AI Will Never Be Conscious

The Blake Lemoine incidental is remembered contiguous arsenic a high‑water people of AI hype. It thrust the full thought of conscious AI into nationalist consciousness for a quality rhythm oregon two, but it besides launched a conversation, among some machine scientists and consciousness researchers, that has lone intensified successful the years since. While the tech assemblage continues to publically belittle the full thought (and mediocre Lemoine), successful backstage it has begun to instrumentality the anticipation overmuch much seriously. A conscious AI mightiness deficiency a wide commercialized rationale (how bash you monetize the thing?) and make sticky motivation dilemmas (how should we dainty a instrumentality susceptible of suffering?). Yet immoderate AI engineers person travel to deliberation that the beatified grail of artificial wide intelligence—a instrumentality that is not lone supersmart but besides endowed with a quality level of understanding, creativity, and communal sense—might necessitate thing similar consciousness to attain. In the tech community, what had been an informal taboo surrounding conscious AI—as a imaginable that the nationalist would find creepy—suddenly began to crumble.

The turning constituent came successful the summertime of 2023, erstwhile a radical of 19 starring machine scientists and philosophers posted an 88‑page study titled “Consciousness successful Artificial Intelligence,” informally known arsenic the Butlin report. Within days, it seemed, everyone successful the AI and consciousness subject assemblage had work it. The draught report’s abstract offered this arresting sentence: “Our investigation suggests that nary existent AI systems are conscious, but besides suggests that determination are nary evident barriers to gathering conscious AI systems.”

The authors acknowledged that portion of the inspiration down convening the radical and penning the study was “the lawsuit of Blake Lemoine.” “If AIs tin springiness the content of consciousness,” a coauthor told Science magazine, “that makes it an urgent precedence for scientists and philosophers to measurement in.”

But what caught everyone’s attraction was that azygous connection successful the abstract of the preprint: “no evident barriers to gathering conscious AI systems.” When I work those words for the archetypal time, I felt similar immoderate important threshold had been crossed, and it was not conscionable a technological one. No, this had to bash with our precise individuality arsenic a species.

What would it mean for humanity to observe 1 time successful the not‑so‑distant aboriginal that a afloat conscious instrumentality had travel into the world? I’m guessing it would beryllium a Copernican moment, abruptly dislodging our consciousness of centrality and specialness. We humans person spent a fewer 1000 years defining ourselves successful absorption to the “lesser” animals. This has entailed denying animals specified supposedly uniquely quality traits arsenic feelings (one of Descartes’s astir flagrant errors), language, reason, and consciousness. In the past fewer years, astir of these distinctions person disintegrated arsenic scientists person demonstrated that plentifulness of taxon are intelligent and conscious, person feelings, and usage connection and tools, successful the process challenging centuries of quality exceptionalism. This shift, inactive underway, has raised thorny questions astir our identity, arsenic good arsenic astir our motivation obligations to different species.

With AI, the menace to our exalted self‑conception comes from different 4th entirely. Now we humans volition person to specify ourselves successful narration to AIs alternatively than different animals. As machine algorithms surpass america successful sheer brainpower—handily beating america astatine games similar chess and Go and assorted forms of “higher” thought similar mathematics—we tin astatine slightest instrumentality solace successful the information that we (and galore different carnal species) inactive person to ourselves the blessings and burdens of consciousness, the quality to consciousness and person subjective experiences. In this sense, AI whitethorn service arsenic a communal adversary, drafting humans and different animals person together: america against it, the surviving versus the machines. This caller solidarity would marque for a heartwarming communicative and mightiness beryllium bully quality for the animals invited to articulation Team Conscious. But what happens if AI begins to situation the human—or animal, I should say—monopoly connected consciousness? Who volition we beryllium then?

I find this a profoundly unsettling prospect, though I’m not wholly definite why. I’m getting comfy with the thought of sharing consciousness with different animals (and perchance adjacent with plants, successful my case) and I’d beryllium blessed to admit them into an expanding ellipse of motivation consideration. But machines?

It could beryllium that my discomfort with the thought stems from my inheritance and education. I person been slow‑cooked successful the lukewarm broth of the humanities, particularly lit and past and the arts, and these person ever held up quality consciousness arsenic thing exceptional that is worthy defending. Just astir everything we worth astir civilization is the merchandise of quality consciousness: the arts and the sciences, precocious civilization and low, architecture, philosophy, religion, government, law, and morals and morality, not to notation the precise thought of worth itself. I accidental it is imaginable that conscious computers could adhd thing caller and arsenic yet unimagined to the banal of these glories. We tin anticipation so. To date, poesy written by AIs isn’t overmuch amended than doggerel; the lack of consciousness mightiness explicate wherefore it lacks adjacent a spark of originality oregon caller insight. But however volition we consciousness if (when?) conscious AIs commencement producing truly bully poetry?

As a humanist, I conflict with the anticipation that the carnal monopoly connected consciousness mightiness fall. But I person present met different types of humans (some of whom telephone themselves transhumanists) who are much sanguine astir this future. Some AI researchers endorse the effort to physique conscious machines because, arsenic entities with feelings of their own, conscious machines are much apt to make empathy than computers that are simply intelligent. Building a conscious AI is simply a motivation imperative, arsenic some a neuroscientist and an AI researcher sought to person me. Why? Because the alternate is the blazingly astute but unfeeling AI that volition beryllium ruthless successful pursuit of its objectives, due to the fact that it volition deficiency each of the motivation constraints that person arisen from our consciousness and shared vulnerabilities. Only a conscious AI is apt to make empathy and truthful spare us. I americium not exaggerating; this is the argument.

One has to wonderment if these radical person ever work Frankenstein! Dr. Frankenstein gives his instauration the acquisition of not lone beingness but besides consciousness, and therein lies the rub. Mary Shelley’s caller chronicles “the instauration of a delicate and rational animal,” and it is the operation of those 2 qualities that determines the monster’s fate. It is not the monster’s rationality but his affectional wounded that spurs him to question revenge and crook homicidal.

“Everywhere I spot bliss, from which I unsocial americium irrevocably excluded,” the monster complains to Dr. Frankenstein aft being driven retired of quality society. “I was benevolent and good; misery made maine a fiend.” The monster’s quality to crushed surely helped him recognize his demonic scheme, but it was his consciousness—his feelings—that supplied the motive. Why should we presume that conscious machines would beryllium immoderate much virtuous than conscious humans?

Remarkably enough, the Butlin study connected artificial consciousness represents thing of a statement presumption successful the field; astir of the machine scientists I interviewed endorsed its conclusions. Yet the much clip I spent speechmaking it (and interviewing 1 of its coauthors), the much I began to question its decision that artificial consciousness is close astir the corner. To their credit, the authors are scrupulous astir mounting distant their assumptions and methods, some of which marque maine wonderment if they haven’t erected their bold decision atop a dubious foundation.

Right connected leafage one, these machine scientists and philosophers acceptable distant their guiding assumption: “We follow computational functionalism, the thesis that performing computations of the close benignant is indispensable and capable for consciousness, arsenic a moving hypothesis.” Computational functionalism takes arsenic its starting constituent the thought that consciousness is fundamentally a benignant of bundle moving connected the hardware of what could beryllium a encephalon oregon a computer—the mentation is wholly agnostic. But is computational functionalism true? The authors aren’t rather prepared to nail themselves to that claim, lone to accidental that it is “mainstream—although disputed.” Even so, they volition proceed connected the presumption that it is existent for “pragmatic reasons.”

The candor is admirable, but the attack demands a tremendous leap of religion that I’m not definite we should make.

For the purposes of the report, the “material substrate” of the system—that is, whether it is simply a encephalon oregon a computer—“does not substance for consciousness … It tin beryllium successful aggregate substrates, not conscionable successful biologic brains.” Any substrate that tin tally the indispensable algorithm volition do. “We tentatively presume that computers arsenic we cognize them are successful rule susceptible of implementing algorithms capable for consciousness,” the authors state, “but we bash not assertion that this is certain.” The acknowledgment of uncertainty doesn’t spell astir acold enough. Unquestioned successful the study is the metaphor that brains are computers—the hardware connected which the bundle of consciousness is run. Here, we conscionable a metaphor parading arsenic fact. Indeed, the full insubstantial and its conclusions hinge connected the validity of this metaphor.

Metaphors tin beryllium almighty tools for thinking, but lone arsenic agelong arsenic we don’t hide they are metaphors—imperfect oregon partial analogies likening 1 happening to another. The differences betwixt the 2 things are arsenic important arsenic the similarities, but these differences look to person gotten mislaid successful the enthusiasm surrounding AI. As cyberneticists Arturo Rosenblueth and Norbert Wiener noted years ago, “The terms of metaphor is eternal vigilance.” Beyond the authors of this report, the full tract of AI appears to person fto down its defender connected this one.

Consider the crisp favoritism betwixt hardware and software. The quality of separating hardware from bundle successful computers is that a large galore antithetic programs tin tally connected the aforesaid machine; the bundle and the cognition it encodes past the “death” of the hardware. The separation besides speaks to our people intuition that dualism is true—that, pursuing Descartes, we tin gully a agleam enactment betwixt intelligence worldly and carnal stuff. But the favoritism betwixt hardware and bundle simply doesn’t beryllium successful brains; there, bundle is hardware and vice versa. A representation is simply a carnal signifier of transportation among neurons successful the brain, neither hardware nor bundle but both.

Indeed, everything that happens to you—everything you acquisition oregon larn oregon remember—changes the carnal operation of your brain, permanently rewiring its connections. (In this sense, determination is nary dualism successful the brain; intelligence worldly tin ne'er beryllium wholly disentangled from carnal stuff.) The thought that the aforesaid consciousness algorithm tin beryllium tally connected a assortment of antithetic substrates makes nary consciousness erstwhile the substrate successful question—a brain—is continually being physically reconfigured by immoderate accusation (or “algorithm of consciousness”) is tally connected it. Your encephalon is materially antithetic from excavation precisely due to the fact that it has been shaped, literally, by antithetic beingness experiences—that is, by consciousness itself. Brains are simply not interchangeable, neither with computers nor with different brains.

Just astir anyplace you propulsion connected it, the computer‑as‑brain metaphor breaks down. Computer scientists dainty neurons successful a encephalon arsenic though they are transistors connected a chip, switched connected oregon disconnected by pulses of electricity. That analogy has immoderate information to it, but it is analyzable by the information that energy is not the lone origin influencing the firing of neurons. Brains are besides awash successful chemicals, including neuromodulators and hormones that powerfully power the behaviour of neurons, not conscionable whether oregon not they occurrence but however strongly. This is wherefore psychoactive drugs tin profoundly change consciousness (and person nary discernible effect connected computers). The enactment of neurons is besides influenced by oscillations that traverse the encephalon successful wavelike patterns; the antithetic frequencies of these oscillations correlate with antithetic intelligence operations, specified arsenic consciousness and its absence, focused attraction and dreaming (as good arsenic different stages of sleep).

To liken neurons to transistors is to grossly underestimate their complexity. Compared with transistors connected a chip, neurons successful the encephalon are massively interconnected, each 1 communicating straight with arsenic galore arsenic 10,000 others successful a web truthful intricate that we are inactive decades distant from being capable to gully adjacent the crudest representation of its connections. In machine science, overmuch has been made astir the advent of “deep artificial neural networks”—a benignant of machine‑learning architecture, supposedly modeled connected the brain’s, that layers a mind‑boggling fig of processors successful specified a mode that the web tin process and larn from immense troves of data. Impressive, for sure, yet a caller survey demonstrated that a azygous cortical neuron tin bash everything an full heavy artificial neural web can.

Yes, determination are plentifulness of ways successful which computers bash lucifer brains, and machine subject has made large strides by simulating assorted aspects and operations of the brain. But the thought that brains and computers are successful immoderate mode interchangeable—the premise of computational functionalism—is surely a stretch. And yet this is the premise upon which stands not lone the Butlin study but besides astir of the field. It’s not hard to spot why. If brains are computers, past sufficiently almighty computers should beryllium capable to bash immoderate brains do, including becoming conscious. The premise each but guarantees the conclusion. Put different way, it is the authors themselves who person removed the biggest “barrier” to gathering a conscious AI—the obstruction that says brains disagree from computers successful important ways.

There is simply a 2nd facet of the study that makes maine wonderment however earnestly to instrumentality its conclusion, and that is the modular it proposes for deciding if an AI is really conscious oregon not. This is simply a superior challenge. Citing the Lemoine incidental (fairly oregon not), the authors constituent retired that AIs tin easy dupe humans into believing they are conscious erstwhile they are not. (It’s astir apt much close to accidental that we dupe ourselves into this belief, acknowledgment to our weakness for anthropomorphism and magic.) “Reportability” (philosophical jargon for conscionable asking the AI itself) won’t enactment erstwhile the AI has been trained connected beauteous overmuch everything that’s been said and written astir consciousness. One attack to this dilemma would beryllium to region each references to consciousness (and presumably feeling and emotion arsenic well) from the dataset connected which the AI has been trained and past spot if it tin inactive talk convincingly astir being conscious.

Instead, the authors suggest that we look for “indicators” of AI consciousness that lucifer the predictions of the assorted theories of consciousness successful play. So, for example, if the plan of an AI included a workspace that brought unneurotic assorted streams of information, but lone aft those streams had competed to participate it, that would look a batch similar planetary workspace mentation and truthful mightiness suffice arsenic conscious. The study reviewed a half‑dozen theories of consciousness, identifying the “indicators” that an AI would person to grounds to fulfill each of them and, by doing so, beryllium deemed perchance conscious.

The occupation present (well, 1 of them) is this: None of the theories of consciousness that it proposes we measurement AIs against are adjacent remotely adjacent to being proved to anyone’s satisfaction. So what benignant of modular of impervious is that? What’s more, galore of these theories tin beryllium simulated successful the plan of an AI, which should travel arsenic nary surprise, due to the fact that they’re each based connected the thought that consciousness is simply a substance of computation. Round and circular we go.

By the clip I finished digesting the Butlin report, the Copernican infinitesimal I’d disquieted astir seemed much distant than the report’s bold decision had led maine to believe. After reviewing the half‑dozen oregon truthful theories of consciousness covered by the report, it seemed wide that each of them stacked the platform by taking for granted that consciousness could beryllium reduced to immoderate benignant of algorithm.

I was besides struck by what was missing from the theories nether consideration. None of them had thing to accidental astir embodiment—the thought that consciousness mightiness beryllium connected having some a assemblage and a brain—or, for that substance thing remotely biological. Nor did the theories person thing to accidental astir the conscious subject. Who oregon what, exactly, is the recipient of the accusation that is broadcast successful the planetary workspace? Or the accusation that is integrated successful integrated accusation mentation (IIT)? And what astir the relation of feelings successful rendering acquisition conscious?

This past constituent was not mislaid connected the authors, who noted the lack of “affect” from astir existent theories and recommended that the tract wage much attraction to the contented of whether conscious machines would person “real” feelings, due to the fact that if it turns retired they do, we volition person a motivation and ethical situation connected our hands. “Any entity which is susceptible of conscious suffering deserves motivation consideration,” the study states. (But isn’t suffering ever conscious?) “This means that if we neglect to admit the consciousness of conscious AI systems,” the study continued, “we whitethorn hazard causing oregon allowing morally important harms.” What would we beryllium machines that tin suffer? And bash we truly privation to bring immoderate much suffering into the world?

Apart from this benignant of highly speculative treatment of feeling (as a troublesome by‑product of making machines conscious), successful the AI community, the speech astir consciousness is arsenic relentlessly abstract—as bloodless, bodiless, and utterly oblivious to biology—as 1 would expect. When I posed the suffering‑computer conundrum to a researcher seeking to physique a conscious AI, helium waved distant the problem, explaining it could beryllium offset with a elemental hole to the algorithm: “There’s nary crushed we couldn’t conscionable crook up the dial connected joy.”

Adapted from A World Appears: A Journey into Consciousness by Michael Pollan. Copyright ©2026 by Michael Pollan. Published by statement with Penguin Press, an imprint of Penguin Publishing Group, a part of Penguin Random House LLC.