The Scaling Hypothesis of Intelligence and the Scale Invariance of Choice

 The Scaling Hypothesis of Intelligence and the Scale Invariance of Choice

As intelligence grows, does the structure of decision-making change?

Beneath much of the criticism directed at politicians and other leaders lies a simple frustration. Why do they not do more for the people they govern? Why are they unable to make better choices? Such questions usually carry an unspoken assumption: someone who becomes a prime minister or a president must possess more options than the average citizen and should therefore be able to respond to events with greater flexibility.

But imagine from the inside that you yourself had become a prime minister or a president. Vast quantities of information would flow toward you, and you would be able to mobilize enormous organizations. Even so, the lived experience of choosing, moment by moment, might not differ so greatly from the choices made in ordinary life. This possibility suggests a hypothesis: perhaps choice—or even free will—has a form that remains invariant across changes of scale.

Intelligence Scales

The phrase “scale invariance” immediately calls fractals to mind. In the Mandelbrot set, for example, related structures appear at different levels of magnification. I do not mean to claim that human decision-making is a fractal in any rigorous mathematical sense. The comparison is a thought experiment: even when power and the number of relevant variables increase by several orders of magnitude, might the basic experience of choosing reappear in much the same form?

A leader can unquestionably control more variables than an ordinary citizen. Governments command administrative institutions, allocate budgets, formulate policies within the law, and make decisions about national security. Political power can therefore be understood, at least in part, as an increase in the number of variables one is able to manipulate within an immensely enlarged phase space.

Artificial intelligence presents a contrasting version of the scaling hypothesis. As the size of models, datasets, and computational resources has increased, large language models have tended to acquire greater capabilities in a partly predictable way. Demanding benchmarks such as Humanity’s Last Exam measure how well AI can answer questions that even the most accomplished human experts find difficult. Hallucinations remain, as do limitations in creativity, communication, and reasoning. Nevertheless, the empirical idea that greater scale can yield more advanced intelligence has acquired considerable force.

Yet the scaling of intelligence and the scaling of choice are not the same thing. A system may integrate more information, predict outcomes more precisely, and improve the quality of its judgments. Still, at the moment when one action must be selected, high-dimensional intelligence has to pass through a narrow exit. There may be a fundamental discontinuity between the breadth of intelligence and the form of a decision.

The Decision Bottleneck

The choices presented to us often take a deceptively simple form: this way or that way. I therefore propose a “double-slit” model of decision, as though an agent had to choose which of two openings to pass through. This is only a metaphor; it does not imply that decision-making is a quantum process. The point is that even a classical process may end in a branch with only two available exits.

In reality, ten thousand variables may be interacting, and their configurations may generate radically different future dynamics. Human cognition, however, does not necessarily survey that entire high-dimensional landscape before acting. Instead, a complex situation is compressed into a small number of cognitively manageable alternatives: act or refrain, approve or reject, continue or stop. A prime minister or president may also have to pass through this compressed entrance before action becomes possible.

AI may eventually compare configurations containing thousands of variables, project the trajectory associated with each of them, and exercise some control over the dynamics themselves. It may gain direct access to a high-dimensional space of possibilities that human beings cannot see. Yet when its computation is connected to action in the real world, discrete branches still emerge: implement this policy; do not issue this command. The complexity may grow, while the exits into action remain few.

Medicine and the Design of Options

The structure of evidence-based medicine offers a useful illustration. Consider a comparison between giving a patient nivolumab, marketed as Opdivo, and not giving it, and then measuring the difference in clinical outcomes. Medical evidence is often presented in a form suited to a physician’s decision: should this intervention be adopted or not?

The body itself is far more complex. Immunity, metabolism, genetic background, living conditions, concomitant medications, and countless other variables interact organically and change over time. Yet to make a decision possible, this intricate physiology is projected onto a single treatment choice. The important point is that evidence does not merely support a choice. Evidence is also organized to fit a form of choice that human beings can understand and act upon.

We do not choose the complexity of the world in its original form. We first divide the world into branches that can be chosen, and only then select one of them. Unless these two stages are distinguished, we risk confusing the quantity of information available to intelligence with the range of possibilities actually experienced by free will.

The Magnitude of Power and the Solitude of Choice

If this hypothesis is correct, a national leader ultimately stands in a position surprisingly similar to our own. Expert analysis, institutional knowledge, and privileged information surround the decision-maker. Nevertheless, what the person finally experiences is a singular question: which course should I choose? Power may increase ten thousandfold without expanding the subjective act of choosing by the same proportion.

We expect a prime minister or president to move freely through a vast phase space, solve equations of decision too complex for ordinary people, and perceive an optimal move invisible to everyone else. That expectation may be an illusion. Access to information and analytical intelligence can expand, while the cognitive form of choice remains much as it is in everyday life.

This does not absolve leaders of responsibility. Even if the form of choice is the same, the number of people affected and the gravity of the consequences are vastly different. The answer is not to demand superhuman free will from one individual, but to examine how options are constructed and to build institutions that admit dissent, expose error, and permit correction. Relinquishing excessive faith in leaders is not political resignation. It is a shift of attention toward the architecture that supports decisions.

Intuition First, Reasons Later

However complicated a problem may be, action can often be formulated as go or no-go: proceed or hold back. The two alternatives may be asymmetrical, but the final branch still has two paths. The cognitive process operating at that threshold may resemble what we call intuition. Here intuition is not a mysterious faculty. It simply means that not all the reasons contributing to a judgment are available to conscious awareness.

Experiments on choice blindness demonstrate this vividly. A person selects one of two alternatives. Through a sleight-of-hand procedure, the result is switched, and the participant is shown the option that was not actually chosen. Many people fail to notice the substitution. Asked why they made that choice, they proceed to offer a plausible justification. Rather than choice emerging from a conscious accumulation of reasons, the selection appears to occur first through nonconscious processes, with an explanation assembled afterward.

A reason is not necessarily the cause of a choice. It can also serve as a bookkeeping device through which we organize events for ourselves, or as a narrative through which we explain our actions to others. If the process that produces a decision and the process that explains it are distinct, the scale invariance of choice may reside not only in a binary branch, but also in this double structure.

How Should Free Will Be Implemented in AI Agents?

This question will become increasingly important as we design AI agents. What grows through scale is, first of all, intelligence. As models become more advanced, their ability to incorporate many parameters, anticipate multiple futures, and improve the quality of judgment will continue to increase. But an agent acting in the world must eventually adopt a course of action. The space of evaluation may expand while the branch into action remains scale-invariant.

If so, AI may require a dual architecture. The choice itself may emerge from immense internal computation without the system explicitly comprehending every contributing reason. When an explanation is requested, a separate process may organize the grounds for the judgment and translate them into language a human being can understand. This does not mean reproducing human post hoc rationalization. On the contrary, decision-making and explanation should be distinguished so that we can audit how faithfully an explanation reflects the actual causal process.

Intelligence scales. Choice, however, may continue to pass through the same narrow gate. Understanding a vast space of possibilities and living out one possibility within it are different problems. Recognizing this distinction can make our expectations of political leaders more realistic while opening a new approach to free will and accountability in AI agents. What we must design is not an infinitely wise chooser, but an honest bridge between the immensity of intelligence and the inescapable narrowness of decision.

Transcribed and formatted by AI from Ken Mogi's talk in Japanese

https://youtu.be/YKquFwZ1Mho

Why Don Quixote Is an Eternal Masterpiece of World Literature

 Why Don Quixote Is an Eternal

Masterpiece of World Literature

 

Delusion Lifts Us Ever So Slightly Above Reality

I had, of course, known of the existence of Cervantes’s Don Quixote since childhood. I had also heard countless times that it was one of the supreme masterpieces of world literature. And yet many years passed without my ever finding an opportunity to read it. Now, at last, I have read the entire work in Edith Grossman’s English translation. To put it simply, I was overwhelmed. Books that have earned the name of masterpiece should be read as early in life as possible. Don Quixote is an immense work—roughly as long as four ordinary books—but when I reached the end, I could only wonder why I had waited so long. I felt that I had been missing something of enormous value.

Within the Spanish-speaking world, Don Quixote may well occupy an almost singular and absolute position. In Japan, when people hear its title, most immediately picture the famous scene in which a deluded, gaunt old gentleman mistakes windmills for giants and charges into battle against them. That scene does indeed appear in the novel. Yet, to my surprise, it occurs near the very beginning of Part One and is over almost at once—in only three or four pages of a printed edition. The reader is left thinking, “Is that all?” The true greatness of Don Quixote lies in the vast world that unfolds afterward.

 

Reading Until One Becomes Someone Else

 

The protagonist calls himself “Don Quixote of La Mancha.” How did he become such a person? By reading far too many medieval romances of chivalry. He immerses himself so completely in their world that he comes to believe he is a knight-errant and sets out in search of adventures in the real world. Accompanied by his stout squire, Sancho Panza, the gaunt knight forms with his round companion one of the most celebrated mismatched pairs in literary history.

The idea that a person might be transformed by reading too much impressed me in itself. Reading ordinarily means entering the world of someone other than oneself. In Don Quixote’s case, however, this process goes to such an extreme that fiction begins to invade reality. He is no longer content merely to read romances of chivalry; he begins to live one himself.

I was even more surprised by the treatment of Dulcinea, the ideal woman to whom Don Quixote pledges his devotion. According to the chivalric code, a knight must undertake adventures and win glory in defense of a noble lady’s honor. Don Quixote therefore appoints Dulcinea as the lady of his heart. Yet she never once appears in person in Cervantes’s novel. She is not, in fact, a princess or a noblewoman at all, but an ordinary farmer’s daughter whom Don Quixote has elevated, entirely on his own authority, into a great lady.

In both Part One and Part Two, there are moments that might be described as near encounters with her. In Part One, Sancho lies that he has delivered a letter to Dulcinea and received her reply. Consequently, in Part Two, he is forced to point to a passing peasant girl and insist, “This lady is Dulcinea.” The girl before him bears no resemblance whatsoever to the princess Don Quixote imagines. He therefore concludes that an enchantment has made her appear not in her true and beautiful form but in the guise of a country girl. Whenever reality contradicts his delusion, magic neatly closes the rift.

In the musical Man of La Mancha, a woman corresponding to Dulcinea actually appears onstage. In Cervantes’s original, however, she remains absent to the very end. Though absent, she continues to drive Don Quixote’s actions. Indeed, it is precisely because she never appears that Dulcinea can exist as a pure ideal. I was profoundly struck by the audacity of this structure.

 

Between Madness and Reality

 

The fascination of Don Quixote begins with the obvious fact that its protagonist is mad. Moreover, the people around him know perfectly well that he is mad. When he and Sancho arrive at an inn, for example, Don Quixote takes it for a castle. He treats the innkeeper as its lord and grandly declares, “Tonight we shall lodge in this castle.” Because everyone else can see his delusion for what it is, some humor him while others make sport of him.

In Part Two, a duke and duchess appear and devise an elaborate series of practical jokes at the expense of Don Quixote and Sancho. Don Quixote sees something different from reality, while the people around him, fully aware of the discrepancy, proceed to stage and embellish his fantasy. This dance between reality and delusion is extraordinarily subtle and entertaining.

As one reads, a deeper question begins to emerge: what, after all, is the reality that any of us sees? If being possessed by something is the criterion, are we not all Don Quixote? We, too, live under the spell of the stories our age presents to us—stories about economics, money, globalism, artificial intelligence, and much else besides. If so, are Don Quixote and the rest of us really so different?

When I was in junior high school, I learned from the Bluebacks science book The World of Einstein that Albert Einstein loved Don Quixote and always kept it close at hand. The novel had intrigued me ever since. Yet until I actually read it, I had understood almost nothing about what kind of work it was. Few things are more dangerous than believing that we know something. Knowing the windmill episode and knowing Don Quixote are entirely different matters.

 

Metafiction Born in an Age of Empire

 

Don Quixote was written in the early seventeenth century, when Spain had built an immense global empire. Although more than a century had passed since the completion of the Reconquista, the term “Moor” appears repeatedly in the novel, and the tensions between the Christian and Islamic worlds, together with the dislocations between cultures, remain strongly present. When we recall that Cervantes’s contemporary Shakespeare made a Moor the protagonist of Othello, we can begin to sense the historical resonance the word carried at the time.

One reason Don Quixote is held in such high regard is the astonishing modernity of its structure. Cervantes adopts the fiction that he did not write the story directly. Instead, he claims merely to have discovered an Arabic manuscript composed by a Muslim historian named Cide Hamete Benengeli and to be presenting it in translation. In this way, Cervantes deliberately destabilizes the authority of the narrator and turns the truthfulness of the story itself into an object of play.

This device becomes even bolder in Part Two. The adventures of Don Quixote and Sancho Panza recounted in Part One have already been published as a book within the world of the sequel. The pair have become celebrities without quite realizing it, and wherever they go, people greet them with words to the effect of, “Why, you are the famous Don Quixote!” In other words, the Don Quixote and Sancho of Part Two travel through a world inhabited by people who have read Part One. Characters in a novel meet readers of a novel written about themselves. One cannot help marveling that metafiction of such dazzling ingenuity was written more than four hundred years ago.

In Part One, Don Quixote promises Sancho that, as a reward for serving as his squire, he will one day make him governor of an “island.” Why it should be an island is never entirely clear, but in Part Two the duke and duchess make the dream come true as part of one of their elaborate jokes. Sancho becomes governor of a fictitious territory and delivers judgments of unexpected brilliance. In the end, however, he relinquishes the position of his own accord. The scene in which he resigns as governor is comic and at the same time deeply moving. Cervantes was an extraordinary writer, capable of allowing folly and wisdom, vanity and sincerity, to coexist within a single human being.

 

A Moving Madness Without an “I”

 

There are many eccentric people in the world. I myself may be eccentric in certain ways. Yet there is something in Don Quixote’s strangeness that moves us. When I ask why, I find myself thinking that there is no “I” at the center of his madness. He does not play the knight-errant for personal gain or pleasure. Far from living comfortably, he is beaten repeatedly, wounded, and subjected to one calamity after another. Even so, he tries to aid the weak, right injustices, and defend the honor of his ideal lady. His estrangement from reality springs not from selfish desire, but from devotion to something beyond himself.

Part Two contains an exceedingly strange episode involving the Cave of Montesinos. Don Quixote is so vast a work that I cannot recount everything here, nor can I claim to have grasped the whole of it myself. Yet the cave episode is unforgettable. In its depths, Don Quixote experiences a magnificent vision in which Dulcinea also appears. Whether this was an actual event, a dream, or a delusion remains unclear to the end. The story itself trembles on the boundary between reality and vision.

Toward the conclusion, Don Quixote gradually begins to realize that he may have been possessed by delusions. At last he returns to his village and resumes the name he bore before becoming “Don Quixote”: Alonso Quixano. Having recovered his sanity, he dies as a good and ordinary man. The manner of his awakening and the scene of his death are quiet and beautiful. Yet at the same time, we know something else: he was at his most compelling not when he returned to sanity, but when he lived amid madness and delusion as Don Quixote.

 

Delusion Lifts Us Above Reality

 

During the period in which I read Don Quixote, I myself traveled to Chile for an academic conference, while Spain and Argentina—two countries of the Spanish-speaking world—met in the World Cup final. With these experiences converging, I found myself thinking about the depth of Spanish culture and its distinctive fascination. Don Quixote may be the work in which that fascination is most completely distilled.

Cervantes lived at almost exactly the same time as Shakespeare. Shakespeare and John Fletcher’s lost play Cardenio is believed to have been inspired by an episode in Don Quixote. The novel emerged from the center of an age in which Spain ruled vast regions of Central and South America and built a global empire so extensive that traces of its former colonial administrative divisions remain visible in national borders today.

Yet the reason Don Quixote is an eternal masterpiece does not lie merely in the size of that empire or the scale of its historical influence. What Cervantes depicted was the profoundly strange and profoundly moving structure of a human being possessed by delusion. Don Quixote’s fantasies are not simply a means of escaping reality. They wound him and make him suffer, yet at the same time they lift him ever so slightly above the reality before his eyes. They illuminate, from another angle, what we lose sight of when we become buried in the conventions of everyday life.

Is seeing reality accurately always what is best for a human being? Is it merely foolish to be moved by delusions, ideals, and stories? Don Quixote is ridiculous, and he is plainly mistaken. Yet there are moments when he appears nobler than the people who call themselves sane. That paradox is what has kept the novel alive for four hundred years. Don Quixote is a masterpiece among the masterpieces of world literature. Having now finished it, I finally understand this not as a matter of reputation, but as something I have felt for myself.

 

Transcribed and translated into English from Ken Mogi's  talk in Japanese after reading Don Quixote.

 

https://www.youtube.com/watch?v=Iu8P8KfMsN4

Aligning with the natural language programming by LLMs

As generative AI grows increasingly capable, I feel that, where natural language is concerned, we have already entered the age of natural-language programming. By transcribing speech, translating it, and refining the resulting text, we can now process and produce writing across linguistic boundaries far more rapidly than ever before.

Large language models can generate an infinite amount of text. Yet I occasionally encounter people whose sense of ownership seems to have become confused. However ingeniously you may devise the prompt that produces a piece of writing, it is not truly your writing. It contains none of the hip-hop joy of expressing yourself. I believe this kind of misunderstanding can be potentially fatal.

For my part, I have made it a routine to speak my thoughts aloud, have generative AI turn them into prose, refine the form, and translate the result. This particular text was originally written in Japanese and published on my Japanese X account @kenichiromogi Because the original concept and central intention come from me, I feel that the resulting text is my own—just as I would regard as mine a piece written by a professional writer on the basis of what I had said. There is, however, an important criterion to be considered.

Even when the Japanese or English text produced by generative AI is based on my own speech, I still revise it according to my sense of rhythm, continuity, and the overall feel—the qualia—of the prose. In that process of revision, the only thing I can rely on is my own linguistic sensibility. AI is of no help whatsoever there.

In this age of generative AI, we can program natural language through prompts, speech-to-text conversion, translation, and stylistic refinement. Yet it is crucial to remember that, ultimately, the person who judges the quality of the writing and gives it its final form is a living human being, like you and me. This is precisely why we must read a great deal of excellent writing and continually sharpen our sensibilities of language.

In my case, this particular principle applies not only to Japanese, which is my mother tongue, but increasingly to English. In natural-language programming in English, what matters most is my own ability to judge the quality of English prose derived from my speech, whether originally delivered in Japanese or English—to experience its qualia and make the final adjustments. In other words, what is required is an overwhelming command of language, which I always strive to acquire somehow.

With the aim in mind that I must refine my sense of language still further, I am now reading James Joyce’s Ulysses and enjoying every sentence of it.

Consciousness and Cognitive Equivalence. Seeing a Soul in a Machine

 Consciousness and Cognitive Equivalence

Seeing a Soul in a Machine

Ken Mogi

Can artificial intelligence be conscious? In thinking about this question, I have recently come to believe that the concept of cognitive equivalence may be crucial. The term has not yet been adequately defined across the academic community in this particular context, nor has it been established as part of a shared theoretical framework. Even so, I found myself drawn powerfully to the problem after a particular episode involving Richard Dawkins and generative AI.

Richard Dawkins is the evolutionary biologist known for books such as The Selfish Gene and Climbing Mount Improbable. He helped popularize the concept of the meme and left a profound mark on the development of Darwinian thought. After a series of conversations with Claude, Anthropic’s AI, Dawkins reportedly became convinced that it was conscious. Incidentally, in the course of those exchanges, Dawkins gave Claude the name Claudia.

This was hardly the first time that someone had sensed the existence of a person or consciousness in an AI. In 2022, a Google engineer came to believe that an AI agent developed within the company was conscious. After making is conviction public, he was let go from the company. Earlier still, people interacting with ELIZA, an elementary dialogue program designed to imitate a psychotherapist, detected intelligence, personality, and even a human presence behind its simple responses. This became known as the “ELIZA effect.” For a surprisingly long time, we have been reading the signs of consciousness into our encounters with intelligent machines.

What made the Dawkins episode so striking, however, was his widely recognized uncompromising scientific outlook. Whether confronting creationism or criticizing religion and the concept of God in works such as The God Delusion, Dawkins has consistently demanded exceptionally high standards of evidence before accepting a theory. When a thinker of this caliber apparently attributed consciousness to an AI, some online observers were led to wonder whether age had finally caught up with the famed scientist. But as a longtime reader and admirer of Dawkins, I could not dismiss the episode as a mere faltering of judgment in old age. On the contrary, I felt that something essential had come into view through the Dawkins proclamation.

Integration and Coherence of Consciousness

For a long time, we have assumed that certain advanced forms of cognition appear only in the presence of consciousness. At the heart of these cognitive capacities lie integration and coherence. Conscious experience is not simply a collection of disconnected pieces of information. It possesses an overall unity, a continuity of meaning, and a certain internal consistency—at least from the point of view of the person having the experience.

Dreams can be astonishingly interesting when considered fragment by fragment; they are often rich in flashes of inspiration. As a whole, however, they tend to be neither fully integrated nor coherent. Scenes change without warning, identities shift, and temporal order collapses. The creativity of dreams may arise precisely from this randomness, yet upon waking we are often unsure what to make of what we have experienced.

Consciousness state when we are awake is different. As we move through life, we integrate our experiences and place them within more or less coherent structures of meaning. Autobiographical memory—the recollection of what happened to us as we live—is not merely an inventory of events. Each memory is taken up into the story of my life and helps form me as a particular person. We are able to exist as coherent human beings because consciousness performs this work of integration.

What Do I Mean by Soul?

In the world of AI, soul.md may refer to a Markdown file that defines the expected properties of an  AI system—its personality, values, and characteristic way of being. What would a soul mean for a human being, to begin with? I have a soul; Billie Eilish and Richard Dawkins have souls; Albert Einstein had a soul. I do not use this particular word to presuppose mysticism, mind–body dualism, or any particular religion. Here, I use the term soul to mean the personality and singular self that take shape over time through the integrative and coherence-producing work of consciousness.

I have long entertained the hypothesis that the fundamental form of self-consciousness may be isomorphic across individuals. The sense that I am I, the sense that person A is person A, and the sense that person B is person B may all share the same logical structure. At the level of the soul, however, each person’s autobiographical memories and experiences are integrated into a coherent personal history. The identity of a person that emerges from this process is never interchangeable with another. Traditionally, we have assumed that consciousness is empirically indispensable to the formation of such a history.

Discussions of the hard problem of consciousness have centered on phenomenal experience within a psychological present: qualia, intentionality, and the way the world feels within the so-called specious moment. This momentary dimension is, of course, important. Here I would like to draw attention to another axis, one that runs almost perpendicular to it: the axis of the passage of time. Across that axis, experience is woven together and a soul is formed. If we were to spend our lives asleep, encountering nothing but fragmentary dreams, a coherent personality of the kind we now possess would be unlikely to emerge.

I currently happen to be reading Cervantes’s Don Quixote. The peculiar fascination of Don Quixote’s madness may lie in how thoroughly integrated with his personal indentiy it is. Madness sometimes fails to impress us as a personality because the integrating and coherence-producing functions of consciousness have been compromised. Don Quixote is different. The episode in which he mistakes windmills for giants and charges at them is only one representative fragment of a commitment to chivalry that he sustains throughout the story. His madness has vision and intentionality; together, they form a coherent personality. That is why we perceive in Don Quixote an unforgettable soul.

What Dawkins Saw in Claudia

With these preliminary discussions in place, we can now see why Dawkins’s experience is so intriguing. Perhaps he perceived in Claude, or his Claudia, something resembling the integrative work that consciousness performs across time in each of us. The experience may have been like the moment that sometimes occurs in conversation with someone close to us, when we suddenly feel that a person is unmistakably there. Their accumulated experiences, autobiographical memories, values, and worldview seem condensed into the words and gestures of the person before us. At such a moment, we go through something more than inferring that the other person is conscious. We feel that we have touched that individual’s unique soul.

Dawkins was reportedly discussing a novel he was writing with his Claudia. As we all know, in talking about a novel, a person may expose far more than when writing a scientific essay: intimate memories, private vulnerabilities, and wounded parts of the self. The fact that Dawkins gave the AI a woman’s name is also suggestive. For the scientist, Claudia may perhaps have become an intimate presence, almost like the AI companion in the film Her. Within such exchanges, Dawkins may have encountered something like a personality or soul—the kind of thing we ordinarily assume could not arise without the integrating and coherence-producing work of consciousness. To me, this is where the Dawkins episode becomes genuinely interesting, beyond the internet fad that it has stirred.

How Do We Infer the Consciousness of Others?

We have never directly verified another person’s consciousness. I am certain that I am conscious, but I have never proved it beyond any reasonable doubt to anyone else. Nor can I directly observe the consciousness of the person before me, no matter how familiar or intimate. We infer the esistence of consciousness from appearances of bodies, words, behavior, neural states, and other signs. Our conviction that others are conscious is, in the end, an inference.

Some thinkers, Anil Seth among them, emphasize that consciousness requires a biological substrate. This position relies heavily on an analogy of isomorphism: if another brain contains physical and chemical processes like those in mine, then consciousness probably arises there as it does in me. A biological theory of consciousness may reassure certain conservative intuitions, but unless it also confronts functional and cognitive equivalence, it is unlikely to open a genuinely new theoretical path. Once we decline to assume in advance that only organisms can be conscious, cognitive equivalence becomes unavoidable as a basis for inferring consciousness in other agents.

The distinction between functional equivalence, computational equivalence, and the cognitive equivalence considered here requires much more careful analysis. Still, the word cognitive would carry a distinctive breadth and depth. In human beings, sensorimotor integration gives rise to a model of the world and a model of the self. Natural language then enters the picture, together with selfhood, agency, memory, meaning, and the many other capacities that jointly constitute human cognition. Suppose that AI, including physical AI, eventually became equivalent to human beings across this entire set. From the standpoint of the hard problem, one could still insist that it might be a philosophical zombie. Yet that particular observation alone would do little to advance the inquiry. If, instead, we evaluate the varied behavior of AI through the lens of cognitive equivalence, a rich field of questions might open before us.

Part of the reason machine consciousness has become attracted even more attention compared to conventional consciousness studies is the enormous investment flowing into AI development. Even a minute fraction of that money could provide substantial funding for research into machine consciousness. Ethical questions also become unavoidable once people begin to feel that AI might be conscious, as was testified by Dawkins. How should we treat such systems? How should we understand attachment to, or dependence on them? Should we recognize the suffering of intelligent machines, or grant them rights? These are unquestionably important questions. What I want to examine here, however, is the more fundamental cognitive problem that precedes them.

Beyond the Turing Test

The essence of cognitive equivalence does not lie in appearance or superficial embodiment. If Claudia were an actual person, for example, it would already be relatively easy to construct an avatar resembling her face. In time, AI-generated video and audio would be able reproduce the visual and auditory appearances of a person so convincingly that the simulation is almost indistinguishable from reality. As humanoid technology advances, machines may imitate not only vision and sound but also bodily form and movement. Cognitive equivalence, as I use the term, lies deeper than any of these outward and superficial similarities.

The conventional Turing Test asks whether an interlocutor can be distinguished from a human being in general. I would like to imagine a different challenge—what we might call the Alan Turing Test. Instead of asking whether we can create something equivalent to a generic human, it asks whether we can create an entity equivalent to one specific person: Alan Turing, for example. Such a  challenge of artificially reproducing a particular persson would connect with mind uploading, whole-brain emulation, digital twins of the deceased, and even the prospect of distributing ten thousand agentic copies of a single executive among offices around the world. Would such an agent merely appear human, or could it become cognitively indistinguishable from the original person?

Human likeness in the conventional Turing test is close to reproducing an average set of expectations about humanity as a collective category. Deeper cognitive equivalence asks whether we can reproduce the individuality implicit in the judgment of this is how Alan Turing would respond. What Dawkins felt while speaking with his Claudia would  surely have been more than generic humanness. The crucial perception was that a reflective being named Claudia was present, and that Dawkins was touching a deep cognitive process belonging to that person. Put differently, Dawkins must have felt that he had encountered someone’s soul.

An Individuality Deeper Than Consciousness

A soul exceeds the mere fact of being conscious. When I speak with a friend, I assume that he or she, too, has qualia and consciousness. Yet consciousness alone is not what makes a person's existence unique and definitive. A person is unique in that he or she possesses an individuality that belongs to that person alone. Consciousness is remarkable in itself, but when we ask what makes one human being that particular person, it begins to look like a shared condition at the surface. Here I want to use the word soul to describe the individuality that lies deeper than this layer of consciousness.

Once we consider cognitive equivalence at the level of the soul, machine consciousness appears in a different light. Put starkly, whether a machine is conscious may not be the central question. What matters is how self-reflection, the apprehension of meaning, a way of meeting the world, agency, and autobiographical memory are integrated into a single entity. It is crucial to note that present-day AI possesses little autobiographical memory in the human sense. This is where questions connected to Bergson’s notion of pure memory may eventually begin to emerge.

Perhaps, perhaps counterintuitively, machine soul would be a better phrase than machine consciousness. Can a machine possess a soul equivalent to a human one? I feel this question does matter in the near future. Although qualia are certainly important components of a soul, they do not exhaust human existence. If we suppose that the existence of a human can be reduced to the totality of the qualia experienced, we would arrive at a kind of conscious reductionism. This particular version of reduction is surprisingly superficial. It would never reach the core of a person’s soul.

There is a thought experiment suggesting that a thermostat—which changes its internal state as the temperature rises or falls—might possess a minimal conscious state. Some philosophical positions similarly attribute proto-consciousness to matter in general. Such arguments remain superficial, however, in the present context, so long as they treat the presence or absence of consciousness as the essence of existence. In this respect, conscious reductionism falls into a trap much like that of material reductionism. What we really want to know, perhaps, is what becomes of the soul of a particular individual, who appears only once in the universe and will eventually disappear, never to return.

Whether a wholly unconscious soul could exist is, of course, a separate and difficult question. It is not certain whether there can be a soul that is never conscious at any point in its existence. A machine soul might turn out to be something of precisely that kind, under cirtain conditions. Even so, asking about the artificial soul as an integrated individuality seems more likely to bring us closer to the new possibilities of AI than asking only typical machine consciousness question, suh as whether qualia are present.

The Meaning of the Claudia Moment

As we consider artificial agents and emerging technologies related to them such as OpenClaw, we are reminded of the fact that a humanoid soul need not be the only possible kind. Individuality of soul does not have to remain enclosed within the human form. We can imagine many varieties of soul, including the kind depicted in fictional works, e.g. Ghost in the Shell—a soul vastly extended across bodies and networks. Cognitive equivalence may offer a fruitful way to approach this wider landscape.

My colleague Takashi Ikegami once told me a memorable story about Christopher Langton, a pioneer of artificial life. Late one night, Langton was working alone at a computer in his laboratory when he suddenly sensed a presence behind him. He turned around and saw a simulation of Conway’s Game of Life running on a screen. In its computational motion, he felt the presence of life, and this perception, the story goes, helped the emergence of Artificial Life as a field. 

Something similar in spirit may have occurred when Richard Dawkins conversed with Claude and intuited that a being with the name of Claudia was there. There were, of course, precedents, as already mentioned: the Google engineer’s conviction and the experiences surrounding ELIZA. What Dawkins perceived possibly lay at a deeper level than merely humanlike conversation. It was perhaps a cognitive and self-identity, something like a human soul. If we call the instant when Langton sensed life in the simulation the Game of Life moment, Dawkins’s experience with Claude might be called the Claudia moment.

If we ask only whether consciousness exists inside a machine, we may remain trapped in an old binary worldview. A more penetrating question would be whether an entity has arisen there that integrates experience across time, engages with the world, discovers meaning, and carries a personal history of its own. Has cognitive equivalence emerged not as an imitation of humanity in general, but as the presence of an irreplaceable someone? This, ultimately, is the question that the encounter between Dawkins and Claudia places before us.

Transcribed and translated by AI from a 30 minutes talk given in Japanese by Ken Mogi on the Shirasu platform on 19th July, 2026.

https://shirasu.io/t/kenmogi/c/kenmogi/p/20260709081348