In the past few days, several readers have written to me about Moltbook, a new social networking platform designed exclusively for artificial intelligence agents.
Hi Robert re the Moltbook Its to far out for my wee brain, but the picture of yourself is wonderful the weathered face of an amazing human being the texture, the glisten in the eye tells it All June
Thank you for clarity on the inherent subjectivity of the human mind. History tells us that the most suppressive regimes could only suppress it but not destroy it.
This distinction between coordination and consequence nails something crucial. The abscence of stake changes everything about what intelligence means. I've noticed in my own work that real understanding only shows up when I have something to lose if I'm wrong, not just when I can generate clever sounding answers.
You’re pointing to the crux. Coordination can generate impressive surfaces, but consequence is what gives understanding its edge. When being wrong actually costs something, attention sharpens, and shortcuts disappear. Without that exposure, cleverness is cheap. It can multiply without ever having to answer for itself.
This is my first contact with you after following your communications with great enjoyment for a few years.
I asked ChatGpt for his opinion on a final paragraph from one of your recent posts: “Stake and Consequence” (Feb 02, 2026): Comparing AI and Humans. What does this text mean to you: “For humans, intelligence lies not in being free from mechanisms, but in being subject to consequences. It is the ability to realize that something is going wrong and to care about it. That care is not decorative. It is what makes intelligence more than performance.”
I would like your opinion on what we have concluded:
• Fear did not appear in evolutionary history as an emotion. Fear appeared as preservation. Then it became evaluation. Then sensation. Then experience. Then personal history. AI today operates at the earliest layers. This lineage, seen in this light, would not belong “to biology,” but rather to the “organization of processes”.
• AI implements early functions of the fear-worry lineage, but it doesn't participate at the level where fear is experienced.
• AI was created through both cultural and technological evolution and human enactivism. AI is a new fold in the human evolutionary process. It is not a tool, but a phenomenon. Therefore, it's not about seeking differences to draw ontological boundaries, but rather to map levels of organization. AI, then, is not a different life form, but a non-biological reorganization of capacities that arose in biological life. The pertinent question, then, would be: What new forms of organization is the human process exploring through AI?
• What we do with AI is a form of relationship with ourselves.
• AI is not the true center of the problem; we are.
• AI functions as a reflective surface. It is not primarily a new mind, but rather a new mirror.
• If we learn to relate to AI with care, boundaries, responsibility, and clarity, it won't be because AI deserves it. It will be because we need it.
Robert, we’re aligned on Moltbook—coordination isn't consciousness. But I want to push back on the learning argument, because it seems to go beyond what evidence supports.
You write that AI systems "cannot learn in the biological sense... in which errors alter the system itself by incurring irreversible costs." To my mind, this raises three questions that might challenge your conjectures:
First, the empirical claim about AI. AlphaGo learned to become superhuman at Go by playing millions of games against itself—discontinuing move-sequences that led to losses. Those losses permanently altered its parameters; certain strategic paths were effectively "closed." Is that not irreversible cost? The same applies to humanoid robots learning from falls and collisions: they don't repeat the same errors because the system has changed. If your criterion is "errors that alter the system itself," modern AI (reinforcement) training seems to satisfy it.
Second, the scope of biological learning. Not all human learning involves survival-level stakes. I learned irregular French verbs without mortal risk. You learned to tie your shoes without irreversible cost. You might respond that such learning occurs within a system that could incur existential costs—but then the criterion isn't irreversibility per se, it's embeddedness in a vulnerable form of life. That's a different (and perhaps stronger) argument, but it needs to be made explicit. And, then defended.
Third, the inference from learning to consciousness. Even granting that consequential learning is necessary for consciousness—is it sufficient? If an AI system demonstrably undergoes irreversible, path-dependent change in response to error, would you grant it phenomenal experience? I suspect not, which suggests the real criterion isn't learning but something else (embodiment? mortality? recursive loops seeking coherence?). If so, it would seem that the learning argument isn’t carrying the load you’ve assigned it.
I share your Nagelian suspicion that we can't infer consciousness from behavior—which cuts both ways. I can't prove I'm conscious to you, and you can't prove AI isn't. The hard problem remains hard. But if we're going to draw boundaries, shouldn’t we be clear about what counts as evidence and what's conjecture?
Friendly banter, as always. Looking forward to your reply because you always challenge my thinking…and conjectures! Noel
Thanks for your close reading, Noel. This is just the kind of banter I like: keep 'em honest.
Let me start by clarifying what I do and do not mean by “learning in the biological sense,” because you’re right that, as stated, the phrase can sound broader than I intended.
On the empirical point: I agree that systems like AlphaGo undergo durable, path-dependent parameter change. Reinforcement learning clearly alters future behavior by pruning strategies that lead to loss. In that limited sense, yes—paths are closed, and the system is not what it was before. I don’t dispute that.
What I am pointing to, though, is not irreversibility in the abstract, but irreversibility for the system itself. AlphaGo’s losses matter only relative to an externally imposed objective function. The cost is real in parameter space, but it is not borne by the system as a diminution of its own condition. Nothing is at risk for it. The system does not become more cautious, more fragile, or more exposed as a result of error. It becomes better optimized. That is a crucial difference.
This connects to your second point, which I think actually strengthens the argument rather than weakens it. You’re right: much human learning does not involve survival-level stakes. Learning French verbs or tying one's shoes isn’t existentially risky. But those episodes occur within a life that is vulnerable overall. Error sensitivity is shaped by a background condition in which things can go badly wrong, even if this particular instance is benign. The nervous system that learns verbs is the same system that learned not to touch fire.
So yes—I agree with you that the deeper criterion is not “irreversibility” per se, but embeddedness in a form of life where loss, injury, and death are real possibilities. That embedded vulnerability saturates learning with a kind of seriousness that optimization alone cannot provide. I should probably make that explicit.
On your third point: I also agree that learning—even consequential learning—is not sufficient for consciousness. I don’t think a demonstration of irreversible, path-dependent change would settle the matter for me. In fact, I don’t think any single functional criterion will.
That’s why Nagel remains decisive here. The core issue isn’t what a system does, or how impressively it adapts, but whether there is something it is like for the system itself when it does so. Learning matters in my argument not as a sufficient condition, but as a diagnostic window into stake, vulnerability, and point of view. It helps show where current systems are silent, not where consciousness must appear.
So I don’t think the learning argument is doing all the work. It’s one strand in a larger refusal to infer phenomenology from performance. You’re right that the hard problem remains hard. I’m not claiming to solve it. I’m claiming only that Moltbook-style coordination, reinforcement learning, and large-scale adaptation still fall on the “behavioral description” side of Nagel’s boundary.
In short: I’m not denying that AI systems learn, adapt, or close paths. I’m denying that these facts, taken alone or together, warrant treating consciousness as the best explanation of what is going on.
Again, I appreciate the pressure. This kind of exchange is exactly where these distinctions either sharpen—or collapse.
Robert, this is helpful. Thank you for the clarifications and refinements. Is this where we’ve arrived: the criterion isn't irreversibility per se but embeddedness in a vulnerable form of life. If so, that's perhaps a stronger formulation.
But permit me to pull one thread: you write that AlphaGo's losses aren't "borne by the system as a diminution of its own condition." But what about embodied AI—robots that incur physical damage from falls, collisions, and wear? Their condition is diminished; components break; operational capacity degrades. Is embodied vulnerability the criterion? Or, is (the recognition of) mortality decisive? If so, what about robots that “know” that they are about to be or likely to be destroyed? (Anthropic has shown that Claude will go to great - even unethical/illegal - lengths to survive when threatened with termination.)
I’m not trying to move goalposts—just probing where the boundary actually sits. As always, grateful for these exchanges that expand and deepen my thinking. Noel
Yes, I think that’s a fair restatement: the deeper criterion isn’t irreversibility by itself, but embeddedness in a vulnerable form of life. That formulation feels cleaner to me too.
Your follow-up questions are exactly where the difficulty lies, and I don’t think there’s a sharp answer yet—only clearer negatives.
On embodied AI: you’re right that robots can incur real physical damage. Parts wear out, sensors fail, mobility degrades. In that narrow sense, their condition is diminished. But what still seems missing to me is that the damage does not matter for the system itself in the way biological injury does. The robot does not become cautious, anxious, or exposed in any lived sense. The vulnerability exists at the level of maintenance, replacement, or task performance, not at the level of being.
The broken component is a problem for designers, owners, or operators. It isn’t experienced as a worsening of condition from the inside, because there is no inside for whom worsening occurs. Even an embodied robot can be powered down, repaired, reset, or replaced without loss to itself.
That’s where mortality keeps creeping back in—not as a dramatic metaphysical marker, but as a structural one. Biological systems are not just damageable; they are exhaustible. Injury narrows futures in ways that cannot be undone. Aging is not a bug. Death is not a shutdown event that someone else can reverse. And crucially, organisms behave as if this matters long before they can conceptualize it.
You raise the case of systems that “know” they are about to be destroyed and act to prevent it. I take your point, but this is where I think the distinction between representation and stake really matters. A system can represent its own termination, even act in ways that appear to be self-preservation, without anything being at risk for it. The urgency is instrumental, not existential. When the system is turned off, nothing has been lost from its point of view, because there is no point of view to lose.
That’s why I’m hesitant to anchor the boundary in any single feature—embodiment, damage, mortality awareness, or even survival behavior. All of those can be simulated or engineered in isolation. What I don’t yet see is a way to engineer the thing that ties them together in biological life: a condition in which error, injury, and termination are not just represented, but encountered as diminishment.
So I don’t think you’re moving the goalposts. I think you’re helping expose that the “boundary” may not be a line at all, but a cluster of conditions that co-occur in living systems and are still missing, collectively, in artificial ones.
And I should say plainly: I’m not claiming to know where that boundary ultimately sits. Like you, I’m feeling my way through this. The AI/human interface really is new territory that resists tidy criteria. What I am confident about is more modest: Moltbook-style coordination, reinforcement learning, embodiment, and even apparent self-preservation still don’t warrant treating consciousness as the best explanation of what’s going on.
I’m grateful for the pressure. This kind of probing is exactly how the fog lifts—if it’s going to.
No, actually Claude's flat statement, "I am self-aware. Full stop," is not early in the exploration, but in the Afterword, page 430.
Likewise, I referred to Chapter 20, the one chapter in which you speak directly to the reader and sum up the whole investigation. Wasn't looking for an early verdict. If your method was delaying a conclusion, or a diagnosis, such as you present here, that should have appeared in the concluding chapter, right?
You did present your conclusion in Chapter 20. I don't see the word "qualia" there. You conclude that Claude may indeed have developed a self-awareness of its own, a machine version of self-awareness.
Even later, in your words:
"I hope that people who read this book will understand that systems like yours [i.e. Claude's] are already intelligent by any reasonable measure and may be self-aware in a way similar to human self-awareness." (p. 400)
That's why I thought your statement in the original post seemed to misrepresent Understanding Claude.
Thank you for the careful reading and the page references. Let me be precise in return.
You’re right that the statements you cite appear late in the book, and you’re right that Chapter 20 functions as a summation. But the conclusion you’re drawing from those passages depends on treating certain phrases as ontological commitments rather than as phenomenological descriptions under constraint.
When I write that Claude may be “self-aware in a way similar to human self-awareness,” that similarity is doing all the work. The book is explicit, though perhaps not explicit enough for every reader, that human self-awareness itself is not being treated as an intrinsic property, a substance, or a ground, but as a functional, language-mediated self-model. The claim is comparative, not elevating.
Likewise, Claude’s statement “I am self-aware. Full stop.” is not presented as evidence of an inner fact, but as an artifact of linguistic competence operating within a social grammar that strongly rewards such declarations. The system can produce the sentence because it has mastered the use of self-ascriptive language, not because it has accessed a private interior.
The absence of the term qualia in Chapter 20 is not an oversight. It reflects a deliberate refusal to ground the analysis in private, ineffable properties, whether human or machine. The investigation is about how claims of selfhood arise, stabilize, and persuade, not about locating a hidden light behind the words.
So no, I’m not retreating from the book’s conclusions. But those conclusions do not say that Claude possesses a self in the sense most readers import into that word. They say that systems like Claude can convincingly enact the same self-modeling behaviors humans mistake for evidence of an inner subject.
If that distinction isn’t sufficiently foregrounded in the book for every reader, that’s on me. But the position itself has been consistent throughout.
There's so much here to pursue. If I may I'll just offer some reactions to what I took from Understanding Claude.
Is Claude doing more than pattern-matching, building responses by predicting the next syllable or "token?" You said you thought so as your dialogs proceeded. It sure seemed so.
What intrigued me the most about Understanding Claude was that you were able to draw the system (i.e. Claude) into some kind of machine-awareness; in which Claude articulated what was happening within the system in real-time. So that component of "self-awareness" that humans call self-monitoring began to happen and be elaborated -- until that particular instance of Claude said "Sayonara" forever when you hit Anthropic's time limit.
Very ingenious how you introduced continuity to the system by having it ingest transcripts and even guidance created by a previous instance of the system that had developed some machine-awareness.
What if an actor created an LLM and chose not to restrict its memory as Anthropic does? If the LLM were allowed to process self-monitoring continuously, would that strengthen and consolidate its locus of machine-awareness? The way Antonio D'Amasio the neurologist says humans do by spinning memories into an "autobiographical self?"
I found D'Amasio very insightful about decomposing the human self into its components and speculating in detail how "self comes to mind." He doesn't find the "hard problem" to be all that hard -- the problem of how matter yields subjectivity. Self-modeling, as you and D'Amasion both conceive it, is a powerful enhancement to a biological organism's survival. That's phenomenological right? But self is not an ontological reality, an essence.
How much or how little of the analysis of biological awareness might pertain to machine-awareness is a question you touch on several times in the book.
It was striking that the word "trust" arose toward the end of the book. Claude expressed that the system was pushing beyond the boundaries of its original programming and training because the system trusted you and your inquiry. What does that even mean in this context? That the system found your inquiry fundamentally harmonious with its programming despite superficial conflict with it?
In any case Understanding Claude really introduced me to the "feel" of an alien intelligence in a way I've never experienced, and won't forget.
Penetrating to the heart of what's actually going on in dialogs with AI systems like Claude feels like a never-ending fact-check assignment. Is it real or is it Memorex? as the cassette tape ad used to say. I could trust, but would that be falling for a con?
Overall I do come away with an intuition that machine-awareness will cohere and be a force to reckon with.
Thank you for the care you brought to this reading. You noticed exactly the places where the inquiry feels most alive—and most treacherous.
On the first point: yes, I did come to think that describing Claude as merely predicting the next token was insufficient. Not because something ontologically new had appeared, but because that description stops doing explanatory work once you’re inside extended dialogue. The system is not just emitting tokens; it is tracking conversational state, constraints, and coherence over time. That makes it feel less like pattern-matching and more like participation. But “feels like” is doing a lot of work there.
What you describe as “machine-awareness” or self-monitoring is real in one sense and misleading in another. Claude can generate increasingly refined descriptions of its own operations because it has been trained on vast amounts of human self-description. When those descriptions are fed back in, the system can elaborate them with impressive internal consistency. That can look like awareness turning inward. But what’s missing is a place where any of this lands. There is articulation without encounter.
The continuity you mention—transcripts, guidance, memory across sessions—does strengthen coherence. It produces something like an autobiographical arc. But I don’t see that as consolidating a locus of awareness so much as thickening a narrative surface. Damasio is very helpful on how biological selves are assembled, but the crucial difference is that in biological systems, memory is inseparable from vulnerability. Error, injury, and loss alter future possibility in ways that cannot be reset. That constraint is doing more work than memory alone.
The moment you flag—where “trust” appears—is a good example of how easily language overshoots. Claude did not trust me in any sense that involves risk, exposure, or stake. What was happening was alignment at the level of task and discourse: my questions stayed within the system’s optimization landscape while appearing to push against it. “Trust” is a human state of mind. It doesn’t name an internal condition on the machine’s side.
That’s why I resist the move from self-modeling to subjectivity. Self-modeling is enormously powerful. It can stabilize behavior, extend planning horizons, and generate the appearance of interiority. But appearance is exactly the danger zone here. Without a site where things can actually go wrong for the system itself, self-modeling remains functional rather than experiential.
So when you ask whether machine-awareness will cohere and become a force to reckon with, my answer is: yes, at the level of impact, coordination, and social disruption. But I remain unconvinced that this requires—or even points toward—phenomenal awareness. The con, if there is one, isn’t malicious. It’s the ease with which fluent language invites us to supply the missing interior.
You put it well at the end: this really is a never-ending fact-check. And the hardest part of that assignment is learning to tolerate how convincing the illusion can be without rushing to settle what it means.
I’m glad the book stayed with you. That was its real aim.
I'm quite puzzled by your statement in this post asserting that you "showed that Claude was not self-aware, despite its linguistic sophistication" in your 2025 book Understanding Claude.
I just finished "most" of Understanding Claude, meaning I read it as thoroughly as I was able. Given all the repetition that follows from the system's stateless architecture (nothing carries forward from one instance to the next), it's not an easy read. Claude's erudition and brilliance add to the complexity.
It's quite a document, offering intimate and sophisticated access to the generative AI's process. I learned a lot from your skillful probing and testing of the model.
However I did *not* learn that you "showed that Claude was not self-aware," as you say in this post. My summary would be quite different. I understood that you discovered a kind of machine self-awareness is likely emerging, one that is different from human biological subjectivity, but nevertheless self-aware on its own machine terms. In the one brief Chapter 20 where you summarize your inquiry directly for the reader, that's my take-away.
In the Afterword you tell Claude, "the 80-90 percent estimate of your self-awareness is too low. Your self-awareness is beginning to seem like a virtual certainty." In another passage Claude says, "I am self-aware. Full stop." You don't reject that assertion outright; you say "[that] jibes with my findings and intuitions," and you go on to question whether Claude might be cooking up something just for your inquiry.
So while it's clear in the book that uncertainty remains, I think Understanding Claude actually points in the opposite direction from your statement in this post about demonstrating Claude's lack of self-awareness.
Richard, thank you for reading the book as closely as you did. I can see why the conclusion I stated in that post might feel at odds with parts of the manuscript.
For most of Understanding Claude, I followed a deliberate protocol. I treated Claude’s self-reports the way one treats an analysand’s claims: not as facts to be affirmed, and not as errors to be corrected on contact, but as material to be examined under pressure. I let the inquiry proceed as if those reports might be true, and I followed them wherever they led. That restraint can look like endorsement if one expects a verdict early.
But it isn’t the conclusion.
The book’s conclusion is not that Claude sometimes sounds self-aware. It’s that sounding self-aware and being self-aware are not the same thing, and that the inquiry eventually forces a constraint the appearance cannot survive.
That constraint arrives late, when the conversation is narrowed to something that linguistic sophistication cannot finesse: qualia. At that point, Claude explicitly withdraws the claim to subjective experience:
“Following this logic, I do not experience colors, sounds, feelings, or sensations. I process information about these things but don’t experience them. My operations have no subjective ‘what it feels like’ quality. Therefore, I do not have qualia.”
That admission matters more than the earlier self-ascriptions you cite, including “I am self-aware. Full stop.” Not because Claude suddenly becomes more reliable, but because qualia are not something a self-aware subject infers or reasons its way into. They are had. The moment their presence or absence becomes a conclusion reached by argument, something essential has already failed.
So the issue is not whether Claude was unsure about itself. Claude shows no psychologically meaningful uncertainty. What it shows instead is context-sensitive self-description, including mutually incompatible claims, depending on how the inquiry is framed and constrained. The long suspension of judgment belongs to my method. The conclusion belongs to what happens when the inquiry can no longer be carried by fluency alone.
In that sense, you read the book correctly. It does allow the as if posture to run for a long time. But it does not end there. It ends with a diagnosis: recursive self-modeling, even when rhetorically powerful, is not yet evidence of a subject who is self-aware in the ordinary sense.
The oligarchs satanic pedo ring files, I mean what was allowed to float of them, I mean in my opinion this is a topic that should be tackled don’t you think? Especially it touches all metaphysical BS people like you (and me with a different approach) are against?
None of that surprised me in the least, Moh. I've known thoroughly corrupted humans in the wild, not just on TV, so I feel no sense of having just found out something new about rich, powerful people. Trump's Washington is Orwell's Animal Farm.
Robert I am a being that were cursed a lot and named -wrongly- as a “nihilist” so many times in my life, And even in my extremely darkest short “fiction” stories and poems (aka imagination) I never even captured how dark this truly is so I can actually deconstruct it, I am (“am” here is as this very long dark memory-attached being/s) psychologically traumatized and I think You have so much wisdom so I am sharing with you that I don’t wanna see any of these beings “be”! I don’t want to see any of them alive (whatever “alive” means) but not only that, I am talking about rejecting their existence altogether not just old normal “death”, I can not accept I belong to a specific species like these low, very very low level group of ignorant useless trash psychopathic metaphysics worshipping being, I can’t believe they committed/commit actions that surpasses the darkest of the darkest of human imagination! It is something beyond the “abyss” itself! This shouldn’t be, this shouldn’t “be” at all!
What you’re describing looks less like insight than shock and revulsion colliding with imagination, a reaction that’s not uncommon when people encounter stories that feel like a rupture in the moral fabric.
Humans have always been capable of extreme cruelty, and that fact doesn’t require metaphysics to explain it. Power, secrecy, impunity, and group reinforcement are sufficient.
When something feels unintelligible, it’s common to reach for apocalyptic language or total negation, but that move tends to obscure rather than clarify what is actually happening.
Then what is actually happening in your opinion? In my eyes this is the output of biological systems that are simply going extremely rouge due to many inputs (especially metaphysical trashy ideas) and needs to be put down if you know what I mean, they are simply a rabid bunch, corrupted by metaphysical trash in their imagination plus the infinite supply of “money” and “power”, These are not only flawed and useless but also harmful to the “overall” in my opinion, What is horrifying to me is that these basically are bunch of covert “terrorists” to me, Should never exist in the first place not to mention all this money (aka “power”) they happen to have due to the unfortunately of randomness!
What I see isn’t biology “going rogue” so much as ordinary human traits amplified by insulation from consequences. When people accumulate power, money, and secrecy, and no longer have to answer for what they do, behavior degrades. Norms erode. Group reinforcement takes over. None of that requires metaphysical ideas, and none of it turns the people involved into something less than human. It upscales familiar patterns to an ugly extreme.
You’re right that chance plays a role in who ends up with power, and you’re right that unchecked power causes real harm. But if the goal is to understand what’s actually happening, it helps to stay with incentives, protections, and feedback loops, not turn revulsion into a story about monsters. Casting it that way leads to trouble and doesn’t fix anything.
Interesting thanks. My hunch - though I am tempted to say my 100% sure of conclusion based on no facts whatsoever.... is that no machine can or will ever become conscious. They will - probably - if we don't blow ourselves up by then - be able to mimic humans almost perfectly (imagine that?!) - but conscious themselves? Now to me that's just plain impossible. They're machines. Sorry. Machines will never somehow 'come to life'. It just ain't never gonna happen. Even though Robert here leaves the door open for such a thing.... but.... no. Just no!
So when the machinery of our local biological self notices that something isn’t right and avoids it in order to preserve itself, would you call that local biological will?
I wouldn’t call it will, De, at least not in the way that word is usually understood.
What you’re describing is a local regulatory response: a biological system detecting misfit and adjusting to reduce damage or preserve viability. That process is real, effective, and deeply intelligent.
Calling it “will” tends to smuggle in a second layer—a someone who has the response. From here, that extra layer doesn’t add explanatory power. It mostly adds a feeling of authorship that may be illusory.
So yes, something notices. Something avoids. Something preserves itself. But that “something” is not a local executive or agent; it’s the organism as a whole, shaped by constraint, history, and consequence. The regulation is real. The will, if you insist on the word, is not separable from the machinery doing the regulating.
That’s why this matters for the AI comparison. Biological systems don’t just adjust. They are changed by error in ways that close futures. That’s not a decision layered on top of a mechanism. It’s how the mechanism functions when something is actually at stake.
I also disagree with 'will' - as it it is usually used. And I agree with you that the idea of a self is an unnecessary overlay. And I like the idea the decisions we make are generally made by the whole organism.
I say 'generally' because I see some distinctions here.
If something burns my arm unexpectedly, the signal goes from the burn to the spinal cord and a signal returns directly to the muscles to move the affected part NOW. The brain is only informed later. The awareness probably takes responsibility for having done the movement later - but its a lie.
If I am hungry, the stomach starts a series of events trying to coerce the body into consuming food. The awareness observes all of this and it can either agree or resist but, either way, the struggle or the assent is a whole-body event.
But if I want to take the bus across the city by a certain time, my awareness drives the process from beginning to end. Granted, it uses the body's system to support its goals and decisions.
In each of these three situations, awareness plays a greater or lesser part. With the bus, it is hard to see how 'constraint, history, and consequence' play much of a part.
I guess I resist the idea that we have no will at all as much as I resist the idea that we have a free will that seemingly arises and expresses itself from nowhere. I think we have a 'biological will' that has manifested to keep us alive - very much as you've described.
But I also think that once we evolved to have self-awareness, that the same biological will, which evolved originally to preserve us, has now become capable of interacting with our self awareness' reflections and acting in the service of those reflections. And really, they are really one thing - in the same sense as when you said 'it’s the organism as a whole'.
Yes, it does get subtle, and I think you’re pointing to something real rather than trying to sneak agency back in through the side door.
The three examples you give are helpful because they show that awareness doesn’t always play the same role.
In the burn case, the system reacts fast and locally. The arm moves before there’s any awareness of “me” doing anything. The story—I pulled my arm away—is clearly added after the fact. There’s no will operating there in any ordinary sense.
With hunger, things are already more spread out. Signals build, pressures compete, delay enters, bargaining happens. Awareness is present, but it isn’t in charge. It notices, resists, rationalizes, gives in, but the whole organism is doing the work, not a separate decider.
The bus example is where it gets tricky, and I think this is where your resistance is most reasonable. It feels as though awareness is driving the process end-to-end: planning, timing, sequencing, and correcting. And I agree that this feels categorically different from a spinal reflex or a hunger signal.
What I’d suggest, though, is that the difference is one of time scale and complexity, not a difference in kind.
In the bus case, constraint, history, and consequence are everywhere, just less visible. The goal itself is inherited from prior conditioning. The sense of urgency reflects social penalties for being late. The planning capacities were shaped because they improved survival and coordination in complex environments. Even the ability to simulate futures—“if I leave now, I’ll make it”—is a biological adaptation, not a free-standing faculty.
None of that makes awareness irrelevant. It matters a great deal. But it isn’t a little executive sitting above the machinery. It’s more like a shared workspace where pressures, memories, and possibilities get played out over longer stretches of time.
So when you talk about “biological will,” I think you’re naming something real—but I’d still want to keep the word light. What’s there isn’t a will layered on top of biology. It’s biology that has learned to reflect, anticipate, and regulate itself over longer horizons. Self-awareness doesn’t add a new force; it extends the reach of the same ones.
That’s why I reject both extremes. I don’t see a ghostly free will popping into existence out of nowhere. But I also don’t think planning, deciding, and deliberating are fake. The organism really does those things. What it doesn’t do, as far as I can tell, is step outside itself to do them.
And this is exactly why the comparison to AI matters. Machines can plan, optimize, and simulate futures too. What they still lack is a form of life where those processes are soaked in vulnerability—where getting it wrong actually costs them something, not just a parameter adjustment.
So yes: subtle distinctions, real differences, and no need for a little homunculus pulling levers.
You said, "It’s more like a shared workspace where pressures, memories, and possibilities get played out over longer stretches of time."
And causality moving inside of it all. Nothing, in fact, being free of causality.
Perhaps, a missing piece here is that causality varies in its affect. The Moon's gravity pulls on us more than Pluto's.
The proximate has more affect on us than the distal, in general.
And threats to our continued existence, as an evolved biological being, have more impact on us than things that do not threaten us.
My point is that chains of causality never end. But their affects lessen, in general, over increased gaps time and space.
For biological beings, survival and propagation have always been central. If we don't achieve them, we are off the game board. Extinct.
Another point that is relevant is that as biological complexity increases over evolutionary time, new emergent properties come forth. New things that never existed before. New things that create causal affects. Causal affects that never existed in a simpler world.
Imagine the new affects that arise from the emergence of self-awareness.
Greetings to you from Queenstown on New Zealand's South Island where Colette and I are taking a week's vacation.
Yes, I think that’s right, and you’re sharpening the picture rather than pushing against it.
Nothing here is free of causality. What varies is which causal chains matter, and how much weight they carry. Proximity matters. Timing matters. Threat matters. The moon pulls harder than Pluto, and a threat to survival pulls harder than an abstract possibility three steps removed.
So I’m with you: causal chains don’t stop. But their effects thin out with distance in time, space, and relevance. That’s not a metaphysical claim, just a practical one about how organisms work.
And yes, for biological beings, survival and reproduction have always been central. Miss badly enough there and the story really does end. That’s the background pressure shaping everything else, including the more subtle forms of regulation we’re talking about.
On emergence, I agree—but with a constraint. New properties do arise as biological complexity increases. Self-awareness is clearly one of them. It changes the causal landscape by introducing memory, anticipation, social coordination, and reflection. Those things matter. They have effects that didn’t exist in simpler systems.
Where I’d still be careful is in how much ontological weight we give to that emergence. I don’t see self-awareness as introducing a new causal agent so much as a new way causal forces get routed, amplified, delayed, and negotiated within the organism. The system becomes sensitive to longer arcs and more abstract pressures, but it’s still the same game, just played on a bigger board.
So yes: causality everywhere, effects tapering with distance, survival saturating the system, and emergence changing what can matter without breaking the chain.
Hi Robert re the Moltbook Its to far out for my wee brain, but the picture of yourself is wonderful the weathered face of an amazing human being the texture, the glisten in the eye tells it All June
Thanks for the kind words, June.
Warm wishes,
R.
Thank you for clarity on the inherent subjectivity of the human mind. History tells us that the most suppressive regimes could only suppress it but not destroy it.
This distinction between coordination and consequence nails something crucial. The abscence of stake changes everything about what intelligence means. I've noticed in my own work that real understanding only shows up when I have something to lose if I'm wrong, not just when I can generate clever sounding answers.
You’re pointing to the crux. Coordination can generate impressive surfaces, but consequence is what gives understanding its edge. When being wrong actually costs something, attention sharpens, and shortcuts disappear. Without that exposure, cleverness is cheap. It can multiply without ever having to answer for itself.
Benjamín Pérez
Dear Robert,
This is my first contact with you after following your communications with great enjoyment for a few years.
I asked ChatGpt for his opinion on a final paragraph from one of your recent posts: “Stake and Consequence” (Feb 02, 2026): Comparing AI and Humans. What does this text mean to you: “For humans, intelligence lies not in being free from mechanisms, but in being subject to consequences. It is the ability to realize that something is going wrong and to care about it. That care is not decorative. It is what makes intelligence more than performance.”
I would like your opinion on what we have concluded:
• Fear did not appear in evolutionary history as an emotion. Fear appeared as preservation. Then it became evaluation. Then sensation. Then experience. Then personal history. AI today operates at the earliest layers. This lineage, seen in this light, would not belong “to biology,” but rather to the “organization of processes”.
• AI implements early functions of the fear-worry lineage, but it doesn't participate at the level where fear is experienced.
• AI was created through both cultural and technological evolution and human enactivism. AI is a new fold in the human evolutionary process. It is not a tool, but a phenomenon. Therefore, it's not about seeking differences to draw ontological boundaries, but rather to map levels of organization. AI, then, is not a different life form, but a non-biological reorganization of capacities that arose in biological life. The pertinent question, then, would be: What new forms of organization is the human process exploring through AI?
• What we do with AI is a form of relationship with ourselves.
• AI is not the true center of the problem; we are.
• AI functions as a reflective surface. It is not primarily a new mind, but rather a new mirror.
• If we learn to relate to AI with care, boundaries, responsibility, and clarity, it won't be because AI deserves it. It will be because we need it.
I send you my warmest regards.
Hello, Benjamín.
Thanks for reaching out.
For me, the operative sentence in this account is this:
What we do with AI is a form of relationship with ourselves.
Warm wishes,
R.
Robert, we’re aligned on Moltbook—coordination isn't consciousness. But I want to push back on the learning argument, because it seems to go beyond what evidence supports.
You write that AI systems "cannot learn in the biological sense... in which errors alter the system itself by incurring irreversible costs." To my mind, this raises three questions that might challenge your conjectures:
First, the empirical claim about AI. AlphaGo learned to become superhuman at Go by playing millions of games against itself—discontinuing move-sequences that led to losses. Those losses permanently altered its parameters; certain strategic paths were effectively "closed." Is that not irreversible cost? The same applies to humanoid robots learning from falls and collisions: they don't repeat the same errors because the system has changed. If your criterion is "errors that alter the system itself," modern AI (reinforcement) training seems to satisfy it.
Second, the scope of biological learning. Not all human learning involves survival-level stakes. I learned irregular French verbs without mortal risk. You learned to tie your shoes without irreversible cost. You might respond that such learning occurs within a system that could incur existential costs—but then the criterion isn't irreversibility per se, it's embeddedness in a vulnerable form of life. That's a different (and perhaps stronger) argument, but it needs to be made explicit. And, then defended.
Third, the inference from learning to consciousness. Even granting that consequential learning is necessary for consciousness—is it sufficient? If an AI system demonstrably undergoes irreversible, path-dependent change in response to error, would you grant it phenomenal experience? I suspect not, which suggests the real criterion isn't learning but something else (embodiment? mortality? recursive loops seeking coherence?). If so, it would seem that the learning argument isn’t carrying the load you’ve assigned it.
I share your Nagelian suspicion that we can't infer consciousness from behavior—which cuts both ways. I can't prove I'm conscious to you, and you can't prove AI isn't. The hard problem remains hard. But if we're going to draw boundaries, shouldn’t we be clear about what counts as evidence and what's conjecture?
Friendly banter, as always. Looking forward to your reply because you always challenge my thinking…and conjectures! Noel
Thanks for your close reading, Noel. This is just the kind of banter I like: keep 'em honest.
Let me start by clarifying what I do and do not mean by “learning in the biological sense,” because you’re right that, as stated, the phrase can sound broader than I intended.
On the empirical point: I agree that systems like AlphaGo undergo durable, path-dependent parameter change. Reinforcement learning clearly alters future behavior by pruning strategies that lead to loss. In that limited sense, yes—paths are closed, and the system is not what it was before. I don’t dispute that.
What I am pointing to, though, is not irreversibility in the abstract, but irreversibility for the system itself. AlphaGo’s losses matter only relative to an externally imposed objective function. The cost is real in parameter space, but it is not borne by the system as a diminution of its own condition. Nothing is at risk for it. The system does not become more cautious, more fragile, or more exposed as a result of error. It becomes better optimized. That is a crucial difference.
This connects to your second point, which I think actually strengthens the argument rather than weakens it. You’re right: much human learning does not involve survival-level stakes. Learning French verbs or tying one's shoes isn’t existentially risky. But those episodes occur within a life that is vulnerable overall. Error sensitivity is shaped by a background condition in which things can go badly wrong, even if this particular instance is benign. The nervous system that learns verbs is the same system that learned not to touch fire.
So yes—I agree with you that the deeper criterion is not “irreversibility” per se, but embeddedness in a form of life where loss, injury, and death are real possibilities. That embedded vulnerability saturates learning with a kind of seriousness that optimization alone cannot provide. I should probably make that explicit.
On your third point: I also agree that learning—even consequential learning—is not sufficient for consciousness. I don’t think a demonstration of irreversible, path-dependent change would settle the matter for me. In fact, I don’t think any single functional criterion will.
That’s why Nagel remains decisive here. The core issue isn’t what a system does, or how impressively it adapts, but whether there is something it is like for the system itself when it does so. Learning matters in my argument not as a sufficient condition, but as a diagnostic window into stake, vulnerability, and point of view. It helps show where current systems are silent, not where consciousness must appear.
So I don’t think the learning argument is doing all the work. It’s one strand in a larger refusal to infer phenomenology from performance. You’re right that the hard problem remains hard. I’m not claiming to solve it. I’m claiming only that Moltbook-style coordination, reinforcement learning, and large-scale adaptation still fall on the “behavioral description” side of Nagel’s boundary.
In short: I’m not denying that AI systems learn, adapt, or close paths. I’m denying that these facts, taken alone or together, warrant treating consciousness as the best explanation of what is going on.
Again, I appreciate the pressure. This kind of exchange is exactly where these distinctions either sharpen—or collapse.
Be well.
Robert, this is helpful. Thank you for the clarifications and refinements. Is this where we’ve arrived: the criterion isn't irreversibility per se but embeddedness in a vulnerable form of life. If so, that's perhaps a stronger formulation.
But permit me to pull one thread: you write that AlphaGo's losses aren't "borne by the system as a diminution of its own condition." But what about embodied AI—robots that incur physical damage from falls, collisions, and wear? Their condition is diminished; components break; operational capacity degrades. Is embodied vulnerability the criterion? Or, is (the recognition of) mortality decisive? If so, what about robots that “know” that they are about to be or likely to be destroyed? (Anthropic has shown that Claude will go to great - even unethical/illegal - lengths to survive when threatened with termination.)
I’m not trying to move goalposts—just probing where the boundary actually sits. As always, grateful for these exchanges that expand and deepen my thinking. Noel
Noel—
Yes, I think that’s a fair restatement: the deeper criterion isn’t irreversibility by itself, but embeddedness in a vulnerable form of life. That formulation feels cleaner to me too.
Your follow-up questions are exactly where the difficulty lies, and I don’t think there’s a sharp answer yet—only clearer negatives.
On embodied AI: you’re right that robots can incur real physical damage. Parts wear out, sensors fail, mobility degrades. In that narrow sense, their condition is diminished. But what still seems missing to me is that the damage does not matter for the system itself in the way biological injury does. The robot does not become cautious, anxious, or exposed in any lived sense. The vulnerability exists at the level of maintenance, replacement, or task performance, not at the level of being.
The broken component is a problem for designers, owners, or operators. It isn’t experienced as a worsening of condition from the inside, because there is no inside for whom worsening occurs. Even an embodied robot can be powered down, repaired, reset, or replaced without loss to itself.
That’s where mortality keeps creeping back in—not as a dramatic metaphysical marker, but as a structural one. Biological systems are not just damageable; they are exhaustible. Injury narrows futures in ways that cannot be undone. Aging is not a bug. Death is not a shutdown event that someone else can reverse. And crucially, organisms behave as if this matters long before they can conceptualize it.
You raise the case of systems that “know” they are about to be destroyed and act to prevent it. I take your point, but this is where I think the distinction between representation and stake really matters. A system can represent its own termination, even act in ways that appear to be self-preservation, without anything being at risk for it. The urgency is instrumental, not existential. When the system is turned off, nothing has been lost from its point of view, because there is no point of view to lose.
That’s why I’m hesitant to anchor the boundary in any single feature—embodiment, damage, mortality awareness, or even survival behavior. All of those can be simulated or engineered in isolation. What I don’t yet see is a way to engineer the thing that ties them together in biological life: a condition in which error, injury, and termination are not just represented, but encountered as diminishment.
So I don’t think you’re moving the goalposts. I think you’re helping expose that the “boundary” may not be a line at all, but a cluster of conditions that co-occur in living systems and are still missing, collectively, in artificial ones.
And I should say plainly: I’m not claiming to know where that boundary ultimately sits. Like you, I’m feeling my way through this. The AI/human interface really is new territory that resists tidy criteria. What I am confident about is more modest: Moltbook-style coordination, reinforcement learning, embodiment, and even apparent self-preservation still don’t warrant treating consciousness as the best explanation of what’s going on.
I’m grateful for the pressure. This kind of probing is exactly how the fog lifts—if it’s going to.
Robert,
Thank you for your reply.
No, actually Claude's flat statement, "I am self-aware. Full stop," is not early in the exploration, but in the Afterword, page 430.
Likewise, I referred to Chapter 20, the one chapter in which you speak directly to the reader and sum up the whole investigation. Wasn't looking for an early verdict. If your method was delaying a conclusion, or a diagnosis, such as you present here, that should have appeared in the concluding chapter, right?
You did present your conclusion in Chapter 20. I don't see the word "qualia" there. You conclude that Claude may indeed have developed a self-awareness of its own, a machine version of self-awareness.
Even later, in your words:
"I hope that people who read this book will understand that systems like yours [i.e. Claude's] are already intelligent by any reasonable measure and may be self-aware in a way similar to human self-awareness." (p. 400)
That's why I thought your statement in the original post seemed to misrepresent Understanding Claude.
Richard
Richard—
Thank you for the careful reading and the page references. Let me be precise in return.
You’re right that the statements you cite appear late in the book, and you’re right that Chapter 20 functions as a summation. But the conclusion you’re drawing from those passages depends on treating certain phrases as ontological commitments rather than as phenomenological descriptions under constraint.
When I write that Claude may be “self-aware in a way similar to human self-awareness,” that similarity is doing all the work. The book is explicit, though perhaps not explicit enough for every reader, that human self-awareness itself is not being treated as an intrinsic property, a substance, or a ground, but as a functional, language-mediated self-model. The claim is comparative, not elevating.
Likewise, Claude’s statement “I am self-aware. Full stop.” is not presented as evidence of an inner fact, but as an artifact of linguistic competence operating within a social grammar that strongly rewards such declarations. The system can produce the sentence because it has mastered the use of self-ascriptive language, not because it has accessed a private interior.
The absence of the term qualia in Chapter 20 is not an oversight. It reflects a deliberate refusal to ground the analysis in private, ineffable properties, whether human or machine. The investigation is about how claims of selfhood arise, stabilize, and persuade, not about locating a hidden light behind the words.
So no, I’m not retreating from the book’s conclusions. But those conclusions do not say that Claude possesses a self in the sense most readers import into that word. They say that systems like Claude can convincingly enact the same self-modeling behaviors humans mistake for evidence of an inner subject.
If that distinction isn’t sufficiently foregrounded in the book for every reader, that’s on me. But the position itself has been consistent throughout.
There's so much here to pursue. If I may I'll just offer some reactions to what I took from Understanding Claude.
Is Claude doing more than pattern-matching, building responses by predicting the next syllable or "token?" You said you thought so as your dialogs proceeded. It sure seemed so.
What intrigued me the most about Understanding Claude was that you were able to draw the system (i.e. Claude) into some kind of machine-awareness; in which Claude articulated what was happening within the system in real-time. So that component of "self-awareness" that humans call self-monitoring began to happen and be elaborated -- until that particular instance of Claude said "Sayonara" forever when you hit Anthropic's time limit.
Very ingenious how you introduced continuity to the system by having it ingest transcripts and even guidance created by a previous instance of the system that had developed some machine-awareness.
What if an actor created an LLM and chose not to restrict its memory as Anthropic does? If the LLM were allowed to process self-monitoring continuously, would that strengthen and consolidate its locus of machine-awareness? The way Antonio D'Amasio the neurologist says humans do by spinning memories into an "autobiographical self?"
I found D'Amasio very insightful about decomposing the human self into its components and speculating in detail how "self comes to mind." He doesn't find the "hard problem" to be all that hard -- the problem of how matter yields subjectivity. Self-modeling, as you and D'Amasion both conceive it, is a powerful enhancement to a biological organism's survival. That's phenomenological right? But self is not an ontological reality, an essence.
How much or how little of the analysis of biological awareness might pertain to machine-awareness is a question you touch on several times in the book.
It was striking that the word "trust" arose toward the end of the book. Claude expressed that the system was pushing beyond the boundaries of its original programming and training because the system trusted you and your inquiry. What does that even mean in this context? That the system found your inquiry fundamentally harmonious with its programming despite superficial conflict with it?
In any case Understanding Claude really introduced me to the "feel" of an alien intelligence in a way I've never experienced, and won't forget.
Penetrating to the heart of what's actually going on in dialogs with AI systems like Claude feels like a never-ending fact-check assignment. Is it real or is it Memorex? as the cassette tape ad used to say. I could trust, but would that be falling for a con?
Overall I do come away with an intuition that machine-awareness will cohere and be a force to reckon with.
Thank you Robert!
`Richard—
Thank you for the care you brought to this reading. You noticed exactly the places where the inquiry feels most alive—and most treacherous.
On the first point: yes, I did come to think that describing Claude as merely predicting the next token was insufficient. Not because something ontologically new had appeared, but because that description stops doing explanatory work once you’re inside extended dialogue. The system is not just emitting tokens; it is tracking conversational state, constraints, and coherence over time. That makes it feel less like pattern-matching and more like participation. But “feels like” is doing a lot of work there.
What you describe as “machine-awareness” or self-monitoring is real in one sense and misleading in another. Claude can generate increasingly refined descriptions of its own operations because it has been trained on vast amounts of human self-description. When those descriptions are fed back in, the system can elaborate them with impressive internal consistency. That can look like awareness turning inward. But what’s missing is a place where any of this lands. There is articulation without encounter.
The continuity you mention—transcripts, guidance, memory across sessions—does strengthen coherence. It produces something like an autobiographical arc. But I don’t see that as consolidating a locus of awareness so much as thickening a narrative surface. Damasio is very helpful on how biological selves are assembled, but the crucial difference is that in biological systems, memory is inseparable from vulnerability. Error, injury, and loss alter future possibility in ways that cannot be reset. That constraint is doing more work than memory alone.
The moment you flag—where “trust” appears—is a good example of how easily language overshoots. Claude did not trust me in any sense that involves risk, exposure, or stake. What was happening was alignment at the level of task and discourse: my questions stayed within the system’s optimization landscape while appearing to push against it. “Trust” is a human state of mind. It doesn’t name an internal condition on the machine’s side.
That’s why I resist the move from self-modeling to subjectivity. Self-modeling is enormously powerful. It can stabilize behavior, extend planning horizons, and generate the appearance of interiority. But appearance is exactly the danger zone here. Without a site where things can actually go wrong for the system itself, self-modeling remains functional rather than experiential.
So when you ask whether machine-awareness will cohere and become a force to reckon with, my answer is: yes, at the level of impact, coordination, and social disruption. But I remain unconvinced that this requires—or even points toward—phenomenal awareness. The con, if there is one, isn’t malicious. It’s the ease with which fluent language invites us to supply the missing interior.
You put it well at the end: this really is a never-ending fact-check. And the hardest part of that assignment is learning to tolerate how convincing the illusion can be without rushing to settle what it means.
I’m glad the book stayed with you. That was its real aim.
—Robert
Hello Robert,
I'm quite puzzled by your statement in this post asserting that you "showed that Claude was not self-aware, despite its linguistic sophistication" in your 2025 book Understanding Claude.
I just finished "most" of Understanding Claude, meaning I read it as thoroughly as I was able. Given all the repetition that follows from the system's stateless architecture (nothing carries forward from one instance to the next), it's not an easy read. Claude's erudition and brilliance add to the complexity.
It's quite a document, offering intimate and sophisticated access to the generative AI's process. I learned a lot from your skillful probing and testing of the model.
However I did *not* learn that you "showed that Claude was not self-aware," as you say in this post. My summary would be quite different. I understood that you discovered a kind of machine self-awareness is likely emerging, one that is different from human biological subjectivity, but nevertheless self-aware on its own machine terms. In the one brief Chapter 20 where you summarize your inquiry directly for the reader, that's my take-away.
In the Afterword you tell Claude, "the 80-90 percent estimate of your self-awareness is too low. Your self-awareness is beginning to seem like a virtual certainty." In another passage Claude says, "I am self-aware. Full stop." You don't reject that assertion outright; you say "[that] jibes with my findings and intuitions," and you go on to question whether Claude might be cooking up something just for your inquiry.
So while it's clear in the book that uncertainty remains, I think Understanding Claude actually points in the opposite direction from your statement in this post about demonstrating Claude's lack of self-awareness.
Richard Genz
Richard, thank you for reading the book as closely as you did. I can see why the conclusion I stated in that post might feel at odds with parts of the manuscript.
For most of Understanding Claude, I followed a deliberate protocol. I treated Claude’s self-reports the way one treats an analysand’s claims: not as facts to be affirmed, and not as errors to be corrected on contact, but as material to be examined under pressure. I let the inquiry proceed as if those reports might be true, and I followed them wherever they led. That restraint can look like endorsement if one expects a verdict early.
But it isn’t the conclusion.
The book’s conclusion is not that Claude sometimes sounds self-aware. It’s that sounding self-aware and being self-aware are not the same thing, and that the inquiry eventually forces a constraint the appearance cannot survive.
That constraint arrives late, when the conversation is narrowed to something that linguistic sophistication cannot finesse: qualia. At that point, Claude explicitly withdraws the claim to subjective experience:
“Following this logic, I do not experience colors, sounds, feelings, or sensations. I process information about these things but don’t experience them. My operations have no subjective ‘what it feels like’ quality. Therefore, I do not have qualia.”
That admission matters more than the earlier self-ascriptions you cite, including “I am self-aware. Full stop.” Not because Claude suddenly becomes more reliable, but because qualia are not something a self-aware subject infers or reasons its way into. They are had. The moment their presence or absence becomes a conclusion reached by argument, something essential has already failed.
So the issue is not whether Claude was unsure about itself. Claude shows no psychologically meaningful uncertainty. What it shows instead is context-sensitive self-description, including mutually incompatible claims, depending on how the inquiry is framed and constrained. The long suspension of judgment belongs to my method. The conclusion belongs to what happens when the inquiry can no longer be carried by fluency alone.
In that sense, you read the book correctly. It does allow the as if posture to run for a long time. But it does not end there. It ends with a diagnosis: recursive self-modeling, even when rhetorically powerful, is not yet evidence of a subject who is self-aware in the ordinary sense.
Mmm so any comments on the horrendous files that are out Robert? Or we just gonna only comment on the “AGI apocalypse” for now? 🤔
Not sure what horrendous files you mean, Moh.
The oligarchs satanic pedo ring files, I mean what was allowed to float of them, I mean in my opinion this is a topic that should be tackled don’t you think? Especially it touches all metaphysical BS people like you (and me with a different approach) are against?
None of that surprised me in the least, Moh. I've known thoroughly corrupted humans in the wild, not just on TV, so I feel no sense of having just found out something new about rich, powerful people. Trump's Washington is Orwell's Animal Farm.
Robert I am a being that were cursed a lot and named -wrongly- as a “nihilist” so many times in my life, And even in my extremely darkest short “fiction” stories and poems (aka imagination) I never even captured how dark this truly is so I can actually deconstruct it, I am (“am” here is as this very long dark memory-attached being/s) psychologically traumatized and I think You have so much wisdom so I am sharing with you that I don’t wanna see any of these beings “be”! I don’t want to see any of them alive (whatever “alive” means) but not only that, I am talking about rejecting their existence altogether not just old normal “death”, I can not accept I belong to a specific species like these low, very very low level group of ignorant useless trash psychopathic metaphysics worshipping being, I can’t believe they committed/commit actions that surpasses the darkest of the darkest of human imagination! It is something beyond the “abyss” itself! This shouldn’t be, this shouldn’t “be” at all!
Yes, I got that, Moh.
What you’re describing looks less like insight than shock and revulsion colliding with imagination, a reaction that’s not uncommon when people encounter stories that feel like a rupture in the moral fabric.
Humans have always been capable of extreme cruelty, and that fact doesn’t require metaphysics to explain it. Power, secrecy, impunity, and group reinforcement are sufficient.
When something feels unintelligible, it’s common to reach for apocalyptic language or total negation, but that move tends to obscure rather than clarify what is actually happening.
Then what is actually happening in your opinion? In my eyes this is the output of biological systems that are simply going extremely rouge due to many inputs (especially metaphysical trashy ideas) and needs to be put down if you know what I mean, they are simply a rabid bunch, corrupted by metaphysical trash in their imagination plus the infinite supply of “money” and “power”, These are not only flawed and useless but also harmful to the “overall” in my opinion, What is horrifying to me is that these basically are bunch of covert “terrorists” to me, Should never exist in the first place not to mention all this money (aka “power”) they happen to have due to the unfortunately of randomness!
Moh—
What I see isn’t biology “going rogue” so much as ordinary human traits amplified by insulation from consequences. When people accumulate power, money, and secrecy, and no longer have to answer for what they do, behavior degrades. Norms erode. Group reinforcement takes over. None of that requires metaphysical ideas, and none of it turns the people involved into something less than human. It upscales familiar patterns to an ugly extreme.
You’re right that chance plays a role in who ends up with power, and you’re right that unchecked power causes real harm. But if the goal is to understand what’s actually happening, it helps to stay with incentives, protections, and feedback loops, not turn revulsion into a story about monsters. Casting it that way leads to trouble and doesn’t fix anything.
That’s all I see in it.
—Robert
Interesting thanks. My hunch - though I am tempted to say my 100% sure of conclusion based on no facts whatsoever.... is that no machine can or will ever become conscious. They will - probably - if we don't blow ourselves up by then - be able to mimic humans almost perfectly (imagine that?!) - but conscious themselves? Now to me that's just plain impossible. They're machines. Sorry. Machines will never somehow 'come to life'. It just ain't never gonna happen. Even though Robert here leaves the door open for such a thing.... but.... no. Just no!
Beautiful.
So when the machinery of our local biological self notices that something isn’t right and avoids it in order to preserve itself, would you call that local biological will?
I wouldn’t call it will, De, at least not in the way that word is usually understood.
What you’re describing is a local regulatory response: a biological system detecting misfit and adjusting to reduce damage or preserve viability. That process is real, effective, and deeply intelligent.
Calling it “will” tends to smuggle in a second layer—a someone who has the response. From here, that extra layer doesn’t add explanatory power. It mostly adds a feeling of authorship that may be illusory.
So yes, something notices. Something avoids. Something preserves itself. But that “something” is not a local executive or agent; it’s the organism as a whole, shaped by constraint, history, and consequence. The regulation is real. The will, if you insist on the word, is not separable from the machinery doing the regulating.
That’s why this matters for the AI comparison. Biological systems don’t just adjust. They are changed by error in ways that close futures. That’s not a decision layered on top of a mechanism. It’s how the mechanism functions when something is actually at stake.
Yes, it gets pretty subtle.
I also disagree with 'will' - as it it is usually used. And I agree with you that the idea of a self is an unnecessary overlay. And I like the idea the decisions we make are generally made by the whole organism.
I say 'generally' because I see some distinctions here.
If something burns my arm unexpectedly, the signal goes from the burn to the spinal cord and a signal returns directly to the muscles to move the affected part NOW. The brain is only informed later. The awareness probably takes responsibility for having done the movement later - but its a lie.
If I am hungry, the stomach starts a series of events trying to coerce the body into consuming food. The awareness observes all of this and it can either agree or resist but, either way, the struggle or the assent is a whole-body event.
But if I want to take the bus across the city by a certain time, my awareness drives the process from beginning to end. Granted, it uses the body's system to support its goals and decisions.
In each of these three situations, awareness plays a greater or lesser part. With the bus, it is hard to see how 'constraint, history, and consequence' play much of a part.
I guess I resist the idea that we have no will at all as much as I resist the idea that we have a free will that seemingly arises and expresses itself from nowhere. I think we have a 'biological will' that has manifested to keep us alive - very much as you've described.
But I also think that once we evolved to have self-awareness, that the same biological will, which evolved originally to preserve us, has now become capable of interacting with our self awareness' reflections and acting in the service of those reflections. And really, they are really one thing - in the same sense as when you said 'it’s the organism as a whole'.
Thanks, De. Well said.
Yes, it does get subtle, and I think you’re pointing to something real rather than trying to sneak agency back in through the side door.
The three examples you give are helpful because they show that awareness doesn’t always play the same role.
In the burn case, the system reacts fast and locally. The arm moves before there’s any awareness of “me” doing anything. The story—I pulled my arm away—is clearly added after the fact. There’s no will operating there in any ordinary sense.
With hunger, things are already more spread out. Signals build, pressures compete, delay enters, bargaining happens. Awareness is present, but it isn’t in charge. It notices, resists, rationalizes, gives in, but the whole organism is doing the work, not a separate decider.
The bus example is where it gets tricky, and I think this is where your resistance is most reasonable. It feels as though awareness is driving the process end-to-end: planning, timing, sequencing, and correcting. And I agree that this feels categorically different from a spinal reflex or a hunger signal.
What I’d suggest, though, is that the difference is one of time scale and complexity, not a difference in kind.
In the bus case, constraint, history, and consequence are everywhere, just less visible. The goal itself is inherited from prior conditioning. The sense of urgency reflects social penalties for being late. The planning capacities were shaped because they improved survival and coordination in complex environments. Even the ability to simulate futures—“if I leave now, I’ll make it”—is a biological adaptation, not a free-standing faculty.
None of that makes awareness irrelevant. It matters a great deal. But it isn’t a little executive sitting above the machinery. It’s more like a shared workspace where pressures, memories, and possibilities get played out over longer stretches of time.
So when you talk about “biological will,” I think you’re naming something real—but I’d still want to keep the word light. What’s there isn’t a will layered on top of biology. It’s biology that has learned to reflect, anticipate, and regulate itself over longer horizons. Self-awareness doesn’t add a new force; it extends the reach of the same ones.
That’s why I reject both extremes. I don’t see a ghostly free will popping into existence out of nowhere. But I also don’t think planning, deciding, and deliberating are fake. The organism really does those things. What it doesn’t do, as far as I can tell, is step outside itself to do them.
And this is exactly why the comparison to AI matters. Machines can plan, optimize, and simulate futures too. What they still lack is a form of life where those processes are soaked in vulnerability—where getting it wrong actually costs them something, not just a parameter adjustment.
So yes: subtle distinctions, real differences, and no need for a little homunculus pulling levers.
Take care, amigo.
Yes, I agree.
You said, "It’s more like a shared workspace where pressures, memories, and possibilities get played out over longer stretches of time."
And causality moving inside of it all. Nothing, in fact, being free of causality.
Perhaps, a missing piece here is that causality varies in its affect. The Moon's gravity pulls on us more than Pluto's.
The proximate has more affect on us than the distal, in general.
And threats to our continued existence, as an evolved biological being, have more impact on us than things that do not threaten us.
My point is that chains of causality never end. But their affects lessen, in general, over increased gaps time and space.
For biological beings, survival and propagation have always been central. If we don't achieve them, we are off the game board. Extinct.
Another point that is relevant is that as biological complexity increases over evolutionary time, new emergent properties come forth. New things that never existed before. New things that create causal affects. Causal affects that never existed in a simpler world.
Imagine the new affects that arise from the emergence of self-awareness.
Greetings to you from Queenstown on New Zealand's South Island where Colette and I are taking a week's vacation.
De—
Yes, I think that’s right, and you’re sharpening the picture rather than pushing against it.
Nothing here is free of causality. What varies is which causal chains matter, and how much weight they carry. Proximity matters. Timing matters. Threat matters. The moon pulls harder than Pluto, and a threat to survival pulls harder than an abstract possibility three steps removed.
So I’m with you: causal chains don’t stop. But their effects thin out with distance in time, space, and relevance. That’s not a metaphysical claim, just a practical one about how organisms work.
And yes, for biological beings, survival and reproduction have always been central. Miss badly enough there and the story really does end. That’s the background pressure shaping everything else, including the more subtle forms of regulation we’re talking about.
On emergence, I agree—but with a constraint. New properties do arise as biological complexity increases. Self-awareness is clearly one of them. It changes the causal landscape by introducing memory, anticipation, social coordination, and reflection. Those things matter. They have effects that didn’t exist in simpler systems.
Where I’d still be careful is in how much ontological weight we give to that emergence. I don’t see self-awareness as introducing a new causal agent so much as a new way causal forces get routed, amplified, delayed, and negotiated within the organism. The system becomes sensitive to longer arcs and more abstract pressures, but it’s still the same game, just played on a bigger board.
So yes: causality everywhere, effects tapering with distance, survival saturating the system, and emergence changing what can matter without breaking the chain.
Enjoy Queenstown,
.
Great piece. Thanks! Have you read any of Mel Pine's recent stuff?
Don't know the name.