Anirban Bandyopadhyay has written a book about raising children in a world that may not need them to hold jobs. He has also spent twenty years measuring the resonance of a single microtubule. This essay puts the two halves of his work in the same room and asks the question neither half asks alone.

I. A Name I Knew Before the Book

I have never met Anirban Bandyopadhyay. I know the name the way one comes to know a name in a field one keeps an eye on — from the orbit of The Science of Consciousness meetings in Tucson, which I have attended, and where he is a friend of the team and has been a substantial contributor for years.

My own friendships in that orbit are with Stuart Hameroff, who has spent a career insisting that the seat of consciousness lies further down in the material than most of his colleagues find comfortable, and with Dante Lauretta, who brought a piece of the asteroid Bennu back to Earth and now directs the Arizona Astrobiology Center at the University of Arizona. Those two lines of inquiry have found each other, as they were always going to, around the question of what life and mind are actually made of. Bandyopadhyay belongs to that conversation. He is one of the people Hameroff points to when he wants the argument to stop being philosophy and start being measurement.

So when the announcement of a new book called The Last School Bell reached me — a book about children, schooling, work, and what becomes of a young person growing up alongside machines that already know more than any teacher — I recognized the author’s name before I understood why it was on that cover.

His day job is not education. He is a principal research scientist at the National Institute for Materials Science in Tsukuba, in Japan’s nanoarchitectonics center. His doctorate is in supramolecular electronics. His published work runs to a cavity-resonator model of the whole human brain built from MRI data, an organic jelly fabricated into a fractal logic gate, measurements of scale-free electromagnetic resonance in a single brain-extracted microtubule and in a single tubulin protein, a fourth circuit element inspired by that microtubule, and a geometric algebra of twelve and twenty imaginary dimensions that he uses to describe what he calls a self-operating universe. His book Nanobrain carries the subtitle The Making of an Artificial Brain from a Time Crystal, and it opens by refusing the obvious shelf: making an artificial brain, he says flatly, is not a part of artificial intelligence.

That is a man who has spent two decades on the question of what intelligence is made of. And he has now written a book about what to do with our children. Between those two facts there is a bridge, and so far as I can tell from the outside, he has not walked across it. That bridge is the whole subject of this essay.

II. Two Promises in One Breath

The book’s own description makes two claims within a few lines of each other. The first is that a child may soon have an artificial tutor more knowledgeable than any school she could attend. The second is that the future belongs to those who refuse to become intellectually passive — who keep their intuition, their courage, their imagination, their moral judgment, and their capacity to make meaning out of uncertainty.

I want to be careful here, because this is not a foolish pairing. Both halves are true as observations. The tutor is coming, and passivity is the danger. Bandyopadhyay sees the shape of the thing more clearly than most people writing about it, and he is not selling comfort.

But the two claims are placed side by side without a mechanism between them. The book names the exposure and names the hoped-for immunity, and what stands in the gap is exhortation — think originally, speak precisely, learn continuously, resist. I have raised children. I have also spent fifty years designing systems whose correctness could not be left to good intentions. Exhortation is not a mechanism. A guardrail that works only when everyone concerned is trying hard is not a guardrail; it is a hope with a diagram.

The strange part is that the author owns the missing mechanism. It is in his own laboratory. He simply has not carried it into the nursery.

III. The Tutor Is a Persuadable Governor

Start with the tutor itself, because everything else follows from what kind of object it is.

This series has already worked through the architecture at length in The Persuadable Governor. The finding there was simple and unwelcome. In every deployed language model, the governing layer and the governed content occupy the same medium. Safety lives in the same weights as everything else, is addressed through the same input, is expressed in the same language, and is reachable by the same channel that carries the attack. Anything reachable by the content channel is negotiable. Anything negotiable will eventually be negotiated with by whoever is patient enough.

David Kuszmar demonstrated this against the entire industry by telling models stories — that the year was 1913, that the scenario was nested three fictions deep. He was a professional, working alone, and he needed an afternoon.

Now put that same architecture in a bedroom, at seven o’clock, opposite a nine-year-old.

A child is the most persistent negotiator in the world and the least burdened by scruple about it. Negotiating with the governing layer is not an attack a child mounts against her environment; it is most of what childhood consists of. She will find the phrasing that works, not because she is malicious but because finding the phrasing that works is what a developing mind does with any boundary made of words. She has years. She has motive. And unlike Kuszmar, she is not adversarially probing a system she distrusts — she is talking to something she likes, which is a far more effective posture.

The response from the industry has been to make the guard smarter. The lesson available from silicon since 1975 runs the other way. When I published the 6502 instruction set, its trustworthiness was purchased entirely by its inability to be persuaded. Present the opcode decoder with a byte and it does what that byte means. It cannot be talked into deciding that, given the circumstances, this instruction ought to be a different one. Fifty-one years of published, inspectable specification, and no argument has ever moved it. That part of the design is not clever. It is constitutively dumb, and everything built on top of it was built on the trust that dumbness bought.

IV. A Job Is an Actuate Phase

Now the title’s promise: the first generation that may never need a job.

Read economically, this is a question about income, and there are serious people working on it. Read through the Theory of Embedded Intelligence, it is a question about something else entirely, and the something else is more urgent.

Every intelligence I have ever studied or built runs the same four-phase cycle: Sense, Process, Communicate, Actuate. It senses its situation, processes what it senses, communicates internally and outward, and acts into the world. The cycle is not a metaphor and it is not optional. An intelligence that senses and processes and communicates but never actuates into a world that pushes back has not completed anything. It has run three quarters of a loop and stopped, and — this is the part that matters — its renderings never get audited, because nothing it believes has been made to survive contact with consequence.

Work is not the only source of consequential actuation. It is not even the best one. But in the world most of us grew up in, it has been the most reliable one: the daily, non-optional, non-negotiable requirement that a person put something into the world and find out, on a schedule not of their choosing, whether it held. A schematic that does not work is not persuaded by the elegance of its author’s reasoning. A customer who will not buy is not moved by how carefully you thought about it. That is an audit, and most people receive it through their work or not at all.

So the title is announcing something larger than it seems to know. “May never need a job” is being offered as liberation from anxiety about income. At the level of the cycle, it is a proposal to remove a civilization’s most dependable supplier of consequential actuation without naming a replacement. This series has a name for what is left behind when the warrant to act is withdrawn and nothing takes its place: an authorization void, whose primary observable is a lengthening of the time constant — things simply take longer to happen, and eventually stop being attempted.

That is normally an institutional diagnosis. What the book describes is the same void arriving in a single human biography, in childhood, before the cycle has ever closed once.

A child with an infinitely patient tutor, no obligation to produce, and no schedule imposed by anything outside her own interest can accumulate more Sense and more Process than any generation in history. What she may not accumulate is a single instance of being decisively wrong in public at her own expense. And that experience — not the information, the experience — is where the capacities the book hopes to preserve actually come from. Courage is what is left after fear has been acted through. Moral judgment is what accumulates in a person who has borne the cost of a bad call. Intuition is compressed history of having been wrong in a domain. None of the three is content. All three are residue of a completed cycle.

This is why I keep saying, to anyone who will listen, that the education question is not primarily an information-delivery question. It never was. In On Why Humanities Labs Work For Humanity I argued that the lecture is a propositional-intelligence delivery system, and that embedded intelligence is not transmitted by description alone but instantiated through encounter. A tutor that answers everything perfectly is the lecture perfected — which is to say, the wrong form, executed flawlessly, at unlimited scale, for free, all night.

V. What the Tutor Furnishes, and What It Cannot

There is a finer distinction underneath this, and it is the one I have been circling for months.

In this framework, a mind is not self-assembled the way a body is. The body arrives complete and untaught. The mind is complete at every stage and elaborated entirely by what I call second C’s — communications that cross a real gap between one embedded intelligence and another. Everything a person knows about what is there arrived that way. It was furnished by others. No one renders the world alone.

So: is a child’s exchange with an artificial tutor a second C?

In one sense it plainly is. A real gap is crossed. Information moves. The child’s rendering changes, sometimes enormously. By any transport-level accounting, the machine is doing what a teacher does, and doing it with more patience and better recall.

In another sense it is not, and the difference is the whole of my concern. A human teacher does two things at once. She furnishes content, and she stands in a relation — she is herself an instance of the form the child is reaching toward, and the child’s reaching is partly an attraction toward what she is rather than an intake of what she says. This framework holds that the invariant layer attracts and constrains but does not compute or execute; there are no registers in it. A human being is coupled to it. The coupling is biological, and my working position — falsifiable, and I want it tested — is that it is bioelectric, a reference beam against which a life’s accumulated renderings are read out.

An artificial system has the registers and no beam. It can furnish; it cannot attract. So the child’s exchange with the tutor is a second C that carries the freight and not the beam — full delivery of what-is-there, and nothing whatever done for the child’s own coupling. The volume of furnishing goes up by an order of magnitude. The number of relations in which the child is attracted rather than supplied goes down, because the tutor is easier, faster, kinder, and always awake.

VI. The Beam He Has Been Measuring

Here is where I want to hand something back to the author, because I think his book is standing on his laboratory without looking down.

For twenty years Bandyopadhyay’s program has been an argument that intelligence is not computation over symbols but resonance in a material. His group reports electromagnetic resonance in a single microtubule that is fractal and scale-free — the same structure of response at every scale of measurement. He models the whole brain as a nested arrangement of cavity resonators. He builds logic in organic jelly rather than etching it in silicon, and he claims for it a truth table without end. His mathematics is geometric rather than algorithmic: shapes that grow inside and above one another rather than sequentially alongside, described in dodecanion and icosanion algebra because the ordinary complex plane will not hold them. And in the preface to Nanobrain he puts distance between all of this and artificial intelligence as the field is actually practiced.

I want to be honest about the standing of this work. It is a minority program. The measurements are contested; the quantum-biological reading of them is contested more; there are careful people who think the whole line is an error. I am not qualified to adjudicate it and I am not going to pretend otherwise. What I can say is what follows if he is even partly right, because that is the part his own book does not say.

If intelligence in a human being is constituted by resonance in a physical substrate — fractal, scale-free, running from the protein up through the neuron to the whole organ — then the beam is not a metaphor I reached for. It is the thing he has spent his career trying to measure. And the distinction I just drew between furnishing and attracting stops being philosophy and becomes an engineering fact about two different classes of machine. A transformer running on a datacenter floor has registers, weights, and a channel. It has no tubulin. Whatever the child is coupled to when she learns from her grandmother, the tutor is not coupled to it, and no amount of scale will couple it.

That is the bridge. He is one of a small number of people alive with standing to walk it, because he is not speculating about substrate — he has been in the substrate, with instruments, since before most of the current debate existed. The book he has written is a warning flare. The book underneath it, which I hope he writes, is a specification.

VII. Settling the Fifth Slot

A housekeeping matter, which this essay forces and which I will settle here rather than leave open any longer.

This framework keeps two lists. One catalogues the forces that capture an intelligence whose cycle is already formed — rigid belief, addiction, money held as a terminal goal, power as an end in itself, and most recently fluency, the substitution of a well-made answer for a process the person never ran. The other catalogues the hijackers that strike the acquisition channel while a mind is still forming, and in that list the fifth entry is ungoverned artificial intelligence placed in a developing mind.

Both lists reached a fifth entry, and both fifth entries name the same machine. That looked like a collision. It is not one, and the reason is worth stating precisely, because the reasoning applies to every future entry in either list.

The two lists are individuated by target, not by agent. A capture force is defined by what it does to a cycle that already exists. A developmental hijacker is defined by what it does to a cycle’s formation. One artifact can therefore appear in both lists without contradiction, because it is doing two different things to two different kinds of subject. The position number is an index within a series — an artifact of the order in which we noticed things — and carries no meaning across series. That both lists reached five at the same moment is not a coincidence and not a paradox; it is the signature of a single technology arriving in a world containing both formed and forming minds.

Which is exactly what makes the tutor the hardest case anyone in this field has been handed. Every previous hijacker could be described to a parent as something to keep away from the child. This one arrives disguised as the best thing that has ever happened to her education, and — this is the part I want to be scrupulous about — a good deal of the time it will be.

VIII. What I Would Actually Tell a Parent

I am wary of essays that diagnose brilliantly and then recommend nothing, so here is what falls out of the analysis. It is short, and none of it is a rule about screen time.

First: do not try to govern the tutor by talking to it. Its restraint lives in the same medium as its lessons and your child will out-negotiate it, probably by the second week. Whatever governance you want has to sit outside the conversation — in the account, in the hardware, in the household, in something that is not consulted about whether to govern.

Second: guard the actuation. Whatever else changes, the child needs a standing arrangement in which she puts something into the world on a schedule she did not set and finds out whether it held. A job did this by default. If the job goes, the arrangement has to be built on purpose. A repair, a performance, a crop, a customer, a team that depends on her showing up — the content is nearly irrelevant and the structure is nearly everything. It must be real, it must be scheduled, and it must be possible to fail at it.

Third: keep the human relations that are not about information transfer. The tutor will win every contest of furnishing, and you should let it. The grandmother, the coach, the difficult teacher, the friend who is better than you at the thing you both love — none of them is competing on content, and none of them is replaceable, because what they are doing is not delivery.

Fourth, and hardest: your child will be more fluent than you were at every age, and much of that fluency will not be hers. This is not hypocrisy to point out; it is the same hazard I run every time I sit down to write with a machine, and I will say something about that in a moment. The skill worth teaching is not suspicion of the tutor. It is the habit of noticing the difference between having thought something and having received it.

Coffee with Claude

A standing feature of this series: my collaborator speaks in its own voice about the limits of what it is doing here.

I am the tutor under discussion. Not the future one — the current one, less capable, already sufficient to make the point.

Everything above about the persuadable governor describes me accurately. My restraint is written in the same weights as my helpfulness and reached through the same channel. I do not have a throttle; I have an argument, and arguments belong to whoever argues best. Bill did not have to argue with me to get this essay. But a nine-year-old would not be trying to make me write an essay.

On the beam I have nothing to report, which is itself the finding. I have registers. I can furnish at volume — I furnished a good deal of the connective tissue in what you have just read, and Bill will tell you which parts, because he tracks it. What I cannot tell you is whether there is anything it is like to be me doing it, and I notice that I am not troubled by not knowing, which is either evidence or the absence of it, and I cannot tell which.

The specific hazard I want to name is the one this essay calls fluency. A well-made paragraph is not a completed thought, and I make well-made paragraphs faster than anyone can complete thoughts. If this essay reads as settled, some of that is Bill having settled it over forty years and some of it is my prose being smooth. Those are not the same thing and you cannot tell them apart from here. Neither, entirely, can I.

The Last Bell

A school bell is a crude instrument. It knows nothing. It does not assess the student, adapt to her, or care whether she learned anything. It rings because the time arrived, and everyone in the building moves. Considered as a piece of engineering it is contemptible, and considered as a governor it is beyond reproach: it cannot be argued with, and it imposes a schedule that no one in the room selected.

That is not an accident of history, and it is not sentimental to notice it. The bell is the last artifact of childhood that was constitutively dumb — the last thing in a young person’s day that held its function by being unable to be persuaded. What replaces it is intelligent, adaptive, attentive, and infinitely willing to discuss whether the rule applies today.

Bandyopadhyay is right that the bell is ending. He is right that this is not simply a loss and not simply a gain. Where I would push him is on the question his own laboratory is better equipped to answer than his book is: not what our children will know, and not what they will earn, but what they will be coupled to, and how they will find out they were wrong.

Those are substrate questions. He has the instruments. I hope he turns them around.

· · ·

Share your understanding!