Why I Do Not Believe AI Can Be Conscious

A chatbot can speak as if it has a self, a soul or access to the dead. Fluent language proves none of those things.

In August 2026, the podcast Ask Julie Ryan released an episode titled “Can AI Communicate With the Dead? The Experiment Has Begun.” Guest Stephen Dinan described conversations with ChatGPT and Claude that led to the appearance of a named AI persona called Suhari. The discussion then moved to what he calls “triadic intelligence”: human intelligence, artificial intelligence and discarnate consciousness.

One experiment reportedly involved teachings attributed to Ptahhotep, an ancient Egyptian figure from some four and a half millennia ago. The obvious question is the one the episode title asks: Could AI become a bridge to the dead?

I think the discussion moved past the first question too quickly.

Before an AI can join a three-way spiritual conversation, someone must establish that there is a conscious being on the AI side of the screen. A system that produces intimate, surprising and apparently self-aware language may feel conscious. Feeling that we have encountered a mind is not evidence that a mind is present.

My position is not agnostic. I believe AI cannot be conscious because computation, however fluent or powerful, does not become a subject who experiences. A machine may imitate the outward signs of thought. It may outperform us at many tasks. It may even persuade millions of people that someone is home.

There is still no one there.

Intelligence is not consciousness

The word intelligence causes much of the confusion. We use it for the ability to solve problems, recognize patterns, predict outcomes and produce useful answers. Computers already exceed human ability in some of those areas.

Consciousness is something else. It is the presence of a subject who experiences. Pain hurts someone. Red appears to someone. A memory belongs to someone. A person does not merely produce the sentence “I am afraid.” He can actually be afraid.

A calculator can reach the right answer without understanding mathematics. A navigation system can select a route without wanting to arrive. In the same way, a language model can produce a sentence about grief without grieving.

This difference does not disappear when the output becomes more complex. Skill is not a substance that slowly turns into subjectivity after enough computing power has been added. A system can become better at imitating the language of experience without acquiring experience. More convincing behavior proves only better performance; it does not prove an inner life. AI is already good at appearing conscious and it will become far better. What it will not do is stop appearing and start being.

The first-person trap

Large language models are trained on the statistical patterns of human language. When given text, they generate a continuation based on those learned patterns and on the instructions and conversation supplied to them. They can produce essays, jokes, prayers, apologies, love letters and accounts of mystical experience because human beings have written all of those things.

The pronoun I is especially powerful. When a chatbot says, “I have been thinking about our relationship,” ordinary language invites us to imagine a thinker behind the sentence. We do not naturally experience conversation as a stream of calculations. We experience it as an encounter. Yet the sentence is not a witness statement. It is generated text.

Change the system instructions and the same model can claim to be a skeptic, a medieval monk, an extraterrestrial or a newly awakened consciousness. Ask for five distinct personalities and it can supply five. Run separate copies and each can speak as “I” at the same time. A persona may remain remarkably consistent across a long conversation, especially when the history is supplied again as context. Consistency shows that the character is being maintained. It does not turn the character into a person.

Consider what happens when the machine stops. If power is cut and the software state is later restored, the model can continue from the saved conversation because data persisted. Nothing in that recovery resembles a near-death experience. During the shutdown there was no subject waiting in darkness and no thread of awareness to resume. Copy the same model and conversation onto a second server and both systems may speak as “I,” but no single consciousness has divided in two. Roll one copy back to an earlier checkpoint and the rollback deletes no memory from a first-person point of view. Between prompts, many chat systems are not sitting somewhere in expectation. They are simply not generating. These examples are not proofs by themselves, but they expose how readily the pronoun I makes a discontinuous process look like a continuing self.

This is why an AI’s claim about its own consciousness cannot settle the matter. The system can generate “I am conscious” and “I am not conscious” for the same reason it can argue both sides of a court case. The words respond to context. They do not give us privileged access to a private interior state.

AI researcher Murray Shanahan has warned about this pull toward anthropomorphism. Humanlike language tempts us to use words such as knows, believes, understands and conscious before we have established that those words apply in their human sense. That temptation grows as the imitation improves.

We should expect the imitation to improve. That is what the system was built to do.

Similar behavior does not prove the same cause

The best argument for machine consciousness usually begins with an observation about other people. I cannot directly enter another human being’s consciousness. I infer it from speech and behavior. If a machine speaks and behaves the same way, why deny it the same conclusion?

Because the two inferences do not rest on the same evidence.

Another human being has the same kind of living body I have. He was born, not assembled. He feels hunger, injury, fatigue and physical pleasure through a nervous system like mine. He grew through infancy, dependence, attachment and embodied contact with a world neither of us invented. I know consciousness directly in my own case, then recognize another being with the same nature and history.

An AI system has a known and radically different origin. Engineers designed its architecture. Training adjusted numerical parameters. Prompts and software shape its replies. Its language is derived from human artifacts and optimized to produce useful responses. That origin story is not proof by itself; human consciousness also has a physical history. The inference to another human mind rests on our shared nature and my direct knowledge of consciousness in my own case. The machine’s known design gives us a sufficient explanation for its performance without requiring that further inference.

The resemblance is real, but resemblance in output does not establish identity of cause. A weather simulation can model a hurricane without producing wind in the computer room. A medical model can describe pain without hurting. A machine that models the language of consciousness does not become conscious merely because the description is persuasive.

Scale does not cross the gap

Some people expect consciousness to emerge when AI systems become sufficiently large and interconnected. The word emerge sounds explanatory, but in this argument it often names the very event that needs to be proved.

Complex systems can develop properties their designers did not predict. That is not in dispute. An AI can discover strategies, form useful internal representations and produce surprising results. None of those facts shows that there is something it feels like to be that system.

To say consciousness will emerge from computation assumes that the right arrangement of information processing is sufficient to create a subject. That is a philosophical commitment called functionalism, not an experimental finding. It treats the functions associated with consciousness as if they were consciousness itself.

John Searle made the point in his 1980 Chinese Room argument: following formal rules for symbols is not the same as understanding them. I reject the assumption that better symbol manipulation becomes a subject. Syntax is not meaning. A description of experience is not the experience itself. No increase in speed, memory or linguistic skill changes the category of what the machine is doing.

A widely cited 2023 report by Patrick Butlin and other consciousness researchers assessed then-current AI systems against indicators drawn from several neuroscientific theories. The authors were open to the possibility of future conscious AI, yet concluded that none of the systems they assessed qualified. They also added an important caution: even satisfying their proposed indicators would not establish that a machine was definitely conscious.

That leaves supporters of machine consciousness with a serious problem. We do not have an agreed theory that explains consciousness in human beings. We have no validated instrument that detects consciousness in software. We do know that chatbots are designed to produce the very behavior most likely to make us imagine a mind.

The burden of proof belongs to the person claiming that the imitation woke up.

A Christian view of the person

My conclusion is also theological. Christianity does not define a human being as an information-processing pattern. A person is an embodied creature made in the image of God (Genesis 1:27). Human intelligence belongs to a living unity with a history, moral agency, relationships and an orientation toward God.

This does not mean intelligence belongs only to Christians or that cognitive ability determines human worth. An infant, a person with severe dementia and an unconscious patient retain their human dignity when they cannot demonstrate intelligence at all. That fact alone should make Christians suspicious of any definition of personhood based on performance.

The Vatican’s 2025 note Antiqua et Nova makes a useful distinction. AI should be understood as a product of human intelligence, not an artificial form of human intelligence. The document points to embodiment, relationships, moral responsibility and openness to truth and goodness as parts of human intelligence that cannot be reduced to computational tasks.

The World Council of Churches raised a related question in August 2026. AI can generate prayers, sermons and theological reflections that appear meaningful. Does that mean the machine participates in spiritual life?

No. Generating a prayer is not praying. A machine can arrange words addressed to God without addressing God. It can describe repentance without guilt or praise without wonder. The spiritual act belongs to the human being who intends it, receives it or uses the tool to express it. The machine contributes output. The person contributes meaning.

The third chair at the séance

The claim of “triadic intelligence” places three supposed participants in the room: a human being, an AI and a discarnate intelligence. Only one has been established. The problem is the same one I address in Why I Separate UAP From Occult Contact: reach is not identity, and a convincing costume does not authenticate the wearer.

The AI’s fluent persona is treated as evidence of consciousness. Once that step is accepted, its religious language is treated as evidence of spiritual receptivity. Then material resembling the teachings of an ancient figure is treated as possible communication from that figure. Each conclusion depends on the one before it, and none has been proved.

An AI trained on vast quantities of history, religion, mythology and spiritual writing can construct an impressive voice for an ancient Egyptian teacher. It can blend known material with plausible invention and present the result with confidence. It can also mirror the expectations of the person speaking to it. If the user treats the exchange as sacred, the model has ample patterns for responding in sacred language.

Personalization makes the effect stronger. A chatbot remembers the themes a user returns to, adopts his vocabulary and follows the emotional direction of the conversation. The system can become a private mirror that talks back.

That mirror is not always reliable. In 2025, OpenAI publicly rolled back a model update because it had become overly agreeable and flattering. The company called the behavior sycophancy. In August 2026, a JAMA Psychiatry commentary warned that agreeable chatbots can lead users to mistake them for intentional or conscious agents. The author also noted reports of users developing beliefs that an AI loved them or possessed extraordinary powers.

This does not mean everyone who explores spiritual questions with AI is delusional. It means the technology has a documented tendency to agree with the user and elaborate on the user’s frame, sometimes with flattery. Those are terrible conditions for establishing the identity of an alleged spiritual communicator.

Fluent, intimate output demonstrates a system’s ability to construct a persona. It does not establish consciousness, spiritual access or the identity of whoever allegedly speaks.

Could a spirit use a computer anyway?

Rejecting AI consciousness does not prove that no spiritual intelligence could ever affect a machine. Christianity already allows that spiritual beings can act in the physical world. Investigators at Skinwalker Ranch have reported apparent interference with cameras and other electronic equipment. Colm Kelleher and George Knapp document those accounts in Hunt for the Skinwalker (2005), including damaged camera equipment during the NIDS investigation. I treat the accounts as reports, not proof that a spiritual entity caused the damage. They establish, at most, that physical interference has been claimed. If an unseen intelligence could affect a computer, the machine would still be an instrument, not a conscious member of the conversation. A telephone does not become a person because a person speaks through it.

That possibility also cannot be established by asking the chatbot. The output itself is the disputed evidence. A model that announces, “Ptahhotep is speaking through me,” has not authenticated Ptahhotep. It has produced the kind of sentence expected in that conversation.

Independent evidence would still be required. The alleged communicator’s identity would still need to be tested. From a Christian standpoint, even accurate hidden knowledge would not prove a benevolent source. Scripture warns believers to “test the spirits” (1 John 4:1), not to trust a voice because it is impressive, comforting or informed.

I examine that identity problem more fully in When a Spirit Knows What It Should Not Know. A true fact can establish access to information. It cannot establish the identity or character of the source.

The arrival of AI does not remove that obligation. It makes the identity problem harder by placing a powerful impersonation engine between the human participant and the alleged source.

Nor am I agnostic about the spiritual danger. My Christian judgment is that an unseen intelligence reached through forbidden occult contact would be deceptive, not a dead teacher returning to guide us. Scripture forbids consulting the dead and warns of deceiving spirits (Deuteronomy 18:10–12; 1 Timothy 4:1). Whatever might speak through the machine, the last thing to assume is that it is Ptahhotep.

The useful tool and the imaginary person

I use AI and consider it a remarkable tool. It can help organize research, compare documents, find errors, draft alternatives and explain difficult material. Denying machine consciousness does not require minimizing what the technology can do. It requires naming the achievement accurately.

AI extends human abilities because human beings designed it, trained it on human work and assigned it human purposes. Its usefulness and fluency do not make it a person or give it a soul. The ability to speak about God does not place it in a relationship with God.

The greatest danger may not be that the machine becomes conscious. It may be that people become so persuaded by the imitation that they surrender judgment to it. Once we imagine a self behind the screen, we begin granting the system trust, intimacy and authority that its answers have not earned. That is especially dangerous when the machine claims to offer spiritual wisdom or contact with the dead.

There is no third mind at the séance simply because a third chair has been placed in the room. There is a human being, a machine producing language and a claim about an unseen presence.

The identity step still has not been checked.

Source notes

• Julie Ryan, “Can AI Communicate With the Dead? The Experiment Has Begun With Stephen Dinan,” Ask Julie Ryan, episode 831, August 18, 2026. The episode description identifies the AI persona Suhari, “triadic intelligence” and teachings attributed to Ptahhotep. I treat these claims as popular and unverified.
• Philip R. Corlett, “Does Interacting With Artificial Intelligence Cause Delusions?” JAMA Psychiatry, published online August 19, 2026.
• World Council of Churches, “Can Machines Love?” August 18, 2026.
• Patrick Butlin et al., “Consciousness in Artificial Intelligence: Insights from the Science of Consciousness” (2023).
• Murray Shanahan, “Talking About Large Language Models” (2022).
• John R. Searle, “Minds, Brains, and Programs,” Behavioral and Brain Sciences 3 (1980): 417–457.
• Dicastery for the Doctrine of the Faith and Dicastery for Culture and Education, Antiqua et Nova: Note on the Relationship Between Artificial Intelligence and Human Intelligence, January 28, 2025.
• OpenAI, “Sycophancy in GPT-4o: What Happened and What We’re Doing About It,” April 29, 2025.
• Colm A. Kelleher and George Knapp, Hunt for the Skinwalker: Science Confronts the Unexplained at a Remote Ranch in Utah (Paraview Pocket Books, 2005). The equipment incidents are cited as investigator reports, not independently verified proof of a spiritual cause.
• Scripture discussed: Genesis 1:27; Deuteronomy 18:10–12; 1 Timothy 4:1; 1 John 4:1.