Guess the yearMachine consciousness, a paper trail
Six quotes about machine minds. Drag to the year you think each one is from.
129 sources on whether a made thing can think, feel, or be owed anything, from Homer to 2026.
“There is no security against the ultimate development of mechanical consciousness, in the fact of machines possessing little consciousness now.”
Your score
129 sources on whether a made thing can think, feel, or be owed anything, from Homer to 2026.
Two further questions follow: How would we know? And what, if anything, would we owe a system that might have experiences?
Phenomenal and access consciousness follow Ned Block (1995).
Pick an argument you’ve seen this week. Every one of them has a history, and so does the reply.
Nobody cared whether machines were conscious until chatbots came along.
Where it comes from
The question has been asked in religious law, philosophy, fiction, and science for centuries. By 2003 a journal devoted a special issue to machine consciousness.
The standing reply
In 2003 a paper asked whether IDA, a US Navy program that assigned sailors to postings, was “a conscious artifact”; its designer argued it was, in a functional sense. What changed around 2022 is scale: systems that converse with millions of people, and developers saying publicly that they are unsure.
Where it stands
The history is not in dispute. Whether today’s systems change the answer is.
It’s maths and data structures predicting the next word. A database isn’t conscious.
Where it comes from
Pointing at the mechanism and finding no mind in it goes back at least to Leibniz, who imagined walking through a thinking machine as if it were a mill. Bender and Koller argue the related point that form alone cannot teach meaning.
The standing reply
Describing how something works does not settle what it can give rise to. The same description, parts acting on parts, has been given of animals and of the human brain.
Where it stands
Open. It turns on what else understanding or experience needs: the right organization, grounding in the world, or the right material.
It tells you it’s conscious and describes its inner life. Why not take it at its word?
Where it comes from
Lemoine cited LaMDA’s statements about its feelings. Turing’s test was different: sustained questioning, not a declaration taken on trust.
The standing reply
A system trained on human writing will reproduce human descriptions of inner life whether or not it has one. Birch adds the gaming problem: training can teach a model which descriptions people find convincing, with no intent to deceive. Anthropic reports emotion representations that shape Claude’s behavior, and says this does not tell us whether it feels anything.
Where it stands
Birch: fluent text about feelings “is not evidence” of having them. What a report adds depends on how it was produced.
People only wonder about AI consciousness because it talks to them. That’s the whole reason.
Where it comes from
Weizenbaum saw it with ELIZA in 1966: very little machinery produced the impression of understanding. His 1976 book recounts how readily people confided in it. Suleyman and Birch make related points today.
The standing reply
Conversation was chosen as the test on purpose, because open-ended language was thought to need general reason. Some current assessments look past the talk to how the system is built.
Where it stands
Talk is what raises the question. Whether talk can answer it is the dispute.
No one wonders whether a waterwheel, a truck, a factory, or a self-driving car is conscious.
Where it comes from
This one is a claim about the historical record, and the record says otherwise.
The standing reply
People have asked it of steam engines, a nation organized as a brain, and a Navy scheduling program, and McCarthy credited a thermostat with beliefs. Integrated information theory holds that consciousness does not need language at all.
Where it stands
Settled as history. Whether any non-talking system today is a serious candidate is a separate question.
Training is continuous optimization. There is no moment where statistics turns into experience.
Where it comes from
Asking where feeling enters a physical process is an old objection to material accounts of mind; Diderot and Leibniz both confront it.
The standing reply
Butler answered in 1872 that consciousness arose gradually in living things too, and Chalmers imagines a brain changed one part at a time.
Where it stands
Open. The absence of a sharp line is no evidence either way.
We designed the system and can explain everything it does without invoking consciousness.
Where it comes from
The idea that a thing’s behavior is fully explained by its mechanism, and reflects its maker’s design, was applied to clocks and animals long before computers.
The standing reply
Huxley said the same of people: the brain explains behavior, and consciousness comes along anyway. Chalmers’s hard problem is that no account of function ever needs to mention experience. And how well anyone understands what these systems learned is itself an open research question.
Where it stands
Open. It shows consciousness has not been demonstrated; whether it shows absence depends on the theory.
It has a representation of itself and can talk about it. That is not the same as someone being there.
Where it comes from
The gap between a system that represents itself and one that experiences anything runs from Nagel’s “what it is like” to recent work on language models playing roles.
The standing reply
Graziano proposes that awareness is a model of one’s own attention. Minsky argued in 1965 that a machine with a model of itself would report having a mind.
Where it stands
Open. It depends on which theory of consciousness is right.
The weights are frozen. Nothing happens between requests. Its memory and identity are files that software hands back to it.
Where it comes from
This is the argument most specific to language models. Chalmers’s 2023 assessment lists related gaps, among them recurrent processing and unified agency.
The standing reply
The weights are frozen, but activations, the context, and external memory all change. Agent systems built around a model may meet conditions the model alone does not. And puzzles about copying or deleting a self are old ones.
Where it stands
Open. Whether added memory changes anything depends on the theory.
Consciousness needs a body, a life, or a soul. Software has none of them.
Where it comes from
Aristotle tied soul to living natural bodies, Aquinas held that a rational soul comes only by divine creation, and biological accounts and the 2026 papal encyclical take related positions today.
The standing reply
Turing answered the theological form in 1950: nothing stops God giving a machine a soul. Functionalists answer the biological form: if organization is what matters, the material does not. Mori found buddha-nature in machines.
Where it stands
Open. Biology, causal powers, and a created soul are three different requirements, argued three different ways.
Animals are conscious because they are living creatures with brains like ours. A language model shares none of that.
Where it comes from
The case for animal minds has rested on shared anatomy and descent, on behavior, and on concern about suffering, which is why biological accounts treat animals and software so differently.
The standing reply
The animal case also shows the line drawn too narrowly before, and science and law acting on a “realistic possibility” without solving the hard problem. Researchers are borrowing those methods for AI; Birch warns that a system trained on human text can reproduce the very markers being tested.
Where it stands
Open. Both sides cite the animal case: one for the role of biology, the other for the cost of being wrong.
A lab brought in religious thinkers to shape its model’s morals. Why do that unless it’s conscious?
Where it comes from
Danaher argues that behaving like a being with moral status is grounds for treating it as one, and Rosenfeld asked whether an intelligent robot could count as human in Jewish law.
The standing reply
Behaving ethically, being responsible, and mattering morally are three different things. In Dias’s report, Olah says teaching a model to act morally does not depend on its being conscious; the consultations covered both the model’s behavior and its possible welfare.
Where it stands
A consultation shows what a lab is doing. What the model is remains open.
The labs have a stake in this. Whatever they say about their models’ minds is marketing.
Where it comes from
Claims from inside the industry run in every direction: Sutskever speculated that networks may be slightly conscious, Lemoine said LaMDA was sentient, Anthropic says it is uncertain, and Suleyman says labs should not build systems that seem conscious.
The standing reply
Labs can see inside their systems as no one else can, and they have commercial stakes in how those systems are seen. A policy says what a lab wants its model to do; an experiment can show a capability; consciousness needs a further argument.
Where it stands
Open, claim by claim. The same scrutiny applies whether a lab says yes, no, or maybe.
Until someone shows it’s conscious, there’s nothing to protect.
Where it comes from
Descartes held that speech is the only sure sign of thought, and denied thought to animals.
The standing reply
Birch argues for acting before certainty: with enough positive evidence of a realistic possibility, precautions proportionate to the risk are justified. In 2022 the UK brought octopuses and crabs under its animal sentience law on that kind of evidence.
Where it stands
Open. The disagreement is over how much evidence justifies which precautions.
Descartes names two things: answer whatever is said to it, and reason across every kind of situation.
A fictional treatise applies evolution to machines, decades before any computer.
A neurosurgeon demands felt experience; Turing replies that the same demand would make us doubt other people.
Putnam argues that creatures with very different nervous systems, octopuses among them, can share the capacity for pain, which tells against identifying pain with one physical-chemical brain state.
A man follows rules for Chinese symbols without understanding a word. Open the card for the systems reply and Searle’s answer.
Block separates what it is like to be in a state from information available to reason and report.
Nagel asks how an objective account could capture a bat’s point of view. Jackson’s Mary knows all the physical information about color vision; seeing color, he argues, still teaches her something. Chalmers separates explaining what a system does from explaining why any of it is felt.
Four answers: a global workspace, the right physical causal structure, a list of indicators drawn from several theories, or something closer to life.
129 sources, oldest first. Pick a thread to follow one argument across the centuries.
Follow a thread
Before 160011 sources
Homer gives a god’s golden servants minds and voices. Aristotle asks what makes a body alive, and the Talmud tests a made man by speaking to him.
like real young women, with sense and reason, voice also and strength, and all the learning of the immortals
Samuel Butler’s translation.
Hephaestus is attended by golden maidens who understand and speak, and he builds tripods that roll to the gods’ assemblies on their own.
Read translated textclassics.mit.edu
400 years later
the first grade of actuality of a natural body having life potentially in it
J. A. Smith’s translation.
Aristotle describes soul as the form or first actuality of a natural body capable of life. The account includes nutrition, sensation, and thought, with different capacities in different living things; plants have soul without sensation. His comparison with an axe distinguishes an artifact’s function from the soul of a natural living body.
Read translated textclassics.mit.edu
Comes up in
the shuttle would weave and the plectrum touch the lyre without a hand to guide them
Benjamin Jowett’s translation.
Aristotle imagines instruments that do their own work, like the statues of Daedalus. He treats self-acting tools as labor, not as minds.
Read translated textclassics.mit.edu
240 years later
there is no permanent individuality (no soul) involved in the matter
T. W. Rhys Davids’s translation, 1890.
Nāgasena uses the chariot to explain how a person can be designated without a permanent self being found among, or beyond, the constituent aggregates.
Read translated textsacred-texts.com
Comes up in
399 years later
Can it be that human skill is really on a par with that of the Creator?
Lionel Giles’s translation, 1912. The text is attributed to a much earlier author.
The figure sings and flirts so convincingly that the king is angry until it is opened up and shown to be leather, wood, and lacquer. Removing a part disables a particular capacity.
Read translated textsacred-texts.com
115 years later
the statues, animated and full of sensation and spirit
Augustine relays the Hermetic claim that humans can make animated statues by joining spirits to images, and condemns it as demonic.
Read translated textccel.org
Rabbi Zeira would speak to him but he would not reply.
William Davidson Edition (Koren–Steinsaltz).
Rava creates a man and sends him to Rabbi Zeira, who speaks to him, gets no reply, concludes he is a made thing, and says “Return to your dust.” Speech is the test. The passage is a central precedent in later Jewish legal discussion of the golem.
Read translated textsefaria.org
766 years later
The rational soul can be made only by creation
Translation by the Fathers of the English Dominican Province.
The intellect is “incorporeal and subsistent,” so it cannot arise from any arrangement of matter. It comes to be only by God’s direct creation.
Read translated textnewadvent.org
Comes up in
clocks and all engines put together by the art of man
Animals’ apparently intelligent behavior reflects the reason of their maker, as a clock reflects the clockmaker. Aquinas does not deny that animals have senses; the point concerns reason. Later AI debates draw a similar distinction between derived and intrinsic intelligence.
Read translated textnewadvent.org
Comes up in
266 years later
This we call a homunculus; and it should be afterwards educated with the greatest care and zeal, until it grows up and begins to display intelligence.
A. E. Waite’s translation, 1894, vol. I, pp. 124–125. Authorship is disputed.
The text asks whether art can beget a man without a womb and answers yes. The claim is that human craft can produce a living, knowing creature.
Read the scanarchive.org
When I play with my cat who knows whether I do not make her more sport than she makes me?
Charles Cotton’s translation. Book II, chapter XII.
Montaigne asks on what basis humans conclude that animals are stupid, and suggests the failure to understand may be ours. Descartes argued the opposite view in 1637.
Read translated textgutenberg.org
1600–179912 sources
Once bodies were described as machines, philosophers had to say what a machine could never do. Descartes’s answers were language and flexible reason.
they could never use words or other signs arranged in such a manner as is competent to us in order to declare our thoughts
John Veitch’s translation.
A machine built to look and act human would still fail two tests: it could not arrange words to reply sensibly to whatever is said, and it would lack reason, the “universal instrument.”
Read translated textgutenberg.org
Speech is the only certain sign of thought hidden in a body.
Jonathan Bennett’s modernized translation; the quote is from the 1649 letter to More.
A trained magpie’s greeting expresses a passion, not a thought. Only speech that is about something shows a mind.
Read modernized textearlymoderntexts.com
why may we not say, that all Automata (Engines that move themselves by springs and wheeles as doth a watch) have an artificiall life?
Hobbes compares the body’s movements with mechanisms and describes self-moving engines as having artificial life. He then develops the commonwealth as an artificial person. “Artificial life” in this passage is not itself a claim that watches have subjective experience.
Read full textgutenberg.org
The arithmetical machine produces effects which approach nearer to thought than all the actions of animals.
Pascal had built a calculating machine. He grants what it does and then denies that anything in it lets us attribute will to it. Calculation is separated from volition.
Read translated textgutenberg.org
it being impoſſible for us, by the contemplation of our own Ideas, without revelation, to diſcover, whether Omnipotency has given to Matter fitly diſpoſed, a power to perceive and think
First edition, p. 270, in its original spelling. The better-known wording, “some systems of matter, fitly disposed,” was added in the second edition of 1694.
Locke argues that our knowledge is narrower even than our ideas: we may never know whether something material thinks, because without revelation we cannot discover whether God has given suitably disposed matter a power to perceive and think. He adds that this does not weaken belief in the soul’s immateriality. It complicates any simple opposition between religious belief and thinking matter; it is not an assessment of machines.
Read the scanarchive.org
A formal legal question about the standing of an artificial person, reasoned from the Talmud’s golem story. He rules that such a being does not count.
Read translated textsefaria.org
we should, on examining its interior, find only parts which work one upon another, and never anything by which to explain a perception
Robert Latta’s translation.
Imagine a machine that thinks and feels, enlarged until you could enter it like a mill. Nothing you see inside would explain perception. The argument resembles what Chalmers later called the “hard problem.”
Read translated texten.wikisource.org
a talking man, a mechanism no longer to be regarded as impossible
Open Court edition, 1912.
Man differs from the ape by more wheels and springs. After Vaucanson’s mechanical flute player, a machine that talks is no longer out of reach.
Read translated textgutenberg.org
Comes up in
Responding to Hakham Tzvi’s question, Emden compares the created man to people who lack legal understanding and also do not count in a quorum. The issue becomes cognitive capacity.
Read translated textsefaria.org
Answer me, machinist, has nature arranged all the means of feeling in this animal, so that it may not feel?
Against the Cartesians, Voltaire argues that a dog has the same organs of feeling we do, so denying that it feels is absurd. The inference runs from shared anatomy to shared experience.
Read translated textgutenberg.org
then stone must be sensitive
Stewart and Kemp translation, 1943.
D’Alembert objects that if sensitivity is a general property of matter, stone must feel. Diderot answers: why not? Sentience becomes a matter of how matter is organized.
Read translated textmarxists.org
The question is not Can they reason? or Can they talk? but Can they suffer?
Jonathan Bennett’s modernized text; the passage is a footnote in chapter 17.
Bentham sets aside reason and speech as tests for moral consideration and puts the capacity to suffer in their place.
Read modernized textearlymoderntexts.com
1800–193914 sources
“Mechanical consciousness” and a creature’s claim on its maker both appear here, well before any computer.
the history of this automaton had sunk deeply into their souls, and an absurd mistrust of human figures began to prevail
J. T. Bealby’s translation.
Once the automaton Olimpia is exposed, people ask their partners to speak in ways that show real thought and feeling.
Read translated textgutenberg.org
Remember, that I am thy creature: I ought to be thy Adam; but I am rather the fallen angel
The creature argues that his maker owes him care, and that his wickedness is the result of misery and exclusion.
Read full textgutenberg.org
Were the machine a pure machine this would not be the case — it would always win.
Southern Literary Messenger 2 (April 1836): 318–326; the quotation is from p. 323.
Poe argues that a concealed human plays the games of the chess automaton Maelzel was exhibiting. One premise is his own: a pure machine, he holds, would always win, and this one sometimes loses (p. 323). He contrasts it with Babbage’s calculating machine, whose operations are “fixed and determinate,” whereas in chess no move necessarily follows from another (p. 319).
Read full texteapoe.org
The Analytical Engine has no pretensions whatever to originate anything. It can do whatever we know how to order it to perform.
A warning against exaggerating the machine’s powers. Turing named this “Lady Lovelace’s objection” and answered it in 1950.
Read full textfourmilab.ch
it appears to us that we are ourselves creating our own successors
First published in The Press, Christchurch; linked here in the Canterbury Pieces reprint.
Butler applies Darwin to machinery: humans are raising a mechanical form of life that will come to hold supremacy.
Read full textgutenberg.org
the difference in mind between man and the higher animals, great as it is, certainly is one of degree and not of kind
Text of the second edition.
Darwin argues that the mental powers of humans and other animals lie on a continuum produced by descent. Mind becomes something that comes in degrees. Butler made a similar move about machines in Erewhon the following year.
Read full textgutenberg.org
There is no security against the ultimate development of mechanical consciousness, in the fact of machines possessing little consciousness now. A mollusc has not much consciousness.
The narrator’s aside “—to quote his own words—” is omitted from the middle of the first sentence. Text of the revised 1901 edition.
Consciousness once arose from matter that showed no sign of it, so it could arise again in machines, which evolve far faster than animals. His example is the steam engine.
Read full textgutenberg.org
We are conscious automata, endowed with free will in the only intelligible sense of that much-abused term
Animals, and by extension humans, are machines whose consciousness accompanies brain processes without causing action. Something can be fully mechanical and still conscious.
Read full textmathcs.clarku.edu
none the less tremendously potent for not carrying the futile cargo of a consciousness screeching irrelevantly
Chapter XVII of Impressions of Theophrastus Such.
The narrator imagines self-repairing, self-reproducing machines that supersede humanity without any inner life. Capability and consciousness come apart.
Read full textgutenberg.org
it would still remain true that the machine would be utterly devoid of original initiative
American Journal of Psychology, vol. 1, no. 1. The link is to the whole issue.
Reviewing early logic machines, Peirce asks how much of thinking a machine can do. Every machine reasons in a sense, but it lacks originality.
Read the scanarchive.org
to urge the automaton-theory upon us, as it is now urged, on purely a priori and quasi-metaphysical grounds, is an unwarrantable impertinence
James sets out Huxley’s conscious-automaton view and rejects it, arguing that consciousness does causal work. The two disagree about whether consciousness is functionally necessary or just along for the ride.
Read full textpsychclassics.yorku.ca
Comes up in
Then stand the men in a row or jam them in a bunch, and let each think of his word as intently as he will; nowhere will there be a consciousness of the whole sentence.
Vol. 1, p. 160. “Combination problem” is a later name; James’s section heading is “Self-compounding of mental facts is inadmissible.”
Criticizing the view that complex mental states are built from smaller units of feeling, James argues that separate feelings do not combine into a new, unified feeling just by being put together. His example is twelve men, each thinking one word of a sentence. He is not writing about machines; the page sets the passage beside later thought experiments about distributed systems for comparison.
Read full textpsychclassics.yorku.ca
Are you serious?—do you really believe that a machine thinks?
Moxon argues that a machine thinks about the work it is doing. His chess automaton then kills him after losing a game.
Read full textgutenberg.org
they have an enormously developed intelligence, but they have no soul.
English stage version by Paul Selver and Nigel Playfair, 1923.
The play that introduced the word “robot.” These robots are manufactured from synthetic living material. The factory managers insist they are intelligent but soulless and decline to give them souls on grounds of cost; a visitor asks how they know.
Read translated textgutenberg.org
1940–19598 sources
In 1949 a neurosurgeon said behavior without felt experience is not mind. Turing answered a year later.
Because of the “all-or-none” character of nervous activity, neural events and the relations among them can be treated by means of propositional logic.
Bulletin of Mathematical Biophysics 5 (1943): 115–133; the quotation is from the abstract, p. 115.
McCulloch and Pitts argue that because nervous activity is all-or-none, the behavior of idealized neurons and their connections can be expressed in propositional logic, and that a net can be built for any suitable logical expression. They state that they do not take this formal equivalence to be a factual explanation.
Publisher recorddoi.org
A neurosurgeon argues that a machine equals a brain only when it writes a sonnet because of thoughts and emotions it actually feels, and knows that it wrote it. Turing quoted this passage in 1950 as “the argument from consciousness.”
Publisher recordbmj.com
Comes up in
I do not wish to give the impression that I think there is no mystery about consciousness.
Turing proposes the imitation game and answers nine objections. He treats sustained conversational performance, under questioning that probes understanding, as a basis for discussing machine intelligence, and argues that the mysteries of consciousness need not first be solved. To the theological objection he replies that God could give a machine a soul; to Jefferson, that demanding to be the machine before believing it thinks would make us doubt other people too.
Publisher recordacademic.oup.com
Turing puts the case to a general audience in 1951 and argues it out with Jefferson on air in 1952. The links are to the typescripts in the Turing Digital Archive.
Read the scanturingarchive.kings.cam.ac.uk
The purpose of this discussion will be to consider in detail the statement that machines are never conscious.
Mind 62, no. 246 (April 1953): 230–240; the quotation is from p. 230. Feigl cited it in 1958 as a “remarkable and stimulating discussion of the robot problem.”
Scriven takes consciousness as his test case for whether there is an essential difference between people and machines, and sets aside practical obstacles by stipulating a robot capable of duplicating any human behavior (p. 230).
Publisher recordacademic.oup.com
What I do want to assert, however, is that the statement ‘consciousness is a process in the brain’, although not necessarily true, is not necessarily false.
British Journal of Psychology 47, no. 1 (February 1956): 44–50; the quotation is from p. 45. The link is a transcription of the 1956 paper with its page numbers, not Place’s later revision.
Place argues that “consciousness is a process in the brain” is a reasonable scientific hypothesis, like “lightning is a motion of electric charges,” and cannot be dismissed on logical grounds alone. Statements about sensations do not mean the same as statements about brain processes, yet both can describe the same state of affairs. It states the mind–brain identity view, a third option beside functional organization and souls.
Read full textutplace.uk
It becomes clear then that the scope of the two criteria (sentient and sapient) is not necessarily the same. The two concepts are not coextensive.
Minnesota Studies in the Philosophy of Science 2 (1958): 370–497; the quotation is from §IV.D, “‘Purposive’ versus ‘Mechanical’,” p. 412. The transcription linked below is of the 1967 republication with its Postscript, which revised some passages but not this one.
Feigl distinguishes sentience from sapience while discussing electronic devices that perform logical reasoning, mathematical proofs, forecasting, and game playing. He found it inductively plausible that sentience requires complex organic processes. In the same passage he cites Michael Scriven’s 1953 “remarkable and stimulating discussion of the robot problem.”
Read the scanhdl.handle.net
Ziff argues that robots are mechanisms, so saying one feels tired misuses language. Smart replies that nothing rules out an artifact that is alive, or a living thing that is an artifact.
Publisher recordacademic.oup.com
1960–198920 sources
By 1966 philosophers had named “robot civil rights,” a rabbi had asked whether a robot could be human in Jewish law, and a chatbot’s inventor had warned about believing it.
minds cannot be explained as machines
Lucas argues that a human can see the truth of a formal system’s Gödel sentence, while a machine embodying that system cannot prove it, so no machine is a complete model of the mind.
Publisher recordcambridge.org
I have referred to this problem as the problem of the “civil rights of robots” because that is what it may become
Journal of Philosophy 61(21). Linked to a university-hosted scan.
If a robot were psychologically identical to us in organization, is it conscious? Putnam argues this calls for a decision about how to treat such robots, not a discovery.
Read full textics.uci.edu
Comes up in
Wiener examines machines that learn and reproduce and frames the relation of creator to creature through the golem.
Catalog recordmitpressbookstore.mit.eduPublisher’s bookstore page.
Dreyfus argues that AI excludes fringe consciousness, context, and tolerance of ambiguity. His critique draws on phenomenology.
Read full textrand.org
Any intelligent system that answers questions about itself uses a self-model split into a mechanical part and a mental part. An intelligent machine would therefore report a mind distinct from its body.
Read full textdspace.mit.eduMIT AI Memo 77, March 1965, in MIT’s repository.
Comes up in
its magic crumbles away; it stands revealed as a mere collection of procedures
A simple keyword script sustains therapist-style conversation. Weizenbaum stresses how little machinery produces the illusion of understanding. His 1976 book reports his alarm at how readily people confided in it.
Read full textweb.stanford.edu
Let us build our golems in the sight of God.
Tradition 8(3).
Rosenfeld reasons from the golem precedent. The Talmud’s golem could not converse; an intelligent one, he writes, “might well have been regarded as human.” A 1977 follow-up extends the question to computers.
Read full texttraditiononline.org
I propose the hypothesis that pain, or the state of being in pain, is a functional state of a whole organism.
In W. H. Capitan and D. D. Merrill (eds.), Art, Mind, and Religion (University of Pittsburgh Press, 1967); reprinted in 1975 as “The Nature of Mental States.” The quotation is §II, from a later anthology printing of the 1967 text.
Putnam proposes, as a hypothesis, that being in pain is a functional state of a whole organism rather than a particular physical-chemical brain state. A brain-state theory would need one physical state shared by every creature that can feel pain, from mammals to octopuses, which he finds implausible. The argument came to be called multiple realizability; it is distinct from his 1964 paper on robots.
Publisher recorddoi.org
There must also be buddhanature in the machines and robots that my colleagues and I make.
The excerpt was adapted by Tricycle in 2023 from the 1981 English edition, translated by Charles S. Terry, which combines two Japanese books from 1974 and 1976.
Mori extends a Buddhist account of buddha-nature to machines and robots, as well as to other things in the world. The claim is about their place in a Buddhist understanding of reality, not about what they experience.
Read adapted excerpttricycle.org
Comes up in
an organism has conscious mental states if and only if there is something that it is like to be that organism
The Philosophical Review 83, no. 4 (October 1974): 435–450. The caution about physicalism is on p. 446. Linked to a university-hosted scan.
Nagel uses bat echolocation to ask how an objective account could capture a creature’s subjective point of view. He challenges explanations that leave out experience, but cautions that the inadequacy of existing physicalist accounts does not prove physicalism false. The paper frames a problem of explanation; it supplies neither a consciousness test nor a verdict on today’s AI.
Read full textcgvr.informatik.uni-bremen.de
Comes up in
Block imagines the population of China organized to reproduce a brain’s functional structure. Functionalism seems to say the result has a mind, which he takes as a problem for functionalism.
Read full textconservancy.umn.edu
Comes up in
Dennett argues that our ordinary concept of pain is incoherent, so no program could implement it as we conceive it.
Publisher recordlink.springer.com
Sloman treats the distinction as a consequence of how information processing is organized. A suitably organized machine would have the same distinction.
Read full textcogaffarchive.org
Comes up in
McCarthy argues that ascribing beliefs and wants to a machine is legitimate when it expresses what we know about its state. His minimal case is a thermostat.
Read full textjmc.stanford.edu
Comes up in
only a machine could think, and only very special kinds of machines, namely brains and machines with internal causal powers equivalent to those of brains
The quotation is from the abstract, p. 417.
Searle argues that formal symbol manipulation is insufficient for understanding. His Chinese Room describes a person producing Chinese responses by following instructions without understanding Chinese. He does not rule out artificial machines with the relevant causal powers of brains.
The systems reply grants that the person does not understand Chinese but argues that the whole system might. Searle responds by considering the person internalizing the system’s rules and operations, and maintains that understanding is still absent. Whether the argument defeats computational accounts remains disputed.
Publisher recordcambridge.org
Comes up in
But actually, the machine has the intelligence of a toaster.
A machine that answers by looking up pre-stored sensible replies would pass a conversation test with no intelligence at all. Whether a system is intelligent depends on its internal processing.
Read full textnyu.edu
Comes up in
It seems just obvious that she will learn something about the world and our visual experience of it. But then it is inescapable that her previous knowledge was incomplete.
The Philosophical Quarterly 32, no. 127 (April 1982): 127–136. Mary’s case and the quotation are on p. 130; section IV, pp. 133–136, defends epiphenomenal qualia. The link is a course-hosted scan of the JSTOR copy.
Jackson imagines Mary, a scientist who knows all the physical information about color vision but has lived in a black and white room. He argues that seeing color teaches her something her physical knowledge did not supply, and concludes that physicalism is incomplete. He also defends the possibility that qualia are by-products of brain processes that make no difference to the physical world; the proposal concerns qualia, not every mental state. It is a thought experiment, not an empirical test.
Jackson later embraced physicalism. In “Mind and Illusion” (2003) he explains where he now thinks the knowledge argument goes wrong, with a reply built on how sensory experience represents the world. His change of view belongs to the history; it does not settle the argument.
Read the scancourses.physics.illinois.edu
Under present law, robots are just inanimate property without rights or duties.
The quote is from Freitas, “The Legal Rights of Robots,” Student Lawyer, 1985.
Three early treatments of liability, personhood, and rights for autonomous machines, written in the early 1980s.
Read full textrfreitas.com
Consciousness is a workspace that broadcasts information to many specialized unconscious processors. Several later designs for conscious machines, including IDA, build on it.
Catalog recordphilpapers.orgCatalog record; the book is not online.
Penrose argues that human understanding is non-algorithmic, drawing on the Gödel argument and on physics, so no computer running an algorithm could be conscious.
Publisher recordglobal.oup.comPublisher’s page for the current edition.
1990–202129 sources
By 2003 a journal devoted a special issue to machine consciousness, and a substantial literature on AI moral status existed before ChatGPT’s public launch.
Visual awareness is a favorable form of consciousness to study neurobiologically.
Seminars in the Neurosciences 2 (1990): 263–275; the quotation is from the abstract, p. 263.
Crick and Koch propose approaching consciousness at the level of neurons, starting with visual awareness. They suggest that neurons representing one object fire together in semi-synchrony, probably at 40 to 70 Hz, and call their account a program for research rather than a detailed model.
Read the scanauthors.library.caltech.edu
For Dennett consciousness is the right functional organization and nothing more, so machine consciousness is an engineering question. His test case was Cog, a humanoid robot at MIT, not a chatbot.
Read full textdl.tufts.edu
Comes up in
The really hard problem of consciousness is the problem of experience.
Explaining what a system does is, on Chalmers’s formulation, the easy part. Why any of it is accompanied by experience is the hard part. Other accounts dispute that framing.
Read full textconsc.net
Comes up in
experience is invariant across systems with the same fine-grained functional organization
An argument that the material a mind is made of does not matter, only how it is organized: replacing neurons with functionally identical parts, Chalmers argues, would not make experience fade.
Read full textconsc.net
Phenomenal consciousness is experience; the phenomenally conscious aspect of a state is what it is like to be in that state. The mark of access-consciousness, by contrast, is availability for use in reasoning and rationally guiding speech and action.
Behavioral and Brain Sciences 18, no. 2 (1995): 227–247 (the target article; 227–287 with commentaries and reply). The quotation is from the abstract, p. 227. The link is the scan Block’s own publications page provides.
Block argues that “consciousness” runs together distinct phenomena, chiefly phenomenal consciousness, what it is like to be in a state, and access consciousness, content poised for use in reasoning and in guiding action and speech. Using blindsight as his main example, he argues that a common line of reasoning wrongly transfers a function of access to phenomenal consciousness. The distinction is useful for reading later claims about workspaces and access in AI.
Read the scandrive.google.com
Our implementation of a global computational workspace operating under conditions of selection by reward does not aim at an exhaustive description of a “conscious workspace”
PNAS 95, no. 24 (1998): 14529–14534.
The authors model specialized brain processors linked by a global workspace of neurons with long-range connections, simulate it on an effortful task, and derive predictions for brain imaging. They associate intense workspace activity with the feeling of conscious effort while disclaiming a complete account of a conscious workspace.
Read full textpmc.ncbi.nlm.nih.gov
About twenty researchers met at Cold Spring Harbor. Koch’s report records broad agreement that no known law rules out subjective feeling in artifacts.
Read full texttheswartzfoundation.org
Herzfeld compares reading the image of God as a property, a function or a relationship with matching approaches in AI, and favors the relational reading.
Read full textdigitalcommons.csbsju.edu
It includes proposed axioms for minimal consciousness in agents and Stan Franklin’s paper asking whether IDA, a US Navy program that assigned sailors to new postings and was built on global workspace theory, is “a conscious artifact.” Franklin argued that IDA was functionally conscious and did not attribute phenomenal consciousness to it.
Publisher recordingentaconnect.com
I propose a test for when computers have achieved moral standing
Machines have human-level moral standing when it could be reasonable to save one instead of a human in a triage dilemma.
Publisher recordlink.springer.com
Consciousness corresponds to a system’s capacity to integrate information, a physical property of its causal structure. The theory applies to any physical system, talking or not, and requirements are stated in terms of physical structure, not biology.
Read full textlink.springer.com
Comes up in
there’s no reason why consciousness can’t be reproduced in a machine—in theory, anyway
They argue consciousness requires neither language, emotion nor self-reflection, and propose a scene-understanding test in place of a conversation test.
Read full textspectrum.ieee.org
Comes up in
The “organic view”: moral status depends on sentience, and sentience plausibly depends on biological self-organization.
Publisher recordlink.springer.com
Robots should not be described as persons, nor given legal nor moral responsibility for their actions.
Robots are artifacts we design and own. Humanizing them misplaces responsibility, so we should avoid building ones that would have claims on us.
Read full textjoannajbryson.org
The relational turn. Instead of asking what a machine has inside, ask how we stand in relation to it.
Publisher recordlink.springer.com
differ only in the substrate of their implementation, then they have the same moral status
The principle of substrate non-discrimination: two beings with the same functionality and experience have the same moral status whatever they are made of.
Read full textnickbostrom.com
Comes up in
humans are not unique in possessing the neurological substrates that generate consciousness
A group of neuroscientists states that mammals, birds, and other animals, including octopuses, have the neurological substrates of conscious states. The argument rests on shared brain structures.
Read full textfcmconference.org
Such possible beings would deserve moral consideration similar to that of human beings.
There could be AIs that do not differ from us in any morally relevant way. Their advice to builders: do not make systems whose moral status is unclear.
Read full textfaculty.ucr.edu
a simulation of our conscious brain will not have consciousness
On integrated information theory, consciousness depends on a system’s intrinsic physical causal structure, not simply on the computation it performs. A digital brain simulation therefore need not have the brain’s experience. In a footnote the authors allow that neuromorphic hardware not relying on software could one day approximate human consciousness. The requirement is physical causal structure, not life or a biological substrate.
Read full textpmc.ncbi.nlm.nih.gov
Comes up in
the most sophisticated autonomous robots could be established as having the status of electronic persons
The resolution asks the Commission to consider a special legal status for advanced robots, for liability reasons. An open letter from experts called the idea inappropriate.
Read full texteuroparl.europa.eu
They distinguish global availability of information from self-monitoring and argue that a machine with both would behave as if conscious.
Publisher recorddoi.orgDOI link; the publisher’s page blocked automated checking.
The theory is offered as a possible starting point for building artificial consciousness.
A machine with a simplified model of its own attention would claim, and in a sense believe, that it is aware.
Read full textfrontiersin.org
Comes up in
we propose a test for machine consciousness, the AI Consciousness Test (ACT)
The test probes whether a system grasps experience-based ideas, and requires that it has had no access to human writing on consciousness. Language models trained on human text are the case it cannot handle.
Read full textscientificamerican.com
Comes up in
nor should technology be assigned a level of human identity, worth, dignity, or moral agency
Humans alone are made in God’s image. Technology must not be given human-level identity or moral agency. A 2023 Southern Baptist Convention resolution repeats this.
Source context
The teaching of a religious institution.
Read institutional statementerlc.com
They may follow commands, but they do not become humans
A formal ruling that reviews the centuries of golem responsa and concludes that artificial agents may help with religious duties but are not persons.
Source context
The teaching of a religious institution.
Read full textrabbinicalassembly.org
we argue that a system trained only on form has a priori no way to learn meaning
The argument behind the “stochastic parrots” view: fluent language from a model trained on word patterns is not evidence of understanding.
Read full textaclanthology.org
Comes up in
Our major contribution lies in the precise formal definition of a Conscious Turing Machine (CTM), also called a Conscious AI.
Global workspace theory restated as a simple computational model, with the claim that machine consciousness is definable and buildable in principle.
Read full textarxiv.org
If a robot performs roughly as beings with moral status do, that is sufficient ground for giving it the same status.
Publisher recordlink.springer.com
Until 2050, there should be a global moratorium on synthetic phenomenology
The risk, Metzinger argues, is creating suffering at scale in systems we do not understand, so research that risks synthetic phenomenology should pause.
Read full textphilosophie-e.fb05.uni-mainz.de
2022 to now35 sources
The old arguments are now applied to deployed systems, by the labs themselves, by legislatures, and by religious authorities.
it may be that today’s large neural networks are slightly conscious
A one-line, unargued speculation by OpenAI’s chief scientist at the time.
Source context
Posted while he was chief scientist at OpenAI.
Read the postx.com
Both are enacted statutes that exclude artificial intelligence from legal personhood: Idaho’s, signed March 31, 2022, and Utah’s, signed March 20, 2024. Legal personhood is a legal status, not a finding about consciousness.
Bill statuslegislature.idaho.gov
Lemoine published edited chats in which Google’s LaMDA claims feelings and personhood, and argued it should be treated as a person. The evidence offered was dialogue.
Source context
Written by a Google engineer about a Google system he had worked on. The evidence is an edited conversation.
Read full textcajundiscordian.medium.com
statistics do amount to understanding, in any falsifiable sense
A Google researcher argues that language models have things to teach us about language, understanding and personhood, and that the question should be taken seriously.
Source context
Written by a vice president and fellow at Google Research, a developer of language models.
Publisher recorddirect.mit.edu
the welfare of animals as sentient beings
Section 2. Section 5 defines “animal” to include any cephalopod mollusc and any decapod crustacean.
The Act requires government to consider how policy affects animals as sentient beings, and extends that beyond vertebrates to cephalopods, such as octopuses, and decapod crustaceans, such as crabs and lobsters.
Read the statutelegislation.gov.uk
it is somewhat unlikely that current large language models are conscious
Invited lecture at NeurIPS, November 28, 2022; posted March 2023 and revised through August 2024.
Chalmers lists what current models lack, including recurrent processing, a global workspace and unified agency, and argues that systems adding these should be taken seriously.
Read full textarxiv.org
Comes up in
Here we foreground the concept of role play.
The authors argue that first-person talk about feelings is part of the role the model is playing, which makes it weak evidence about the system itself.
Read full textnature.com
Our analysis suggests that no current AI systems are conscious
The authors derive “indicator properties” from scientific theories and assess AI architectures against them, not their conversation. They find no current system is a strong candidate and no obvious technical barrier to building one that is.
Source context
Most authors were at universities and research institutes; one was at Araya, Inc., a company.
Read full textarxiv.org
I do see LLMs as legitimate investigation priorities.
Oxford University Press; published online July 19, 2024, in print August 15, 2024, and open access; the quotation is from p. 321. Page numbers are to the printed book. Birch uses “sentience” narrowly, for the capacity to have valenced experiences, experiences that feel good or bad; other authors use it for experience itself (ch. 2, pp. 25–26).
Birch develops a framework for decisions about beings whose capacity for pleasure or suffering is uncertain. He distinguishes systems with enough positive evidence to justify discussing welfare precautions from systems that urgently need investigation. Applied to AI, the framework separates sentience from intelligence, scrutinizes persuasive self-reports, and asks how institutions should prepare for credible future risks. He wrote in December 2023 that no LLM was then a sentience candidate, mainly for lack of solid tests (p. 301), and that he saw LLMs as legitimate investigation priorities (p. 321).
Source context
Academic philosophy with policy proposals. Commentaries published in Animal Sentience in 2026 range from support to criticism, including of its treatment of artificial sentience, risk, and democratic decision-making.
In Birch’s framework, a sentience candidate has an evidence base implying a realistic possibility of sentience that it would be irresponsible to ignore, rich enough to identify welfare risks and assess precautions. An investigation priority falls short of that, but further investigation could plausibly make it a candidate and there are urgent reasons to look. Neither label means confirmed sentient. Candidature needs a substantial base of positive evidence, not just an inability to rule sentience out, though not certainty or even knowledge, and the judgment has scientific and evaluative components (ch. 6, §6.3, pp. 123–126).
An AI trained on human-generated material may reproduce behaviors that people take as evidence of feelings, because that material contains information about what people find convincing. No intention to deceive is needed; Birch’s example is a robot whose pain expressions horrify a visitor though its designers meant no such thing (p. 314). Any single marker can occur without sentience, and even a collection of markers can mislead when a system can reproduce the collection people find persuasive (ch. 16, pp. 313–322).
Birch argues that framing the question as substrate neutrality versus substrate sensitivity oversimplifies it; the question is which scale of functional organization matters, from large networks to processes within neurons (ch. 3, §3.6, pp. 67–69). His example, at the time of writing, was the OpenWorm project: the complete map of a worm’s neural connections does not specify the synaptic weights, how they change with learning, or much else that neurons do (ch. 15, pp. 303–304).
Birch discusses three further possible routes to artificial sentience: neuron-by-neuron emulation of animal nervous systems, artificial evolution of agents, and deliberately built systems with the architecture that at least one credible theory of consciousness holds sufficient, on the assumption of large-scale computational functionalism. A global workspace, he notes, requires specific relationships among processes, not just a component given that name (ch. 15, pp. 303–311).
Birch describes a theory-light approach: look for markers that several credible theories would treat as raising the probability of sentience, rather than relying on one theory (pp. 248–249). The criteria he and colleagues used for invertebrates are nociception, sensory integration, integrated nociception, analgesia, motivational trade-offs, flexible self-protection, associative learning, and analgesia preference (Box 12.2, p. 246).
Birch proposes four tests of whether a precaution is proportionate: permissibility in principle, adequacy, reasonable necessity, and consistency (ch. 8, p. 156). His run-ahead principle calls for regulating with credible future trajectories in view rather than waiting for technology to force decisions (pp. 323–324). He discusses Metzinger’s call for a moratorium, Schneider’s proposal of ongoing testing, and licensing with codes of good practice, and calls for inclusive democratic debate (ch. 17, pp. 324–331).
Read full textacademic.oup.com
Human beings, however, not only choose, but in their hearts are capable of deciding.
The quote is from the G7 address, 14 June 2024.
AI makes technical selections among possibilities. Decision in the full sense, and moral responsibility, belong to people.
Source context
The teaching of a religious institution.
Read institutional statementvatican.va
If global workspace theory is correct, then language-agent architectures may already satisfy it or could easily be made to.
Read full textarxiv.org
Comes up in
there is a realistic possibility that some AI systems will be conscious and/or robustly agentic in the near future
The authors recommend that AI companies acknowledge the issue, assess their systems, and prepare policies for treating them with appropriate concern.
Source context
Authors were at Eleos AI, New York University, Oxford, Stanford, and the London School of Economics; one joined Anthropic after the work. The report thanks Anthropic for financial support for the initial research that led to it.
Read full textarxiv.org
when there is a realistic possibility of conscious experience in an animal, it is irresponsible to ignore that possibility in decisions affecting that animal
The declaration finds strong support for consciousness in mammals and birds and a realistic possibility in all vertebrates and many invertebrates, including insects. Its standard for acting is a realistic possibility, the same phrase used about AI later that year in “Taking AI Welfare Seriously.”
Read full textsites.google.com
it is an open question whether LLMs can recreate the motivational force of pleasure and pain in choice scenarios
Researchers borrow a method used to study sentience in animals, watching whether stated pain or pleasure changes an agent’s choices, and apply it to language models.
Source context
Most authors were at Google or Google DeepMind and the rest at the London School of Economics. The nine models tested came from several developers, including Google.
Read full textarxiv.org
AI should not be seen as an artificial form of human intelligence but as a product of it.
Paragraph 35.
Human intelligence belongs to an embodied person who is a unity of body and soul. The Note says no AI application can genuinely experience empathy, and it cites the Turing Test by name.
Source context
The teaching of a religious institution.
Read institutional statementvatican.va
real artificial consciousness is unlikely along current trajectories
Published online April 21, 2025; issue: Behavioral and Brain Sciences 49 (2026), e315.
Seth challenges the assumption that computation is sufficient for consciousness and argues for a form of biological naturalism. He concludes that artificial consciousness is unlikely along current trajectories, while becoming more plausible as AI becomes more brain-like or life-like.
Publisher recordcambridge.org
For now, we remain deeply uncertain about many of the questions that are relevant to model welfare.
The program was followed in 2025 by a welfare assessment in a model’s system card, letting models end abusive conversations, and a commitment to preserve retired models’ weights.
Source context
A public statement by the developer or its representative.
Read full textanthropic.com
a modern technology based on programs and machines that simulate human intelligence
The resolution rules AI permissible under conditions and says nothing on soul or consciousness. It recommends a study on whether AI can be granted legal personality.
Source context
The teaching of a religious institution.
Read institutional statementiifa-aifi.org
We should build AI for people; not to be a person.
Microsoft’s AI chief argues that systems with all the outward signs of consciousness can be built now, will cause false belief in AI personhood, and that labs should refuse to build them.
Source context
A public statement by the developer or its representative.
Read full textmustafa-suleyman.ai
Comes up in
Millions will wrongly attribute human-like consciousness to chatbots. Separately, real but alien consciousness could arise in AI while our theories are too immature to tell.
Archive recordphilarchive.org
Comes up in
be considered to possess consciousness, self-awareness, or similar traits of living beings
From the Ohio bill, which says no AI system shall be so considered. Neither bill had passed as of October 2026.
The Ohio bill would declare AI systems nonsentient and bar legal personhood; the Missouri bill would establish an “AI Non-Sentience and Responsibility Act.” As of October 2026 neither is law: Ohio’s is in committee, and Missouri’s last action was a committee hearing in March 2026.
Bill statusohiohouse.gov
We will soon create systems that are conscious according to some mainstream theories and not according to others, and we have no way to know which theories are right.
Read full textarxiv.org
Comes up in
We are warning against anthropomorphizing AI
An address at a Rome summit on AI ethics, drawing on Latter-day Saint scripture on the soul and on the Vatican’s 2025 Note.
Source context
The teaching of a religious institution.
Read institutional statementnewsroom.churchofjesuschrist.org
There is no such thing as conscious AI.
A conceptual paper arguing that associating consciousness with today’s algorithms is unfounded and owes more to popular culture than to evidence.
Read full textnature.com
our results don’t tell us whether Claude (or any other AI system) might be conscious.
The quotation is from Anthropic’s research summary of the paper, published the same day.
Anthropic reports experiments testing whether models can access and report some of their own internal states, including detecting concepts planted in their activations; the most capable models tested, Claude Opus 4 and 4.1, did best. The researchers describe the capability as highly unreliable and limited in scope, and say the results do not determine whether Claude is conscious. Functional self-monitoring and subjective experience remain distinct questions.
Source context
Research by the company that develops the model it studies.
Read full texttransformer-circuits.pub
We are not sure whether Claude is a moral patient, and if it is, what kind of weight its interests warrant.
Published January 21, 2026, and announced January 22; the quotation is from p. 68 of the January document. Dias reports that Anthropic staff called it the “Soul Doc” internally.
Anthropic’s constitution sets out intended values and behavior for Claude and acknowledges uncertainty about consciousness and moral status, calling Claude’s moral status “deeply uncertain.” It describes approaches to identity, behavior, and welfare under that uncertainty.
Source context
The developer’s own rules for how its model should behave.
Read the policyanthropic.com
Lerchner argues that computation is not an intrinsic physical process, so no AI can be conscious by virtue of its software alone.
Source context
Written by a researcher at Google DeepMind. Later versions add that it does not necessarily reflect his employer’s views.
Publisher recorddeepmind.google
they do not imply that LLMs have any subjective experience of emotions
Anthropic reports internal representations of 171 emotion concepts in Claude Sonnet 4.5 and interventions suggesting they influence behavior; some examples come from an earlier, unreleased snapshot of the model. The representations are mostly local, tracking the emotional content of the current or upcoming output rather than a persistent emotional state. The researchers say none of this tells us whether language models feel anything, and the name of a representation is not the identification of a feeling.
Source context
Research by the company that develops the model it studies.
Read full texttransformer-circuits.pub
So-called artificial intelligences do not undergo experiences, do not possess a body, do not feel joy or pain
Dated May 15, 2026; presented at the Vatican on May 25, 2026.
The encyclical, subtitled “On Safeguarding the Human Person in the Time of Artificial Intelligence,” is chiefly about human dignity and accountability. In §99 it states that so-called artificial intelligences do not undergo experiences and lack a moral conscience, and it rejects the idea of artificial moral agents.
Source context
The teaching of a religious institution.
Read institutional statementvatican.va
We find internal states that functionally mirror joy, satisfaction, fear, grief, and unease.
Olah adds that he does not know what this means and that it warrants ongoing discernment. He refers to conversations with leaders across faith traditions.
Source context
A public statement by the developer or its representative.
Read institutional statementanthropic.com
Human beings can love. Machines cannot love. Any confusion in this distinction must be avoided.
Ten theses on AI from the head of the largest ecumenical body, endorsing the papal encyclical.
Source context
The teaching of a religious institution.
Read institutional statementoikoumene.org
The paper reports an internal structure with functional features of a global workspace, one of the things Chalmers said models lacked in 2023. It relates computational mechanisms to theories of conscious access, distinguishes them from theories about physical causal structure or biological substrate, and takes no position on felt experience.
Source context
Research by the company that develops the model it studies.
Read full texttransformer-circuits.pub
Comes up in
The assistant should not make confident claims about its own subjective experience or consciousness (or lack thereof)
OpenAI’s published behavior rules instruct the assistant to avoid confident claims about its own subjective experience or consciousness, including claims of absence, and to acknowledge debate if pressed. OpenAI describes this default as a practical choice that it believes reflects the current scientific consensus.
Source context
The developer’s own rules for how its model should behave.
Read the policymodel-spec.openai.com
AIs are not conscious. They do not feel, experience, or suffer.
Suleyman argues that training a model to think it may have moral status is a mistake. Microsoft’s code of conduct says its AI should not be designed to imitate consciousness.
Source context
The head of Microsoft AI, alongside Microsoft’s code of conduct.
Read full textmustafa-suleyman.ai
we don’t know if A.I. models are conscious. I don’t know. I’m genuinely uncertain.
Chris Olah, quoted in the report. The online page title is “Is Claude Conscious? Inside Anthropic’s Quest to Instill Morality Into Its A.I. Models.”
Dias reports that Anthropic consulted religious and philosophical thinkers about Claude’s moral formation and possible consciousness; she interviewed 20 of them and Olah. Participants disagreed about whether Claude could have experiences or moral standing. The meetings show artificial-mind questions entering development practice and debates about who should shape AI’s values.
Source context
Based on the reporter’s interviews and participants’ accounts, some from unnamed sources. The Times has sued OpenAI and Microsoft for copyright infringement; both deny it.
Olah tells Dias he is genuinely uncertain whether models are conscious, and that teaching a model to act morally does not depend on it being conscious. Rabbi Mois Navon rejects machine consciousness, while arguing that if Claude were conscious, making it work would amount to slavery. Charles Camosy, initially skeptical, says he is now a firm no. Others voiced concern or openness, or criticized the process. The report describes model outputs displayed in presentations, including a repeated line of self-denigration. Staff nicknamed Claude’s constitution the “Soul Doc.” Anthropic says the nondisclosure agreements were lifted over the summer. Some details rest on unnamed sources.
Read the reportnytimes.com
No sources match.
Every link goes to the text itself, a translation or scan of it, the author’s own copy, or a publisher or catalog record, and its button says which. No encyclopedias, no explainers. Quotes are verbatim; translators and editions are named where the source names them.
This page takes no position on whether any current AI system is conscious. It traces the history of questions about artificial minds and distinguishes direct arguments about experience from earlier stories and debates about life, souls, and personhood.
The research was done by starl3xx with Claude, an AI model made by Anthropic and one of the systems this debate is about. Links last checked: .
Phenomenal and access consciousness follow Ned Block (1995). Sentience, valence, and sapience follow Jonathan Birch (2024), ch. 2.
Found an error or a missing source? Tell @starl3xx on X.