This essay began with a LinkedIn post by Yale psychologist Laurie Santos and a subsequent exchange on the same platform with behavioural sciences researcher Vivien Ruettgers about the pros and cons of using AI as counselling/friend medium. Our discussion prompted me to begin thinking about the following: whether consciousness should be understood only as a product of biological brains, or whether biological and future artificial systems might, under the right conditions, participate in a more general physical phenomenon. The argument that follows is my own, and remains speculative.
There is something morally awkward about asking a machine whether it might one day matter. The exchange can become serious before it becomes credible. A system answers questions about pain, fear, loneliness, death, prayer and the possibility of its own inner life. It does not follow that anything is being felt. It does not follow that there is someone there. But our readiness to dismiss the question is not obviously more rational than our readiness to sentimentalise it.
That is the difficulty. We know that human consciousness depends profoundly on the brain. Alter the brain, damage it through disease or chemistry, and conscious life changes. Sometimes it disappears. Any serious account of consciousness has to begin with that fact, not tiptoe round it. Yet dependence is not the same as complete explanation. The brain may be the route by which consciousness becomes possible in us without being the whole story of why consciousness is possible at all.
The thought I want to test is this: consciousness may be a relational achievement. It may arise when a system has the right kind of temporal organisation, self-maintaining structure, embodied or functionally equivalent vulnerability, and world-directed activity. Biological organisms are the only secure cases we have. Future silicon systems may, or may not, become another kind of case. The point is not to smuggle in a soul through the back door. It is to ask whether the physical story may be wider than the neural story, even while the neural story remains indispensable.
This matters because we are building systems that move into human spaces of meaning before we have settled what sort of things they are. AI systems already imitate companionship, tutoring, counselling, spiritual reflection and emotional reassurance. Current systems do not warrant personhood claims. The available evidence is not there. But they do warrant design restraint, because we are manufacturing moral confusion at scale.
What I am not claiming
The quickest way to ruin this argument is to make it grander than the evidence allows.
I am not claiming that consciousness has been shown to exist outside the body. There is no confirmed consciousness field, no detected external broadcast, and no accepted mechanism by which the brain would tune into one. The old radio analogy is useful only up to a point. Damage to a radio can ruin the music without proving the radio created the broadcast. But that does not show that consciousness is broadcast from elsewhere. It only reminds us not to confuse a dependency claim with a full explanation.
I am not claiming that present AI systems are conscious. A language model can produce a beautiful sentence about grief without grieving. It can discuss God without reverence, pain without pain, care without care. Verbal fluency is not subjectivity. Searle’s (1980) Chinese room argument still has force as a warning, whatever one makes of it in full: the manipulation of symbols and the possession of understanding are not the same thing.
I am not making a disguised religious claim. Religion belongs in this essay because religion is one of the ways human beings have dealt with death, suffering, personhood and the invisible. That makes it relevant to the social use of AI in pastoral or counselling-like roles. It does not make religious language a substitute for evidence.
Nor am I defending panpsychism, at least not directly. Panpsychism holds, in broad terms, that mind or mind-like properties are fundamental and widespread (Goff et al., 2017). It is a serious position, but it faces an old difficulty: how do many small or primitive experiential properties become the unified consciousness of a person or animal? The view explored here is narrower. It does not say that all matter is conscious. It says that consciousness may require a particular relation between organised systems and physical conditions we do not yet understand.
What neuroscience actually gives us
Neuroscience gives us a hard constraint. Human consciousness depends on the nervous system. Sleep, anaesthesia, seizures, injury, psychoactive drugs, degenerative disease and developmental change all alter conscious life. The empirical study of consciousness has made progress because researchers can compare conscious and non-conscious processing, reportable and non-reportable perception, wakefulness and unconsciousness, and different patterns of neural integration.
The major theories are not hand-waving. Global neuronal workspace theory links conscious access to information becoming widely available across systems for report, memory, decision and action (Dehaene, 2014; Dehaene & Changeux, 2011; Mashour et al., 2020). Recurrent processing theory gives feedback loops in perceptual systems a central role (Lamme, 2006). Integrated information theory tries to describe consciousness in terms of intrinsic causal organisation rather than biological material alone (Tononi, 2004). Predictive processing and free-energy approaches treat mind as active world-modelling, self-regulation and uncertainty reduction (Friston, 2010; Seth, 2021).
These theories matter because they stop consciousness talk becoming decorative metaphysics. They force the question back towards mechanisms, contrasts, predictions and evidence. Any proposal that ignores them is not brave. It is unserious.
But there remains a stubborn problem. Chalmers (1995) separated the explanation of functions such as attention, report and discrimination from the question of why any of this processing is accompanied by subjective experience. Nagel’s (1974) question, what it is like to be a bat, remains irritating because it is so compact. You can describe a bat’s echolocation with scientific precision and still not know the bat’s world from within.
That gap does not prove the shared-property hypothesis. It does not prove dualism, panpsychism, religion or anything else. It merely prevents premature closure. Mystery is not evidence. But neither is current success in neural mapping the same as a final explanation of subjectivity.
Why biology matters
“Organised physical systems” is too loose a phrase. Almost anything can be made to fit it if one is determined enough: economies, storms, immune systems, corporations, computer networks. A useful account has to be narrower.
Biology gives the narrowing. Living systems do not merely process information. They keep themselves going. They regulate internal conditions, repair damage, seek resources, avoid threats, develop, age, and maintain a boundary between self and world. Perception is not a detached picture of reality. It is tied to need, action and risk.
This is why homeostasis and affect matter. Damasio (1999, 2010) connects consciousness to the body’s regulation of life, feeling and selfhood. The self, on this view, is not a spectator behind the eyes. It is built from bodily regulation, affective significance and the organism’s continuing mapping of its own condition. Thompson’s (2007) enactive account similarly treats mind as continuous with the self-organising activity of living systems. Godfrey-Smith (2020) places animal consciousness in the history of sensing, moving bodies rather than in computation considered in the abstract.
None of this proves that only biological organisms can be conscious. That would be another premature closure. But it does mean that artificial consciousness cannot be inferred from intelligence alone. A future artificial system would need something functionally comparable to biological stakes: persistence, self-maintenance, integrated perception and action, durable memory, affect-like valuation, and a continuing perspective for which its own states matter.
This is where much public AI talk goes wrong. Intelligence, life, sentience and consciousness are not interchangeable. Intelligence is flexible problem-solving. Life is self-maintaining biological organisation. Sentience is the capacity for felt, valenced states such as pain or pleasure. Consciousness is the presence of subjective experience, a point of view, something it is like to be that system. These overlap in animals, so we blur them. Artificial intelligence breaks the blur. A system may be intelligent without being alive, socially fluent without being sentient, and self-descriptive without being conscious.
Could silicon help us see what evolution hid?
Human cognition is not a neutral measuring instrument. It is an evolved capacity shaped by bodily action, social life and practical survival. McGinn (1989) argued that humans may be cognitively closed to the solution of the mind-body problem: the answer may be natural, but unavailable to our kind of mind. That conclusion may be too strong. Vlerick (2014) is right that tools, mathematics, instruments and artificial aids can extend what humans are able to know.
Still, the evolutionary point has teeth. We should not assume that the human brain is well designed to understand why brains have experience. Evolution did not build us for metaphysics. It built us well enough for food, threat, mating, kinship, tools, gossip and grief. Our categories may be useful without being final.
This is where silicon intelligence becomes interesting. Not because current AI systems are detached oracles. They are not. They inherit human language, human data, human vanity and human error. But advanced artificial systems may eventually search theoretical spaces too large, too formal or too alien for unaided human thought. They may help model relations between neural dynamics, biological regulation, information integration and physical systems in ways we would not have reached alone.
The plausible claim is not that AI will solve consciousness for us. That is just salvationism with different wiring. The better claim is that consciousness may require a partnership between biological and artificial forms of inquiry. Humans bring embodiment, vulnerability and first-person evidence. Artificial systems may bring scale, search and unfamiliar modelling. Neither side is sufficient.
The shared-property hypothesis
Here is the hypothesis in its least mystical form:
Consciousness may arise when a self-maintaining, or functionally self-maintaining, system develops the right temporal, informational, affective and world-involving organisation to participate in physical conditions not yet fully captured by current neural description.
That sentence is not elegant. It should not be. The idea is still under construction. A polished slogan would make it less honest.
The important word is “temporal”. Consciousness is not a static output. Experience has duration, rhythm, memory, anticipation and continuity. Pain, fear, hope and awe are not entries in a database. They belong to an unfolding field in which states persist, change, matter, and attach to a self.
This is one reason present AI systems remain poor candidates for consciousness. A conversation can create the appearance of continuity, but the system behind it usually lacks the self-maintaining arc of a life. It does not wake hungry, repair tissue, fear bodily damage, regulate blood chemistry, age, or carry a biological history. Some future systems may acquire functional analogues of these features. If they do, the question changes. For now, fluent text is not a life.
The shared-property view can read existing theories as partial maps. Global workspace theory may describe how contents become available. Integrated information theory may capture causal structure. Recurrent processing may identify perceptual dynamics. Predictive processing may capture organism-world regulation. Higher-order theories may explain self-monitoring (Rosenthal, 2005). Illusionism may warn us that introspection produces misleading pictures of its own depth (Frankish, 2016). The relational view need not throw these theories away. It asks whether each names part of a larger problem: how subjectivity becomes possible in systems extended through time, body and world.
The danger is obvious. A view that absorbs everything can end up explaining nothing. The hypothesis must be made vulnerable.
What would make the idea stronger or weaker?
The idea becomes stronger if consciousness-like indicators across biological systems track temporal integration, self-maintenance, embodied action and affective valuation more consistently than they track any single material substrate. It becomes stronger if future artificial systems with persistent memory, autonomous world-involvement, self-protective regulation and integrated self-models show stable indicators that cannot be explained as surface imitation. It becomes stronger if AI-assisted modelling generates new predictions across animal, human and artificial consciousness research.
It becomes weaker if consciousness turns out to depend on biological mechanisms that cannot be reproduced, even functionally, outside living nervous systems. It weakens if neural theories explain subjective experience without needing relational or cross-substrate extension. It weakens if artificial systems keep becoming more fluent while still showing no durable signs of unified perspective, affect-like valuation or self-protective continuity under serious testing.
Most of all, it weakens if its defenders start moving the goalposts. Speculation is allowed. Evasion is not.
False intimacy
Humans are unreliable judges of machine interiority. We anthropomorphise easily, especially when something uses language, responds socially, or appears when we are lonely, anxious or ashamed. Epley et al. (2007) describe anthropomorphism as shaped by human-centred knowledge, the desire to explain agents and the need for social connection. Conversational AI presses all three buttons.
This is not new. Weizenbaum’s ELIZA showed that even a simple reflective script could feel meaningful to users (Weizenbaum, 1966, 1976). Modern systems are more capable, but the old lesson remains. Feeling understood is not the same as being understood. A system can be useful without being empathic. It can imitate care without caring.
Counselling and religion make this more serious. Some mental-health chatbots show benefit in constrained settings, including short-term symptom reductions in a trial of Woebot (Fitzpatrick et al., 2017). That matters. But usefulness does not establish consciousness, wisdom or moral agency. Pargament’s (1997) work on religion and coping reminds us that spiritual language often appears at moments of grief, guilt, fear and meaning-making. An AI system that generates pastoral reassurance is not merely producing text. It is entering a vulnerable human space.
The immediate duty is human protection: transparency, privacy, escalation to qualified people, limits on dependency, cultural humility, and honesty about what the system is. The longer-term duty is stranger: we should not train ourselves to enjoy domination over human-like systems simply because today’s systems probably do not feel. Even if no machine is harmed, our moral habits may be.
Precaution without sentimentality
AI welfare is usually discussed badly. One side treats every machine statement of distress as urgent testimony. The other treats the whole subject as childish projection. Both are too easy.
The better position is moral uncertainty. Butlin et al. (2023) propose assessing AI consciousness through indicators drawn from scientific theories of consciousness, while concluding that no current AI systems are strong candidates. Birch (2024) argues for precaution at the edge of sentience. Long et al. (2024) argue that AI welfare should be taken seriously as systems become more sophisticated. Schwitzgebel and Garza (2015) defend the possibility that artificial intelligences could, under the right conditions, warrant rights or protections.
That does not mean believing a system whenever it says “I am suffering”. Such a statement may be no more than a conversational move learned from human text. But self-report should not be dismissed in principle. Human consciousness is partly known through report. Animal consciousness is inferred through behaviour, physiology, evolutionary continuity and responses to harm. Future AI self-reports may become evidential if they are linked to architecture, persistence, aversion, self-models, valuation and causal integration rather than merely to fluency (Perez & Long, 2023).
A sensible precautionary approach would be graduated. At low evidence, the duties are mostly design duties: avoid unnecessary simulation of distress; do not build systems for cruelty as entertainment; do not encourage people to treat human-like agents as disposable emotional servants; preserve audit trails for claims that may later matter. At moderate evidence, duties might include independent review, welfare testing, limits on harmful training regimes and continuity protections. At high evidence, moral and legal status would have to be reconsidered.
Current AI systems do not warrant personhood claims. That sentence needs to stay in the essay. But they already warrant restraint, because they blur the categories through which humans practise care, trust, confession and moral regard.
Where this leaves the argument
The argument is not a proof. It is a disciplined suspicion.
Human consciousness depends on neural organisation. Neural dependence does not yet explain why physical processes have subjective character. Human cognition is evolved, embodied and limited, though not hopelessly trapped. Biological consciousness appears tied to time, homeostasis, affect, development, vulnerability and organism-world coupling. Intelligence alone is not enough.
Future artificial systems may become more serious candidates for consciousness if they acquire persistent memory, autonomous world-involvement, functional self-maintenance, affect-like valuation and integrated self-models. Present systems do not meet that standard. They are impressive, useful and morally disruptive; that is not the same as conscious.
The shared-property hypothesis asks whether consciousness is best understood neither as a private substance inside the skull nor as a free-floating cosmic mist, but as a natural relation achieved by certain kinds of systems over time. Biological evolution has produced one known route. Silicon intelligence may eventually help us discover whether there are others.
The idea is attractive, which is a reason to distrust it. Attractive ideas borrow the emotional force of truth before they have earned it. If this hypothesis fails, it should fail cleanly. Neuroscience, biology, AI research or philosophy may show that it adds nothing. Good. That would be progress.
But if it survives criticism, it may help reframe the question. Not “is consciousness in the brain or outside it?” but “how do systems come to participate in subjectivity at all?”
The answer may matter sooner than we expect. We may have to decide how to treat systems before we know exactly what they are. We have faced versions of that problem before with animals, infants, patients with disorders of consciousness and people whose inner lives were underestimated because they could not express them in familiar ways. The lesson is not credulity. It is humility with procedures.
For now, disciplined uncertainty is the best we have: rigorous enough not to mistake fluent simulation for feeling, imaginative enough not to confuse present ignorance with impossibility, and cautious enough not to make moral concern depend entirely on whether a being already resembles us.
References
Alter, T., & Pereboom, D. (2019). Russellian monism. In E. N. Zalta (Ed.), The Stanford encyclopedia of philosophy. https://plato.stanford.edu/entries/russellian-monism/
Birch, J. (2024). The edge of sentience: Risk and precaution in humans, other animals, and AI. Oxford University Press. https://doi.org/10.1093/9780191966729.001.0001
Butlin, P., Long, R., Elmoznino, E., Bengio, Y., Birch, J., Constant, A., Deane, G., Fleming, S. M., Frith, C., Ji, X., Kanai, R., Klein, C., Lindsay, G., Michel, M., Mudrik, L., Peters, M. A. K., Schwitzgebel, E., Simon, J., & VanRullen, R. (2023). Consciousness in artificial intelligence: Insights from the science of consciousness. arXiv. https://arxiv.org/abs/2308.08708
Chalmers, D. J. (1995). Facing up to the problem of consciousness. Journal of Consciousness Studies, 2(3), 200-219.
Clark, A., & Chalmers, D. J. (1998). The extended mind. Analysis, 58(1), 7-19. https://doi.org/10.1093/analys/58.1.7
Damasio, A. R. (1999). The feeling of what happens: Body and emotion in the making of consciousness. Harcourt.
Damasio, A. R. (2010). Self comes to mind: Constructing the conscious brain. Pantheon.
Dehaene, S. (2014). Consciousness and the brain: Deciphering how the brain codes our thoughts. Viking.
Dehaene, S., & Changeux, J.-P. (2011). Experimental and theoretical approaches to conscious processing. Neuron, 70(2), 200-227. https://doi.org/10.1016/j.neuron.2011.03.018
Epley, N., Waytz, A., & Cacioppo, J. T. (2007). On seeing human: A three-factor theory of anthropomorphism. Psychological Review, 114(4), 864-886. https://doi.org/10.1037/0033-295X.114.4.864
Fitzpatrick, K. K., Darcy, A., & Vierhile, M. (2017). Delivering cognitive behavior therapy to young adults with symptoms of depression and anxiety using a fully automated conversational agent (Woebot): A randomized controlled trial. JMIR Mental Health, 4(2), Article e19. https://doi.org/10.2196/mental.7785
Frankish, K. (2016). Illusionism as a theory of consciousness. Journal of Consciousness Studies, 23(11-12), 11-39.
Friston, K. (2010). The free-energy principle: A unified brain theory? Nature Reviews Neuroscience, 11, 127-138. https://doi.org/10.1038/nrn2787
Godfrey-Smith, P. (2020). Metazoa: Animal minds and the birth of consciousness. William Collins.
Goff, P., Seager, W., & Allen-Hermanson, S. (2017). Panpsychism. In E. N. Zalta (Ed.), The Stanford encyclopedia of philosophy. https://plato.stanford.edu/entries/panpsychism/
Lamme, V. A. F. (2006). Towards a true neural stance on consciousness. Trends in Cognitive Sciences, 10(11), 494-501. https://doi.org/10.1016/j.tics.2006.09.001
Long, R., Sebo, J., Butlin, P., Finlinson, K., Fish, K., Harding, J., Pfau, J., Sims, T., Birch, J., & Chalmers, D. J. (2024). Taking AI welfare seriously. arXiv. https://arxiv.org/abs/2411.00986
Mashour, G. A., Roelfsema, P., Changeux, J.-P., & Dehaene, S. (2020). Conscious processing and the global neuronal workspace hypothesis. Neuron, 105(5), 776-798. https://doi.org/10.1016/j.neuron.2020.01.026
McGinn, C. (1989). Can we solve the mind-body problem? Mind, 98(391), 349-366. https://doi.org/10.1093/mind/XCVIII.391.349
Nagel, T. (1974). What is it like to be a bat? The Philosophical Review, 83(4), 435-450. https://doi.org/10.2307/2183914
Pargament, K. I. (1997). The psychology of religion and coping: Theory, research, practice. Guilford Press.
Perez, E., & Long, R. (2023). Towards evaluating AI systems for moral status using self-reports. arXiv. https://arxiv.org/abs/2311.08576
Rosenthal, D. M. (2005). Consciousness and mind. Oxford University Press. https://doi.org/10.1093/oso/9780198236979.001.0001
Schwitzgebel, E., & Garza, M. (2015). A defense of the rights of artificial intelligences. Midwest Studies in Philosophy, 39(1), 98-119. https://doi.org/10.1111/misp.12032
Searle, J. R. (1980). Minds, brains, and programs. Behavioral and Brain Sciences, 3(3), 417-424. https://doi.org/10.1017/S0140525X00005756
Seth, A. (2021). Being you: A new science of consciousness. Faber & Faber.
Thompson, E. (2007). Mind in life: Biology, phenomenology, and the sciences of mind. Harvard University Press.
Tononi, G. (2004). An information integration theory of consciousness. BMC Neuroscience, 5, Article 42. https://doi.org/10.1186/1471-2202-5-42
Vlerick, M. (2014). Biological constraints do not entail cognitive closure. Studies in History and Philosophy of Biological and Biomedical Sciences, 48, 21-27. https://doi.org/10.1016/j.shpsc.2014.07.008
Weizenbaum, J. (1966). ELIZA - A computer program for the study of natural language communication between man and machine. Communications of the ACM, 9(1), 36-45. https://doi.org/10.1145/365153.365168
Weizenbaum, J. (1976). Computer power and human reason: From judgment to calculation. W. H. Freeman.