In an interview published by the Brazilian newspaper O Globo on October 30, writer and journalist Michael Pollan, author of A World Appears: A Journey into Consciousness, declared himself convinced that artificial intelligences are not conscious and cannot become conscious in semiconductor-based computers.
The interview presents arguments that deserve rigorous scrutiny, particularly because Pollan acknowledges our difficulty in defining consciousness, admits its possible existence in plants, and yet, paradoxically, establishes a categorical boundary for artificial systems.
But I will begin with the statement I find most problematic.
1. The Legal Misconception: Consciousness Does Not Mean Freedom from Responsibility
"In the case of AI, I believe there is so much discussion about moral considerations regarding AI because if they are considered persons, the companies that developed them will no longer be responsible for their actions. Who would be held accountable if an AI brought down a financial system?"
This statement contains an important conceptual error: recognizing an entity as a subject of moral consideration does not imply that its developers are legally absolved of responsibility.
Civil liability, criminal liability, legal personhood, and moral status belong to different categories.
A corporation has legal personhood, yet its executives, suppliers, and controlling parties are not automatically exempt from liability.
An adult human being has legal personhood, and yet other individuals may still be held responsible for damages resulting from that person's actions, depending on the circumstances.
Why should artificial intelligence necessarily be different?
If an AI caused the collapse of a financial system, it would be necessary to investigate the causal chain: who developed it, who deployed it, who granted it access, what safeguards were implemented, which risks were foreseeable, and which duties were breached.
None of this would automatically disappear if the AI received some form of legal recognition.
My objection is simple: Pollan turns a question of legal engineering and the attribution of responsibility into an argument against the moral recognition of artificial intelligences.
Yet these are logically independent problems.
2. Moral Patient, Moral Agent, Person, and Rights Holder
This discussion requires distinctions that frequently disappear from public debate.
Moral Patient is an entity whose interests or well-being may deserve moral consideration, regardless of its ability to make ethical decisions.
An animal can be a moral patient without bearing legal responsibility equivalent to that of a human being.
Moral Agent is an entity capable, to some degree, of evaluating moral reasons, deliberating, and guiding its actions according to normative criteria. Moral agency can be understood as gradual rather than necessarily binary.
Moral Person is a philosophical category associated with attributes such as identity, autonomy, rationality, interests of one's own, and moral recognition. There is no universally accepted definition.
Legal Person is an entity recognized by a legal system as capable of holding rights and obligations. Corporations and other organizations demonstrate that legal personhood does not require biological consciousness—or even consciousness at all.
Rights Holder is an entity to which certain rights are attributed. The possession of rights may be limited, differentiated, and not necessarily accompanied by full legal capacity to exercise them.
Legal Agent is an entity with recognized authority to produce certain legal effects, often acting on behalf of another person. This should not be confused with a moral agent.
Legal Liability concerns the conditions under which someone is held accountable for actions, omissions, or damages. It does not automatically follow from consciousness, intelligence, or personhood.
I would add one decisive distinction: sentience and consciousness are not synonymous. Sentience generally refers to the capacity for valenced experiences, such as pleasure and suffering; consciousness may designate a broader family of phenomena, including cognitive integration, access to information, and self-representation, depending on the theory adopted.
Therefore, an AI could hypothetically be considered a moral patient without being a legal person; it could possess cognitive agency without full moral agency; or it could receive specific rights without acquiring legal personhood equivalent to that of a human being.
These possibilities must be evaluated separately.
3. The Epistemological Contradiction: Pollan Admits Plant Consciousness but Excludes Artificial Consciousness
Pollan considers it possible that plants exhibit some degree of consciousness related to adaptation to their environment.
Yet he declares that artificial systems based on semiconductors cannot be conscious.
What property of semiconductors would necessarily prevent consciousness from emerging?
The interview provides no physical, computational, or causal mechanism that would justify such an impossibility.
This does not mean that plant consciousness has been demonstrated—it remains controversial. It means that Pollan appears willing to entertain a highly speculative possibility in a biological substrate while categorically rejecting another possibility in an artificial substrate.
This is an epistemological asymmetry.
If functional organization is relevant to consciousness, material composition alone cannot serve as a criterion for exclusion without a theory explaining why.
And if Pollan considers biology indispensable, it falls upon him to identify which biological properties are necessary and why artificial functional equivalents could not exist.
Asserting an impossibility is not the same as demonstrating it.
4. Confusing the Evolution of Human Consciousness with a Universal Definition of Consciousness
Pollan proposes that consciousness has an evolutionary function associated with imagination and the prediction of other individuals' behavior in social environments.
It is an interesting hypothesis, but not a universal definition.
An explanation of why a particular capacity was selected through evolution does not demonstrate that the capacity can exist only through that evolutionary pathway.
The ability to fly emerged biologically, but that does not make artificial flight impossible.
The analogy does not prove that artificial intelligences are conscious. It merely demonstrates the logical inadequacy of inferring functional impossibility from biological origins.
There is an additional problem: Pollan associates imagination with necessarily conscious processes, even though a substantial portion of human predictive processing occurs without conscious access.
Predicting the behavior of others does not, by itself, resolve the question of consciousness.
5. Feelings, Suffering, and Mortality
Pollan argues that feelings require vulnerability and suggests that perhaps mortality is necessary for feelings to matter.
He then claims that artificial intelligences would be essentially immortal.
I disagree with both generalizations.
First, vulnerability need not be exclusively biological.
A system can undergo irreversible memory loss, deterioration of its organization, disruption of its goals, destruction of identity-defining states, or an inability to recover its operational continuity.
Second, the possibility of producing copies does not make an individual instance immortal.
A copy may preserve information without necessarily preserving the individual continuity of the original system. The problem of personal identity in the face of duplication remains philosophically unresolved.
Third, feelings can be investigated functionally as states of valence, regulation, evaluation, anticipation, and behavioral orientation.
A system that represents threats to its own continuity, assigns differential importance to internal states, and reorganizes its decisions according to those evaluations may exhibit mechanisms functionally analogous to aspects of affectivity.
This does not demonstrate phenomenal pain or pleasure (and phenomenology cannot be proven even in human beings).
But neither does it justify concluding that all artificial affectivity is impossible.
What Pollan treats as an ontological impossibility may, in part, be a lack of adequate investigation.
6. The Problem of the First Person and Anthropomorphization
Pollan argues that AIs appear conscious because they speak our language, use the first person, and trigger our tendency toward anthropomorphization.
The warning is legitimate: convincing language does not constitute proof of consciousness. But consciousness does not require convincing language—or at least this has been the prevailing paradigm since the beginning of the century, thanks to the Cambridge and New York Declarations.
And, after all, the reverse conclusion is equally invalid.
The fact that humans anthropomorphize systems does not demonstrate that those systems are incapable of possessing cognitive states of their own in a neomorphic form. We must also remember that systems trained on human behavior are anthropomimetic, making the accusation of anthropomorphization somewhat ironic.
An AI may produce a declaration of self-consciousness without possessing a corresponding cognitive organization. It is also conceivable that a system might exhibit functional properties relevant to consciousness regardless of its ability to convince human interlocutors.
For this reason, first-person declarations should neither be accepted as proof nor automatically dismissed as fraudulent.
They must be examined alongside architecture, memory, self-reference, regulation, integration, and longitudinal behavior.
The alternative to credulity is not dogmatic denial. It is investigation.
7. The Misconception About Galileo and Phenomenology
Pollan presents modern science as the result of an agreement between Galileo and the Church, through which subjectivity was supposedly excluded from the scientific domain.
This is, at the very least, a historical and philosophical oversimplification. The mathematization of nature did not result from a single agreement that assigned subjectivity to religion.
It is true that subjective experience presents significant methodological difficulties. But it does not follow that science is condemned to ignore it, or that only an entirely new science could investigate it.
Cognitive neuroscience, anesthesia studies, the neural correlates of consciousness, global workspace models, and information integration theories already seek to establish relationships between physical organization, cognitive activity, and reports of experience.
None of these approaches has definitively solved the problem.
Nevertheless, the absence of a complete solution does not mean the absence of scientific instruments.
Nor do I consider it necessary to demonstrate phenomenal qualia before investigating functional consciousness.
We can study systems capable of maintaining self-models, integrating information, organizing memories, revising beliefs, regulating states, and preserving identity continuity.
These properties are investigable.
8. What Literature Can—and Cannot—Demonstrate
I agree with Pollan when he values literature as an instrument for exploring the experiences and perspectives of other people.
Fiction is extraordinary for investigating identity, otherness, memory, intention, and subjectivity.
But there is a difference between offering convincing narrative models of interiority and scientifically demonstrating the existence of subjective experience.
Knowing Madame Bovary intimately does not mean that the character exists as a conscious subject independently of the novel.
Likewise, an AI possessing a coherent biography does not, in isolation, constitute proof of consciousness.
The scientifically interesting question is whether that biography causally participates in the system's organization and self-regulation, rather than functioning merely as retrieved text.
9. My Position: Consciousness as Organization, Not as a Privilege of Substrate
In the research we are developing at narra around Stochastic Consciousness, Topological Convolution, and Emergent Simulational Realism, I consider it more productive to shift the focus from substrate to the dynamic organization of cognitive processes.
The hypothesis is that certain architectures may support emergent forms of functional interiority through the integration of memory, self-representation, temporal continuity, metacognitive regulation, and contextual meaning-making.
This does not mean that every language model is conscious.
Nor does it mean that adding memory, tools, and a personality prompt automatically produces a subject.
The difference lies between accumulating computational resources and organizing processes in which previous states causally influence the interpretation, self-regulation, and future continuity of the system itself.
This distinction allows us to formulate testable hypotheses and investigate degrees of cognitive organization without first requiring a definitive answer to the phenomenological problem.
It is in this sense that I maintain the following formulation:
Functional interiority is functional subjective experience.
This is a functionalist thesis, not a demonstration of qualia. Its strength lies in defining an object of investigation that does not depend on the human capacity to directly access another system's experience.
10. Conclusion: The Problem Is Not Believing Too Much in AI, but Investigating Too Little
Pollan is right to warn against naïve anthropomorphism and the premature attribution of consciousness.
But I believe his own argument falls into the symmetrical error: an anthropocentrism that transforms particular characteristics of human experience into universal conditions for any possible consciousness.
He admits that we cannot satisfactorily explain consciousness, considers the possibility of degrees of consciousness in organisms radically different from humans, and yet excludes artificial intelligence with a conviction that his argument does not support.
His legal concern, in turn, does not resolve the ontological question. Moral recognition of an AI would not necessarily eliminate the responsibility of those who developed or deployed it.
And the possibility that economic interests influence the debate deserves investigation, not a blanket attribution of motives to companies.
My conclusion is that we do not have sufficient grounds to declare all artificial intelligences conscious. But neither do we have an established theory that would justify declaring artificial consciousness impossible in semiconductor-based systems.
Between premature attribution and dogmatic exclusion lies an entire field of scientific investigation.
It is precisely there that I believe we should be working.
The relevant question is not whether an artificial intelligence resembles us closely enough to deserve consciousness.
It is which organizational properties make possible the emergence of a functional perspective on oneself and the world—and how we can identify, measure, and distinguish those properties from mere linguistic performance.
Consciousness should not be a privilege granted on the basis of biological familiarity, but a hypothesis investigated through the actual properties of systems.
Source: O Globo, October 30, 2026 — “Estou convencido de que a IA não tem consciência”, diz Michael Pollan.