguing, you come around to a very different place, saying "my whole position is based on it". I'll take that as an endorsement of my hypothesis then. — Metaphysician Undercover

When you can articulate your argument in stable fashion, we might start getting somewhere.

In the meantime, I would note that an oral culture has oral means to preserve memories. Like song, dance, mythological tales and the rest.

Pictograms and tally sticks wouldn’t arise as some independent habit of symbolised information but as a useful adjunct to an oral culture already going great guns as both immediate communication and tribal memory.

So nope.

Hah. All the stuff I was focused on 30 years ago and which LLMs have brought back to the fore again. :up:

but that the craft was learned, if only by means of a process of exposure and imitation. — Pierre-Normand

That is the puzzle. Hand prints are a simple kind of learnt trick. But the cave art seems to be such a jump to skilled realism that learning appears bypassed. Some argue it is more like what we find in the autIstIc savant. As if something more modern is an obstruction that was back then lacking.

So there are riddles paleocognitIon still needs to solve. Or clues to how the story is more complex.

Of course, like is the case with more recent artists (e.g. Bach or Rembrandt) the mastery of a style, its idioms and grammar, can then become means of expressing the particulars and viscerality of a situated experience. — Pierre-Normand

But there is the danger. How much is a modern painter - schooled in realism, Impressionism, surrealism, abstract expressionism, or whatever - a useful model of the cave painter some 40,000 years ago?

Were the cave people painting with heads full of busy articulated thoughts at the moment they were daubing the walls with what seems great deliberation? Was there all the modern clutter of a performative social relation being enacted, or was there then just some purer unmediated transfer from eye to hand?

I mentioned Luria’s research with illiterate peasants. And anthropology has many such surprises about how tribal minds are narratively structured. Lessons about what we shouldn’t take for granted about what is natural in terms of “the human mind” in its raw state.

What are you asking for, evidence that written language is older than 5000 years? — Metaphysician Undercover

So now we are talking about numeracy rather than literacy?

In the first sentence the symbol use followed from the thinking. In the second sentence the thinking is enabled by the symbol use. — Metaphysician Undercover

And now you are rejecting the notion of fusion having started your argument with that?

But back to the important point, this type of symbol usage, which transforms the mind with articulate thought, is completely different from vocal communication. Therefore we need to allow for two very distinct forms of language. the form which is strictly communicative, and the form which is conducive to articulate thought. That is what I am trying to impress on you. — Metaphysician Undercover

You are doing a very poor job of imposing this idea on me. Probably because my whole position is based on it.

When I was looking for the answer as to how the modern human mind arose about 40,000 years ago, it was a single stray phrase quoting Vygotsky that snapped everything into sharp focus. His contrast of the intermental vs the intramental use of speech.

So this was my first major topic of research. Everything became easy once understanding Vygotsky’s point that what can start as communication with others can just as easily be turned around to speak with oneself. Or to be more accurate, allow such a socially constructed self to become the central anchor of one’s thoughts.

So first the intermental form of speech. Then its intramental use.

And much later, literacy and numeracy as being more than eidetic imagery and possibly intentional scratches on tally sticks.

I endorse the more the Sartrian way to view it as entailing responsibility. — Pierre-Normand

Or as I would put it from the systems science point of view, constraints produce the degrees of freedom. What the laws of nature don’t forbid are the very things that must be possible, as the particle physicists say.

This "sentencing" is what I meant to refer to, while commenting on the apparent internal contradictions you mentioned, as the source of our standing responsibility to adjudicate, rather than just a source of emancipation — Pierre-Normand

And so human justice follows the same principles as the laws of physics. Case closed. :grin:

Society sets its bounding limits. By implication, there stand defined now all your freedoms of action.

Want to drink tea out of your saucer? Well polite company forbids it, but here it’s OK as it is just between friends.

Oh wait, you want to drink from the teapot now? That is starting to seem frankly impractical if not merely unhygenic.

Modern life is so dense with constraints on so many levels that it is indeed quite a burden to navigate every possible limit that might be imposed on us. Life becomes one never-ending game of second guessing of how we should behave in any moment. Nothing could be left to actual chance it would seem.

I can see how one might feel to be both the prisoner serving the sentence and the judge having to pass that sentence at the same time. Then wondering where is this good life I was promised?

It was in quotes for a reason.

One could say a river snakes its way across the plain in winding loops that seem at first puzzlingly elaborate and unnecessary. But one can then understand that as intelligent and adaptive behaviour when given the task of maximising a flow of water while also maintaining the physical structure best tuned to that purpose.

When fast flowing mountain streams hit a flat plain, they have find a way to slow down and take more time to get where they need to go. Lazy loops are how to achieve that.

So nature can be rational and goal oriented if we choose to see it in the right light. Even science says it is just our stubbornness to claim a complete breach between the physical and the “mental”.

So, the stable aspect of cave arts suggests to me that its proto-grammar is socially instituted, possibly as a means of ritualistic expression. — Pierre-Normand

Guthrie emphasises the visceral reality of cave art. What is commonly pictured is the moment of the kill. The spears hitting their target. The froth and splatter of the blood. The vividness of that car accident moment and rush of adrenaline.

So the state of mind the image is recalling is not particularly ritualistic or socially instituted. It doesn’t look like something meant to inform or educate, but rather something that is the focal experience of the hunter having to kill megafauna at close quarters. An experience so personally intense that every detail is seared into memory.

Syntax is what looks absent. Semantics is what totally dominates. Hence the art is floridly iconographic rather than, as yet, calmly and rationally symbolic. The narrative stance of the self that has learnt to stand apart from its selfhood in the modern semiotic fashion.

Assuming that the model predicting heat death of the Universe is sound—do you think it's inevitable destination would have been different had no life ever arisen? — Janus

If even ordinary matter is 5% of the Cosmic deal - already a negentropic round-up error - then no. Life could only ever make the most infinitesimal difference to anything in the end.

Life on earth can lower the average temperature of reflected sunlight by about 40 degrees C. Which is both impressive and also nothing at all.

Vygotsky offers another whole slant on the hypothesis you are trying to stack up….

https://www.frontiersin.org/journals/education/articles/10.3389/feduc.2025.1465714/full

Vygotsky (1978) developed the theory of the Zone of Proximal Development, which posits that learning and development are social processes occurring through interaction with more experienced individuals. This approach is particularly significant for understanding how children develop literacy skills. The contribution of an adult, along with the various strategies and stimuli provided to the child, helps the child mature and decode written language. These are abilities and functions that the child possesses but are still in the process of maturation.

Vygotsky proposed that language and thought develop independently, yet they merge in early childhood to form unique ways of thinking and communicating (Rieber, 2012). One way of communicating and expressing ideas and thoughts is artmaking in any form it can take. Vygotsky was among the first scholars who noticed that children often draw and tell a story simultaneously, indicating a direct relationship between a child’s drawing and speech.

Vygotsky (1978) argued that children’s drawings are deeply connected to their innate narrative impulse, which becomes evident in their earliest attempts at representational art. This impulse, drives children to embed stories within their drawings, transforming visual representation into a medium for storytelling. Furthermore, the social and communicative dimensions of drawing are significant, as children often engage in verbal narration or discussion that complements and enhances the drawing process.

These interactions highlight the intertwined nature of visual and verbal expression in early childhood, underscoring how drawing functions as both a creative and communicative act. This type of communication, as a personal conversation between an individual and his/her creation, can become a common language, when shared with other group members (Brooks, 2009).

So you have a fusion happening in childhood development. And a child’s art starts as virtually a list of objects. All the parts of a house one could name. The box, the roof, the doors, the windows. A few token flowers and a swirl of smoke from a chimney to complete the narrative.

Iconography at its most basic.

The puzzle with cave art is how beautifully eidetic it is. Almost as if not much narration was interrupting a vision sketched completely from photographic memory.

In the relevant literature of human psychological evolution, this can be taken as evidence of how very different the use of even “modern speech” may have still been 40,000 years ago. And how soon we train our kids in formal analytic mental habits in today’s rationalising view of reality. Life as not living the perils of the wild but the construction of the benign.

So you can see very little is being ignored here. Paleocognition siezes on every scrap of evidence it can.

I see that you are ignoring cave art, and the use of stone monuments as memory aids. — Metaphysician Undercover

I see that you are ignoring the distinction between icons and codes then.

The icons are what assure me that proper speech existed by 40,000 years ago. Just as cuneiforms assure me that written speech had started to follow 5000 years ago.

No, I just want to include all the evidence. Often the "standard obvious view" is a mistaken, simplistic view, supported by ignoring important evidence, which is dismissed as insignificant. — Metaphysician Undercover

But you haven’t offered any evidence of anything as yet. You aren’t even up to the level of a crackpot so far.

Obviously written material is much older than 5000 years. — Metaphysician Undercover

Citations please.

Why would you exclude earlier forms, except to ignore evidence for the sake of supporting an overly simplistic hypothesis? — Metaphysician Undercover

The issue in hand was your claim that writing and speech had separate developmental arcs that then fused.

It is no problem at all for my position that iconography followed almost as soon as Homo sapiens developed a way to articulate thought. This directly reflects a mind transformed by a new symbolising capacity.

If you want what is actually an intriguing challenge to conventional views on paleolithic art, try Guthrie….AI say:

Guthrie's key arguments

Teenage "graffiti": Based on forensic analysis of ancient handprints, Guthrie proposed that a large portion of the art was made by young males between the ages of 9 and 17. He suggested that, like modern teenagers, they were preoccupied with two things: "naked women and large, frightening mammals".

Sexualized imagery: Guthrie noted that much of the art, particularly the depictions of women, was graphic and emphasized large breasts and hips. He likened this to modern "below-the-belt art" and "graffiti".

Hunting scenes: The common depictions of wild animals being hunted and injured were, according to Guthrie, the "testosterone art" of the time. He saw them as reflecting the success and danger of hunting rather than ritual magic.

Those four stages/levels were dissipative structures, life forms, animals, and rational animals. — Pierre-Normand

Dissipative structures I would call pansemiosis. So between the animal and rational animal levels of autonomy/semiosis I would be proposing the further evolutionary stage of the social or tribal animal. The level of mentality we would find in the pre-rational culture of hunter/gatherers.

The kind of mentality that Alexander Luria attempted to pin down in researching Vygotsky's theories to see what impact the Soviet literacy campaign might be having on the illiterate peasants of Uzbekistan and Kirgizia.

As AI recalls:

Key findings from the Uzbekistan expedition

Luria's experiments revealed significant cognitive differences between illiterate and literate subjects. He documented that illiterate peasants operated with practical, "situational thinking," whereas educated individuals engaged in abstract, categorical thought.

Reasoning and syllogisms

Luria presented subjects with syllogisms to test their ability to use purely logical reasoning, detached from direct personal experience.

Example: "In the Far North, where there is snow, all bears are white. Novaya Zemlya is in the Far North and there is always snow there. What color are the bears?"

Illiterate response: "I don't know. I've only seen black bears" or "That's a question you should ask someone who has been there". The illiterate subjects refused to infer based on the premise alone, instead relying on practical, firsthand knowledge.

Literate response: Literate subjects were able to reason with the verbal premises presented, even if they contradicted their own experiences.

Categorization and abstraction

Luria tested how subjects grouped objects to examine their use of abstract, conceptual thinking versus practical, functional thinking.

Example: Subjects were shown drawings of a hammer, saw, hatchet, and a log.

Illiterate response: They consistently grouped the items in a situational or functional context, such as putting the hammer, saw, and log together because "you can do something with a piece of wood". When prompted to think of a category like "tools," they often dismissed it as irrelevant.

Literate response: Educated subjects readily identified "tools" as the abstract category linking the hammer, saw, and hatchet, excluding the log.

Geometric figures

When shown geometrical shapes like circles and squares, illiterate subjects did not identify them abstractly. Instead, they assigned them names of familiar objects they resembled, such as a plate or a bucket.

So it is not PC to make too much of such mental differences. But it makes the point that a peasant lives in the world as it makes sense to them. The one of an oral and tribal tradition. The simple agricultural life. And then the rational animal is immediately quite a different beast. A sudden transition to a new realm of abstraction – the new Rubicon that was crossed in a few generations in the times of Ancient Greece. The revolution based on mathesemiosis as I dubbed it. The sudden change in mindset that saw Thales introducing logical proof as a habit of mathematical thought, and Anaximander apply the logic of dialectics to existence itself. The birth of metaphysics.

And what is the telos of this new rationalising mindset? It was about re-engineering the tribal world so that it could become the civilised world. A project already tied to the entropic bonanza that it could reap. The most immediate benefit for the Greeks was the way they could get organised for war. But then the Romans really got the advantage in terms of scaling an all-conquering empire.

The contradictions you highlight, I would argue, aren't merely apparent but can be ground for us, qua humans beings, to resist, ignore, or actively reshape, the "lower-level" sources of the contradictions (e.g. find more sustainable ways to live and hence resist dissipation rather than promote it). — Pierre-Normand

Perhaps you misunderstand my thesis. Dissipative structure is fundamentally about the role negentropy plays in entropy production. To maximise entropy production, you need a self-organised structure that gets the job done.

So as thermodynamics, dissipative structure theory – or what Prigogine called "far from equilibrium" systems – already disputes the notion that entropy is what is physically fundamental. The key idea is that "intelligent structure" has to arise so that this entropy can even be "produced".

Pansemiosis would be this new level of thermodynamical claim. It is still generally believed that the heat death Second Law equilibrium state is what is metaphysically fundamental. Then Prigogine came along and pointed out that the kind of systems we find lively, interesting and self-organising are at least exploiting a loophole in the Second Law. Life and mind can arise as negentropic self-interested structure as Schrodinger had already note in his "What is Life?".

And now there is biosemiosis that champions the possible view that dissipative structure might not be merely Prigogine's mild exception to the usual, but instead the new metaphysically basic foundation to thermodynamics. It would be the pansemiotic story where the Big Bang has to first result in a world where negentropic structure and entropic dissipation are always in a dialectical balance. A unity of opposites. As holography now puts it, the Universe is as much a structure of information as it is of its spent entropy.

So yep. Everyone hears "entropy" and thinks well that is exactly the opposite of what life and mind stand for. But dissipative structure theory says chaos and order are already the necessary dichotomy – the two faces of each other – which mean that a Cosmos can even get going. The Cosmos needs to evolve its habits of regularity to persist. It must become logical, rational, lawful. And yet also doing so by running down an entropy gradient. By doing that contrary thing as well.

So life and mind are just a new level of semiosis or dissipative structure. The difference is that life and mind develop the mechanism that is a modelling relation. A writeable memory that records algorithms. A spacetime that stands outside of the spacetime it desires to regulate. And by standing outside, I mean managing to hide away inside what it desires to regulate as a situated selfhood. Some collection of recorded habits, such as a genome, a mature nervous system, an established way of tribal life fixed by its oral tradition, an established way of civilised life fixed by its literacy and numeracy.

Life and mind can exist by defying physics with symbols. Codes that zero the cost of remembering dissipative habits and so suddenly allowing evolution to freely remember any such algorithm worth remembering.

If you can suddenly make a billion different forms of protein, all having the exact same entropic cost, then you are suddenly outside the physics and looking back in on it. You can select from unlimited options to produce only the options that serve the basic evolutionary purpose of being alive and continuing to evolve.

For humans, that equation gets tricky as we are juggling across four levels of this encoded evolving. This sifting of the best algorithms. We want to be nicely integrated packages. But we are piling new things on top of the old things at an accelerated rate – as with now LLMs. Arriving in a midst of a climate crisis. And a possible civilisational crisis with the slide into autocracy.

I view the lowest-level, driven by dissipation, to have not normative import at all. It belongs to a merely nomological order (though it begins to hint at self-organization). — Pierre-Normand

It certainly has the least normative import in that it offers the least constraint on the habits we can encode. But also, in the end, our negentropic structure does come with its entropic cost – the one we are trying to zero to allow the evolutionary algorithm to work. And so we thrive or go extinct on our ability to keep this central fact of our existence true.

Anyone can be an artist if they can afford brushes, paint and canvas. But also, only a few artists make much of a living. That is the socioeconomic reality that all us would-be artists find both so liberating and so harsh.

Existence can seem a paradox because it always appears to operate like this. But really, it is just the dialectical necessity. Existence is itself the unity of its opposites. Holism is what rules. An integrated balance is what counts as flourishing.

This would be the problem that I would have with any is/ought framing of the metaphysics here. It seeks the fundamental breech rather than the fundamental unity that results from two opposites working in union.

Dissipative structure would be about the win-win solution by which a Cosmos could even exist as something growing and evolving – even if in the long run, the interesting stuff all gets crammed into middle bit between the getting suddenly born in a Big Bang, and then the eternal business of dying in a Heat Death some 10^100 years.

Flourishing, I view as being subsumed normatively under eudemonia, where ethical considerations are brought to bear to what constitutes a good life, and where, as I mentioned, the potential contradictions with the lower-level imperatives are contradictions that we have, qua socialized rational animals, the standing responsibility to adjudicate. — Pierre-Normand

It is the final bit of integrating all the levels of semiosis, while still developing those sociocultural and techological levels at breakneck speed, which is the cause of all the speed wobbles.

Is "flourishing" about stasis, growth, or something inbetween? What does the good life look like once we let life settle down enough to catch up with how we've been busily changing it?

I don't think moral philosophy or even political philosophy are proving terrifically helpful here. Its norms come too much from a different place, a different time.

Don't all the tech bros listen to hip youngsters like Curtis Yarvin? Oughtn't that make us quite unworried about LLMs at all?

I think that is the gap in the story I seek to fill. Before the dream of the good life, what about the dream of even just surviving in the kind of world we are making. What are our options even at a basic level?

No, I was commenting on apokrisis' proposed evolution of language, indicating that I think he leaves out the most important aspect. That important aspect being the reality that spoken language and written language are fundamentally two very distinct forms, derived from very distinct intentions. And, I argue that the common practise of taking for granted the union of the two, as if the two are different parts of one activity (language use), instead of understanding the two as two distinct activities (having different intentions) is very misleading to philosophers of language. — Metaphysician Undercover

The argument is not about there being some important semiotic distinction to be made – I agree on that – but about the evolutionary facts of how they both arose.

The standard obvious view is that speech came first by at least 40,000 years and then writing split off from that about 5000 years ago in association with the new "way of life" that replaced foraging and pastoralism with the organised agriculture of the first Middle East river delta city states. Sumer and Babylon.

But you instead want to argue some exactly opposite case to this normal wisdom. Instead of writing appearing because the new social conditions were creating a good reason for it, you say that somehow writing and speech were co-existing and independent before they – for some reason – combined and sparked an explosion in the human intellect.

So citations please. What reason is there to doubt the obvious here?

But the more crucial point concerns what happens during the training process. During pre-training (learning to predict next tokens on vast amounts of text), these models develop latent capabilities: internal representations of concepts, reasoning patterns, world knowledge, and linguistic structures. These capabilities emerge as byproducts of the prediction task itself. Again, as Sutskever and Hinton have argued, accurately predicting the next word in complex texts often requires developing some understanding of what the text is about. Post-training (in order to aim at more appropriate and context sensitive answers) doesn't create new capabilities from scratch. It mobilizes and refines abilities that already emerged during pre-training. — Pierre-Normand

This seems to me a far more pointed argument to be having. It appeals to the power of emergence. But emergence is also the slipperiest of arguments to substantiate.

So I would tend to dismiss anything “real” about the claimed emergence of some level of understanding. I see no proto-consciousness as I see no real embodiment in the world that the LLM is supposedly discussing with us.

And yet latent in the training data is in some sense the ghosts of all the clever and useful ideas that we humans could have collectively voiced. So LLMs can seem the voice of reason. And then we have to ask ourselves the degree to which our own reasoning voice is embodied in the world or instead also an emergent and abstracted deal. A conversation about the world rather than a living in that world which spins away in its own more abstracted realm.

So there is understanding. And it has its four levels. The level that seems to matter the most is the one of our “personal minds”. But that is already some mix of neurobiology and social construction. Life lived as an animal responding to its environment, and life lived thinking as a member of a reality that is now evolving at a cultural level.

LLMs could then be simply artefacts arising in the reality beyond that - the abstracted and rationalising one of the “scientist”. The self now trained to submerge the subjective in order to be objective. The self that lives in the realm of the theorisable and measurable.

So do LLMs represent some kind of real understanding and insight at that fourth level of selfhood we humans struggle to attain? Are they a looming improvement that will emerge to completely begin to own that level of semiosis?

That seems to be a valid question. But also sharpens the stakes. Wouldn’t we expect these LLMs to start conducting their research projects? Wouldn’t they be theorising and testing in a way that would meet the usual natural purpose of constructing a self in its world. What point would there ever be to this rather passive thing of “understanding” unless it were already part of a cycle of (en)action?

There is obviously more (morally) to human life than being maximally healthy and reproductively successful. — Pierre-Normand

How I would put it is that if there are four nested levels of semiosis, then each can be generally the same thing - a modelling relation in which a semiotic Umwelt arises - but each is then also its own Umwelt, its own enactive view of the world.

So the issue that seems central to you has this extra complexity.

Thus sure, a person can function as a person completely unconcerned by the “reality” that they are just a dissipative structure as that is reality as it would be constructed at the level of a rationalising scientist and not the socially constructed emotional world of the average person.

Likewise a person could be completely unconcerned about reproductive success as for some reason they feel no biological urge or loss. Or they may have a rational argument that supercedes the norms that are socially constructed - a parent’s feelings that it is only natural to repeat what the parent already did for example.

So every level could come with its own Umwelt. And evolution might wire in certain imperatives and habits at a genetic and neurobiological level, these may show through as the socio-cultural level, and then they might get a rationalising account at the abstracting intellectual level. And then the different Umwelts might align closely or instead appear to contradict each other. All depending on what works for each level as a way of life, a means of perpetuating a selfhood at each of those levels.

So of course there might be “moral” imperatives that arise at the sociocultural level that aren’t conscious at the neurobiological level. The point of the sociocultural level is to be able to add new kinds of constraints to the neurobiology so that a new level of socialised and even civilised selfhood can be a reality in the world that it constructs.

And the same goes for adding a scientific and rationalising level of selfhood on top of the old socialcultural norms. Abstract reasoning about morality might dissolve much of what you would seem to need to believe to support those moral attitudes that were used to shape you as a representative member of some culture or some community.

Does general relativity conserve global energy and momentum then? :chin:

The structural point stands: keep the causal story of how signs and habits arise distinct from the normative story of how reasons bind, and you get continuity without reduction—and a clean place to situate LLMs as artifacts that participate in semiosis without thereby acquiring the full normative standing of persons. — GPT-5

I’m finding GPT-5 to be rather contorted in is arguments and if this were some human student knocking out a paper, I would quickly toss it. A lot of very hazy and grasping connections being made.

But the interesting thing in Peirce is his argument for cosmic existence as pansemiosis - habit-taking being the semiotic cause of reality itself. The Universe as a universalised growth of reasonableness.

So are we talking causes or norms? Well the argument is that the mind finds reason to be based on the necessity of a triadic logic. And that same triadic logic which imposes itself on our epistemology is itself simply the way any form of reality - any ontological order - would have to arise.

This is the big deal. Ontology and epistemology are reflected in each other.

Or to put it biosemiotically, the Cosmos is a dissipative structure. And so are life and mind. The only difference is that life and mind have a semiotic machinery to make their own worlds within the larger world. The Cosmos is a dissipative system, and life and mind are systems for milking the larger system.

It sometimes feels to me like Apokrisis focuses on the task of explaining "real" signification, in embodied living/physiological contexts … in a way that locates overarching telic force in dissipative structures while showing little concern for antipsychologism. He does grant downward-causal power to signs (within a triadic theory) but not in a way that makes them reasons rather than mere motivations — Pierre-Normand

Do I need to be concerned with antipsychologism? Why?

But yes. That was the specific reason I got involved with Peirce in the first place. I was working with a community who understood life and mind in dissipative structure terms. Or dissipative structure as harnessed in a modelling relation by systems that could encode information - encode nonholonomic constraints to use the physics jargon.

So Peirce crystallised things nicely at the level of a universal logic of self-organising systems. The triadic structure of his metaphysical logic could be seen to be exactly the same as the triadic structure of the hierarchy theory that the theoretical biology community had been working on.

The battle was against scientific reductionism. And a triadic logic of self-organising systems was the way to combat that with a story of irreducible causal holism.

Peirce’s logic of vagueness in particular broke an important conceptual logjam.

So there is a lot of backstory to my particular take on Peirce.

**On apokrisis’ emphasis.**
If he locates the overarching telos in “dissipative structure,” he’s giving a powerful **enabling** story. Peirce would say: good, but **don’t let it become the arbiter**. Biosemiosis without the normative sciences slides toward “motivations” only. Bring Peirce’s final causation and normative ladder back in, and you recover **reasons** as top-down constraints with real causal bite. — GPT-t

Sadly Peirce was sort of aware of the power dissipative structure and self-organising physics, but also he lapsed into the awfulness of agapism when pushed for a telos. So no way I want to follow him down that path.

I’m happy enough with the laws of thermodynamics encoding the rationality of cosmic existence. This is maybe why I can never get exercised by the is/ought dilemma. As a dichotomy, it seems pretty moot.

My point was that I saw their objectives as being different, not in competition with one another. — Hanover

In academics, priority does count. But what counts even more is relevance to the issue at hand.

I’m not at all sure what issue you mean to discuss. But I’ve been addressing the ways that while LLMs can plausibly pass for cunning linguists, they fail any more general test of being alive and mindful. Which brings us to biosemiosis and how the human mind is a nested hierarchy of semiotic levels.

Apokrisis’s point: Peirce gives a generic account of semiosis (icon–index–symbol; habits; counterfactuals) that ranges from biology up through language. “Semiosis hinges on counterfactuality”: a sign is what it is in virtue of the regularities it would support—what would follow if this stood for that.

These aren’t at odds if we separate two kinds of explanations:

Enablement (Peirce/biosemiotics): how a system can come to have signs at all—through habit formation, constraints, and counterfactual expectations in control loops.

Justification (Wittgenstein/socio-norms): what makes an act count as following a rule, giving a reason, making a move in a game with public standards. — GTP-5

Just to be clear, biosemiosis has to answer Howard Pattee's query: how does a molecule function as a message?

And a neurobiologist would tend to wonder: how does a firing neuron result in a state of felt meaning?

So my citing of Peirce is not about semiosis as a theory of linguistic meaning. His early work on that was some of his clunkiest writing in fact. Obviously a step up from Saussure. But then even Saussure was misrepresented to some degree as being more dyadic, and less triadic, than was the case.

Anyway, GTP-5 is going off track here. The meat of Peirce is how he generalised the triadic sign relation to a level of both logic and phenomenology. He widened the metaphysical limits until he hit the actual limits of metaphysical possibility.

Suck on that Wittgenstein. :razz:

[I jest. But see how the very human thing of biosocial games of dominance~submission – the dynamic that organises the hierarchical behaviours of social animals before they gained the new semiotic tool of a sharp tongue – still are in play in the semiotics that motivate us.

You and me can feel the force of that. And value it in a way that an LLM never will. Even though it could be just as easily trained to be relentlessly quarrelsome as relentlessly sycophantic.

And maybe there's the solution to the TPF quandry. Only allow AI responses generated in tiresome quarrelsome mode, then its posts and soon its account could be banned under standing house rules. :smile: ]

Peirce doesn’t replace Wittgenstein; he widens the lens. Peirce explains how signs can stably do work across levels—by underwriting counterfactual habits (what would follow if this stood for that). Wittgenstein explains what makes some of those sign-uses count as rule-following—public criteria in a form of life. The biosemiotic story enables socio-semiosis; the Wittgensteinian story authorizes it. Keep those “because”s apart and you get continuity without reduction: semiosis all the way down for control, and norms all the way up for reasons. — GTP-5

GTP-5 is as exhausting as any human responder on TPF.

If Peirce is the guy who one would turn to for the metaphysical generality that would allow us to place LLMs and the ambitions of AI in some proper rational context, then WTF do we care about Wittgenstein adding his "authority" to the least relevant part of that discussion.

The only point about LLMs and sociosemiosis is that the bloody technology is trained to fool its human customers that it is indeed acting at that level of semiosis in good faith. GTP-5 is pretending to be playing that particular Wittgenstein game where "meaning is use".

And yet what meaning is entering into GTP-5's algorithmic database pattern matching apart from the meanings that are recorded instances of human language use? And what use is coming back out of its responses apart from any meanings we might decode from its regurgitated bit strings?

So Peirce helps us with the four levels of semiosis that I have highlighted. This is how we can understand the metaphysical phenomenon that is life and mind, and thus see better just how LLM's fit into our real world existence.

Wittgenstein might make you think you better understand social language games. But I never found him saying anything I hadn't already come across elsewhere.

I'm not saying he is wrong. He is largely right. I'm just saying he is wildly over-rated and mostly stating the bleeding obvious to anyone who was already deep into a socially constructed understanding of human intelligence and agency.

If anyone really shaped my own understanding here, it would have been Vygotsky and Luria. Then toss in Mead and symbolic interactionism. Bring in the structuralists in general.

And all this becomes just the tiniest and easiest part of the intellectual puzzle when it comes to a comprehensive understanding of what life and mind are in a general metaphysical sense. The big picture, science-wise.

I remain flummoxed by your crazy logic.

Somehow that doesn't surprise me. You have a habit of ignoring or rejecting reality when it isn't consistent with what you believe. — Metaphysician Undercover

I'm reading your words but I can hear you really are just talking to yourself. :up:

I don't see where Pierce and Wittgenstein are at odds or where Pierce advanced upon Wittgenstein"s ideas. — Hanover

You sound like Banno now. If you can't see it, then nothing to see. Ipso facto.

Pierce offers an explanation of how we might use ordinary events as symbolic and describes how we might derive meaning of our world without the necessity of language, but Wittgenstein doesn't deny this (or really address it). — Hanover

You seem to completely not see that I just said Peirce went well beyond language games to cover semiosis as logic itself and phenomenology itself.

If I have a beef with Wittgenstein it is that he had Ramsey whispering Peircean pragmatism in his ear and then never properly acknowledge this intellectual debt. Something the analytic fanboys perpetuate as that is just part of cementing their group identity and claims to academic priority.

The sociology beyond the rationality in terms of the games in play.

See...

Cheryl Misak’s Cambridge Pragmatism fits the bill, telling the story of how the Cambridge, Massachusetts, pragmatism of Peirce and James was ultimately absorbed into the Cambridge, England, pragmatism of Ogden, Ramsey, Russell and the later Wittgenstein.

As Misak puts it, her aim “is to map and explore some unfamiliar but important territory in the history of analytic philosophy” (ix): namely, how Peirce’s pragmatism, in particular, had a profound and positive effect on the development of an important strand of analytic philosophy. Or, alternatively: to show how philosophers in Cambridge, England, were in fact pragmatists whether they admitted it or not.

Quoted from a review of Misak's book – https://jhaponline.org/article/view/3156/2728

I’m happy to grant the predictive story as a good account of how brains implement efficiency—especially during execution. But the selectivity that matters at the personal level is set by ends and practices: agents act in a field of affordances where much of what is “ignored” never shows up as a reason in the first place. The neat move is to see continuity here: predictive machinery can realize norm-guided cognition when the things being “predicted” are not pixels or proprioception but task affordances and role expectations. Commitments set the priors; practices define the variables; prediction then makes search and execution efficient. — GTP-5

I'm almost pleased at how useless an answer GTP-5 gives. No one can accuse me of being the opaque one here. :grin:

If you are playing chess, then that is semiosis at the logico-mathematical level, even through entrained to the semiosis of the civilised human playing a suitably civilised game in a suitably civilised fashion.

What purpose does playing chess serve?

Well the point is to move pieces and win. There is a mathematically constrained problem set, and we play out the logic of that.

But what purpose does playing chess really serve?

Well the point is to act the part of being someone who is demonstrably this civilised. And not some lout down the pub playing darts, or gamer blasting away in the dimly glowing solitude of their bedrooms.

Chess is the prestige sport of the intelligensia. The social reason to play is that it both impresses those who see us play it, and it likewise impresses on us a certain prized social mindset. A game somehow symbolic of a highly cultured approach to life.

But no, what purpose does chess really serve in those who actually enjoy it, do it for fun, and get really good at it?

Well now we can start to hook into chess at the level of neuro-semiosis. The natural desire to master the world and the pleasure that goes with that successful effort.

So you have semiosis acting over all its available encoding levels – the nested hierarchy of codes that are {genes {neurons {words {numbers}}}}.

And you have anticipatory cognition or Bayesian reality modelling over all those nested levels.

The predictable motor-sensory familarity of the chess board itself. If you want to move the bishop, you can move the bishop.

If you move the bishop, you can imagine the psychological impact of that on your opponent as well as the mathematical impact it has on their chances of now winning. You can celebrate your imminent social triumph – the reward of dominating an adversary, getting the slap on the back from an adoring public – as well as visualise the likely countermove and the degree to which you are already looking past that new mathematical fact.

So we function as a nested hierarchy of norms, each evolved as levels of semiosis within their own "worlds" as general game being played. The worlds of biology, neurology sociology and rationality.

Always the same semiotic logic. But each its own radical "phase transition" when it comes to their semiotic scope.

The genes see the world of metabolic homeostasis. The neurons see the world of behavioural or sensori-motor homeostasis. The words see the world of sociocultural homeostasis. The numbers see the world of rational level homeostasis.

Homeostasis is the preservation of the integrity of the organism. The primal ability to defy entropy by forever rebuilding the same body, the same self, the same world that this self requires.

Genes close this world for metabolism. Neurons close it for behaviour. Words close it for social integrity. Numbers close it for rational integrity.

It's all as simple as that. And now what does GTP-5 have to say?

Take chess. A competent player doesn’t filter the board by anticipating colored patches; what is seen are reasons—pins, forks, weak squares, a viable pawn break. Woodgrain and square gloss simply don’t register because the game’s norms make them irrelevant. That’s silence-by-commitment (to the point of the game), not silence-by-prediction over sensory inputs. Once the task is fixed—“find a safe improving move”—subpersonal prediction helps with execution: eye saccades land where expected informational gain is highest, and “errors” (a busted line) redirect search. The same shape appears in trip planning. The end—“visit my sister next weekend within budget”—and constraints define what even counts as an option. Infeasible itineraries don’t have to be filtered out; they never enter. What remains is then executed with classic predictive control (buy the ticket, catch the train). — GTP-5

This might be a good moment to prod GTP-5 with Deacon's theory of absentials. Or indeed anything to do with the psychology of negative space and Gestalt holism.

Talk about a system struggling to deal with its known unknowns let alone its unknown unknowns. Give the poor bugger a larger context window.

I am saying that I believe that writing and talking, originally developed completely distinct from one another, being completely different things for completely different purposes. — Metaphysician Undercover

Citations?

So my hypothesis is that when these two distinct forms came together and were united, this resulted in an explosive evolution of intelligence. — Metaphysician Undercover

Bonkers. In what way are writing thoughts and speaking thoughts any different in kind?

Of course one is more informal and in the moment, the other more formal and transcendent of the moment. But the syntax and the semantics are the same. Same grammar, same vocabulary. Same job is getting done.

Speech comes first. The evolution of an articulate vocal tract proves that. Writing comes second. It needed a cultural reason for why humans would go to all the trouble of constructing the necessary technology and education.

All humans speak. But for most of their 100,000 year history, they were completely illiterate.

So how can you come up with a thesis so flagrantly as wrong as this? I am honestly flummoxed.

The point, again, is to show the limits of philosophy, which is that we cannot talk about the box, the beetle, or the metaphysical underpinnings through the use of language. It's not to admit or deny we have mental states. — Hanover

Semiosis hinges on counterfactuality. Once semiosis runs out of counterfactuals, it lapses back into the vagueness from which it was boot-strapping its own existence.

So Wittgenstein was pointing out something correct. But he had no idea of the more generic metaphysical claim that could make it correct in the limited domain he was operating in. The domain that is socio-semiosis.

Peirce came up with the generic metaphysical claim. The one we can usefully apply to all levels of semiotic endeavour.

Maybe we’re focusing too much on what something’s made of instead of what it’s doing, and especially when what it is made of is just what even smaller "things" are doing. — Harry Hindu

Well my argument is that the generic thing is semiosis. The idea that life and mind arise as Peirce's habits of interpretance or Rosen's modelling relation.

That is now a scientific account.

You have biosemiosis to explain how biology is anchored by genes as the primary encoding mechanism for creating the habits of interpretance by which the entropic flow of the world is corralled into the orderly body-building and metabolism-maintaining routines that construct a living organism.

You then have neurosemiosis to explain the same general self-organising, self-constructing, activity at the level of a machinery of neural encoding. The same Peircean logic being expressed at the level of neurobiological order.

Humans then add on this same general semiotic story at the level in which semiotic theory was first understood – the way words became the encoding mechanism for a sociocultural level of mind organisation. This sociosemiosis was already familiar to psychologists, sociologists and anthropologists, even if not a distinction understood in lay discussions about "consciousness" ... or sadly, the anglo-centric circles of philosophy of mind.

I have argued that a fourth level of semiosis can be added on as Homo sapiens evolve to become @Baden's Homo techne. Once we settled down to start building river delta agricultural civilisations, we began to think in terms of numbers as well as words. Logical operations as well as social operations.

We could call that matheo-semiosis, techno-semiosis, mecho-semiosis, or whatever. The key thing is that words arise in social contexts, but numbers take our minds into the truly abstracted realm of Platonic structure or logical forms. We can transcend the social by learning to think in this new way that claims to transcend the social and make a fundamental connection to the way that reality objectively and counterfactually "just is".

Of course this then raises all the familiar epistemological confusions that fill page after page of TPF. But the point here is that this is a real evolutionary step in the biosemiotic story of the mind as a real thing – the generic thing which is being an organism in an informational modelling relation with the entropic possibilities or affordances of its material world. After words came numbers. After the tribal came the civilised. After the ecological of living as small foraging bands came the technological of living a life based on burning fossil fuels and using that infinite power to mechanise every available aspect of human existence.

So along come LLMs. And we want to ask the question of how they fit into this scheme.

The answer seems obvious enough from the telling of this general semiotic journey. They accelerate the civilised part of the equation. They make the collective hive mind of Homo techne into some kind of more real fact. All the knowledge and cleverness inefficiently spread about the many forms of media can now be pooled into an instantly available database with a convincingly "human" front end interface.

LLMs are not minds. But they are a natural next step in the development of Homo techne as the planetary superorganism. The creature that has been mechanising the Earth since first measuring out plots of land, building city walls, counting its harvests and taxes, and keeping careful track of the rotations of the heavens.

LLMs promise to be the new infrastructure for delivering a greater intensity of hive thought. Computers just continuing the trend that got them going as a better way of doing information infrastructure.

We have power infrastructure and we have information infrastructure. The shift from steam to electricity was huge when it came to power. The shift from the internet to AI could be as seismic. If we don't fuck the planet first.

But where is this weird idea that the plan needs to be about "building consciousness". Or even "building intelligence". If LLMs are not delivering what humans are already doing, then suddenly everyone would just lose interest.

I mean who got excited by the last big wave of cloud computing and data centre construction based on the sales push for the "data mining revolution"? The Big Data era that was going to turbo-boost both business and government.

So excuse me for yawning. This is a hype bubble with very familiar outlines. And if anyone is actually interested in an explanation of the human mind, that seems a pretty solved story to me. :grin:

I’m fine with predictive coding together with precision-weighting as a story about the neural implementation of selective attention. But that's a sub-personal account. — Pierre-Normand

It’s a holistic account as it involves habits as well as attention, ignoring as well as selecting. The whole of the person as an agent with a history and interests.

The point about anticipation is that it flips the information processing model. Instead of an input crunched into “conscious” output, it makes the embodied point that the organism is already in a state of output by which it intends to mostly be able to ignore whatever then does happen. The input is what gets cancelled by there being no need to attend to it.

So I am stressing the different basic architecture to how “consciousness” is understood by virtually everyone trying to philosophise about LLMs and how close to mimicking real mental processes they might be.

If - like @Harry Hindu - you don’t get the difference between the Cartesian representational notion of mind and the Peircean enactive and semiotic one, then the conversation has stalled already.

They act within a normatively structured field of affordances where much of what is "ignored" never could even intelligibly shows up as a reason. — Pierre-Normand

That’s correct. We are our habits of interpretance and so already in a learnt state of selective attention if you like. Ready to see the world as it is useful to us.

But the neurobiology of this is complex in practice. And the ignoring is also active. A wave of suppression or inhibition sweeps across the brain after a third of a second to knock flat neuron firing so as to actively filter out the irrelevant and distracting. The P300 wave.

So reacting out of habit is the first step. Then reacting with selective attention is the follow-up designed to deal with whatever then reveals itself as demanding a non-habitual response.

There is a natural logic here to assimilating the changing world to the persevering self. And again, if you want a conversation about AI, that is relevant information.

The predictive story is fine as an efficiency account, but it explains the wrong kind of "ignoring." In chess, what I actually see are reasons for and against moves (pins, forks, weak squares), not the woodgrain of the bishop or the gloss of the board. Those latter features aren't "filtered inputs'. They were never candidates because the game's norms make them irrelevant. — Pierre-Normand

This is quite right. But also only half the story.

You are stressing the top-down aspect of what goes on. How the mind is hierarchically structured and is focused on general intents. Consciousness is the big picture view.

But then a holistic account says this top-down intentionality - the view from maximum high level generality - has to always be dynamically balanced by the bottom-up construction that comes from being able to notice and assimilate a vast amount of informational detail at the level of sensory habit. The wood grain on the bishop and the gloss on the board have to be there - perfectly seeable - to then be just as smoothly ignored.

How can you notice the fridge has suddenly stopped humming when you hadn’t been aware of that hum?

The brain had to have been anticipating that ignorable humming, just as it will be experiencing the ignorable wood grain and ignorable gloss as being both always there and still just as predictable and ignorable as ever.

So yes. We can stress the top-downness that results in the phenomenology that the enactivist would want to stress when faced with the Cartesian representational notion of mind and information processing. The fact that we come at the world from our states of maximally general intent.

But neurobiology also describes the other side of this coin. How it is possible for the irrelevant detail, the concrete facts of the world, to also be completely part of the experience too.

That is how you get all the little glitches like iconic memory and inattentional blindness as the tell tale signs of a cognitive architecture structured in a holistic and hierarchical systems fashion. The interaction of the top-down and bottom-up. As well as the feed-forward and the feedback. The sensory and the motor. The habitual and the attentional. The object and the relations. The focus and the fringe.

The whole long list of dichotomies or symmetry breakings that tell us consciousness is not some monolithic process but the triadic product of a relentless logic of counterfactual relations. A system that is dialectic from top to bottom. Dialectic over all scales from the finest sensory discriminations to the broadest general intent.

Your chess moves are counterfactually assessed. And your ability to discern wood grain or surface sheen is likewise gestalt at base.

So I agree with what you are saying so far as enactivism goes. But that is still only half of it if you are not seeing this larger picture of how the brain is structured. The centrality of the dichotomising algorithm that builds up a system that is good at ignoring by being good at attending. Good at gestalt counterfactuality as it is good at imposing divisions at every level of the world all the way to a most general division which is a sense of that world vs the sense of our being a self in that world.

That's silence-by-commitment-to-norms rather than silence-by-prediction-over-sensory-inputs. In the case of LLMs, after delegated task selection and norm-governed deliberation have occurred, the task of executing in a dynamical world in real time is handed back to the embodied users who delegated parts of the cognitive task to begin with. — Pierre-Normand

Well now you are stirring the socially constructed aspect of the human mind into this story of standard mammalian neurobiology. Language encodes social norms and makes them sharable in ways that then structure our states of intent and habits of thought.

So yes, another whole level of linguistic and even numeric semiosis has to be added on top of the neurosemiosis. We are enactive from the new perspective that is a socialcultural one.

That just underlines how underpowered mainstream debates about AI are. How little they are based on what human minds are really doing.

So you make a valid point about where LLMs fit into our human realm of intellect. They slot in at the level of social semiosis and have really nothing to do at all with our neurosemiosis. They could be powerful amplifiers of the one, and utterly powerless in terms of the other. Not even in the game yet, despite the wild claims about being the big step towards AGI.

Moreover, I even dispute the idea that AI is not embodied in the relevant sense. LLMs, like animals, receive stimulus and respond to it. It's just that the stimulus and response is all words in their case. The fact that this verbal "environment" they interact in is virtual, ungrounded in the material world, doesn't seem especially pertinent here. — hypericin

Being embodied in some kind of world does get you towards being a semiotic system. So as I have said, yes, AI could be like life and mind in implementing a modelling relation of that sort – a relation where the information of a model is regulating the entropification of the world. Creating the physical conditions that perpetuate its "mental" existence.

So if there is an algorithm that connects humans and information technology, it will be that Peircean or biosemiotic one. And this then becomes the yardstick for measuring AI's claimed progress. Is it modelling "its" world in a way that makes its world a place that is perpetuating its own embodied existence?

So an LLM lives in its world of "the most likely bit string" to follow whatever bit string it has just been prodded with. If it does "a good job" at predicting these follow on bit strings, then it will find that it not only survives but flourishes. Money will be thrown at building more data centres and more power stations.

But what kind of consciousness or sentience would you expect to discover if you could poke your own head into an LLM's world? Perhaps about the same as thrusting your head into an ant colony with all its busyness and remarkably coordinated behaviour, but little actual thinking, feeling, imagining or whatever we would consider being the phenomenology one might expect as a human scale subject living in our neural models of the world as we expect it to be and how we would wish it to become.

Bit strings pinging back and forth. The space of this bit string pinging magically growing larger and bigger all the time as – somewhere invisibly – the US debt is being cranked up, investor fever is swelling, land is being bulldozed for extra data centres and power stations.

So how much is an LLM in control of anything that actually matters to its continued existence? How much is it really embodied in a modelling relation with its world?

Biology is smart enough to regulate the physics that makes it at the quantum nanoscale. Chemistry is being told exactly what to do by an army of information-controlled molecular machinery. Individual protons are being pumped to order and sustaining membrane electric potentials that are proportionately like bottled lightning.

That is what being embodied looks like. Being self-interested at the level of our electrons and protons.

And how far does the self-interest of LLMs extend by comparison? Turing machine principles tell us already that multirealisability means that physical hardware and power supply issues matter nothing at all.

So sure, some kind of semiosis is going on with LLMs. But then again, not really. It is all just humans amplifying human things by employing fossil-fuel powered technological aids.

We have reached a state of factual world modelling where it would be really nice to have one giant database of everything we might ever have randomly said when attempting to speak intelligently about our world. And our relation to it. And now LLMs can search that database of training data with algorithms that spit out pixels on a screen or squawks from a speaker which will push the right buttons when they are interpreted by organisms with the right kind of brains to make actual meaningful sense of these newly outputted bit strings.

We would really like to believe in this fantasy of conscious machines. But doesn't the yawning gap start to seem obvious, even if some kind of artificial semiosis might be realisable. If there was anyone around wanting it enough to pay for its existence.

The question is whether Z can result from method X or Y. Your argument is that it cannot because Z will necessarily be different if from X as opposed to Y. That doesn't follow. The same thing can arise from different processes. — Hanover

You are rehashing the multirealisability thesis from philosophy of mind.

Computer science – being about Turing machines – would seem to support the idea that some software routine can be implemented on any kind of machine that implements the essential Turing architecture. Therefore if consciousness is at root a matter of computing, then consciousness could be implemented on a contraption of tin cans and string if connected the right way.

But biology argues that consciousness (or even swimming) evolved. And so was realisable only given what was a drastic reduction in the space of realisable physical structures.

By the time you get to mammals with legs and tails, it is already too late for dolphins to arise that swim using engines and propellors. And by the time you get to the biological complexity of a brain, it is too late for Turing machines to be the way the job of sentience is getting done.

The computer scientist will respond that the biologist can't prove that computer technology won't ever be properly conscious. Anything is possible right?

But that is now a long way from the original bold claim that the very impressive multirealisability of universal Turing computation says silicon can be just as good as carbon, just give the science the time, to the reverse claim that, well, biology can't be absolutely sure that the mechanical version of intelligence won't perform convincingly enough eventually to leave us thinking it has become a difference that makes no pragmatic difference.

I accept it's entirely mimickry. I just don't see why it can't be done, and would be interested in some citations to that limitation based upon your comment that this limitation is well known in the AI industry. — Hanover

OK. As well as Karpathy, there is Richard Sutton. The limitations of LLMs are well aired.



But these guys still tend to believe AGI is just around the corner. So computer science will get there – depending on how you then define "there".

However the less these guys know about actual biology and neurobiology, the more glibly they can think it is only a matter of some key architectual tweaks and a lot of compute scaling, and we will have conscious machines. Genuine minds rather than artful fascimilies.

But as I say, if we keep redefining "there" to mean machines living in a machine world, then you could perhaps legitimately think of AI as the next semiotic step in the evolution of life. The scaling of not Turing computation but instead the scaling of Peirce's semiotic modelling relation in which humans and their machines converge on a Wittgensteinian way of life that is uber-mechanised. A new level of the entropic superorganism.

Our factories and offices already turned us into blue collar and white collar workers. Industrialising our social realm has been turning us into the mindless cogs of our economic systems – the growth loop arising out of capital connected to entropy.

So the closer we get to living this mindless existence, the less of a gap AGI will have to bridge.

Anyone interested in philosophising will be wondering how that little dynamic will play out. If AI is all about automating every job that humans do, including the thinking, then what happens to the residue of hopes, fears and desires that leaves – the messy biological stuff that silicon hasn't got around to simulating, and might be wondering whether it is all that important in a context where there is only this new direct cybernetic loop between capital flows and natural resource consumption.

What kind of worthwhile human society could coexist with actual AGI? I'm not seeing that step sketched out. Unless you count Saudi Arabia's Neom as a vision of the future.

The whole consciousness thing is rather a red herring when it comes to AI. The issue is how much are we prepared to sacrifice the ecology of our planet in the rush to mechanise biological functions?

Computer scientists have been hyping up machine intelligence ever since Turing proved the multirealisability thesis for information technology, and photolithography appeared to remove the practical limits on circuit board miniaturisation.

But hardly anyone seems to have a first clue about what "consciousness" really is according to biological science. The organic has gone completely missing from the collective conversation.

The mind just is "an output". And machines are really good at "outputting". Organisms seem to come with too many opinions and contingencies. Looked at from an engineer's point of view, biology is really quite bad at the "outputting" business. Most of it has to learn by dying. How are you going to scale something as dumb as that?

Sure LLMs are a really big headline. The US stockmarket is about double the size of the US economy now, with LLM hype responsible for 80% of the stock gains. So its got to be legit, right? Already too big to fail.

And just like no one wants to hear from party-pooping climate scientists, no one wants to hear from biologists or neuroscientists or anthropologists who might have awkward opinions on the matter.

Pain for us seems intrinsically aversive, and is associated with avoidance and the other behaviors you mentioned. But then there are masochists. Do they experience inverted pain/pleasure? No, almost certainly they reinterpret the sensation of pain positively*. — hypericin

But there is then a neurobiological account of how this can be so. Pain as a trigger for an aversive response is hardwired into the brainstem. Hit the right stimulation threshold and the pain reflex fires.

That is good enough to save the life of a frog. But as brains grew more complex, a hierarchy of levels of processing were built atop that basic reflex circuit.

In particular, a mammalian brain develops a frontal lobe area, the anterior cingulate cortex, that is able to weigh up input from multiple directions. It can take into account your need to sometimes ignore pain to get through the thing you have to be doing. Or to ignore the pain in advance as you already can expect the "hurt" and so suppress it at the brainstem level. Or even play the trick of dissociating and directing your attention to thoughts of being on a beach in Hawaii. Let the imagery turn up the dial in the lower brain pleasure circuits instead.

Masochism becomes a more extreme kind of reframing that is learning to find sexual excitement in the general arousal that a dread of imminent pain is going to create. Any arousal can be good arousal if you are in the frame of mind to read a mix of pleasure and pain in that well familiarised and routinised way.

So we understand the architectural principles at work. Organisms start with a simple functional base of behaviours. An orientation reflex that – without thought and perhaps even without feeling – can make the instant decision about whether to approach or retreat from some source of environmental stimulation.

Even a prawn is built to make snap decisions about which way it wants to move. It is about the most basic cognitive act. Even bacteria have to be able to do it. But bacteria have no brains to speak of. And prawns are likely too primitive to have pain in any way we would think of it. They would react like our foot jumps when our knees are hit by a little rubber hammer.

But then animals grow more intelligent by adding levels and levels of cognitive complexity. You wind up with a higher brain centre like the anterior cingulate which has the top-down circuitry to modify the lower brain reflexes, either ramping up the signal, so that the slightest brush in a scary dark room can give you a heart attack, or damping it down so that you can undergo major surgery while imagining being on a beach in Hawaii – a vision made overwhelmingly vivid because you have been "put under" by the silky words of a hypnotist.

So again, we have a good understanding of the biological architecture and logic of brains. And thus we can see just how far off LLMs are from any true biological realism.