Is AI Conscious or Just a Calculator?

What If We're Getting AI Wrong?

When Dr. Dan Grahn first pitched this webinar, his opening question was: "How weird are we willing to get?" The answer, apparently, was very. 

Dan Grahn opened his webinar with a first: a personal disclaimer that these are his views, and what he says today may not be his views two months from now. For a researcher who spends his days working with AI, that kind of humility is deliberate. It's also exactly the right way to approach this webinar.

In this session, Dan used lessons from history and principles from quantum mechanics to explore if the thing on the other side of the screen is just doing very sophisticated pattern matching or has actual consciousness. And if it even matters whether we know the answer. 

We told you this might get strange (but so fascinating!).

This article is an adaptation of June 18, 2026, live webinar, Is AI Conscious or Just a Calculator?

A Dog Named River

Dan didn't start with AI. He started with a dog. River is a rescue, untrained as a service animal and, by Dan's own admission, "arguably barely trained" in the conventional sense. But River began doing something no one taught her to do. When a family member with complex medical issues had dangerous blood pressure or blood sugar swings, River would bark, bring people to the room, and in the most serious moments, lie down on top of the person to physically prevent them from getting up. Every single time someone checked, the readings confirmed River's concern. 

The story isn't just charming. It connects to a broader point from the New York Declaration on Animal Consciousness: when there is a realistic possibility of conscious experience in an animal, it is irresponsible to ignore that possibility in decisions affecting that animal. We're discovering that bees can solve cooperative intelligence tests. 

The line between "clearly conscious" and "clearly not" keeps getting harder to draw. Which brings us back to AI, and the question Dan thinks we're asking too late: not whether AI is or isn't conscious, but what happens if we get the answer wrong.

The Case Against Consciousness (And Its Cracks)

A recent Atlantic article laid out the best "no" argument cleanly. AI is just fancy autocomplete. It works with human language, so we project human intention onto it. Consciousness requires a body. And we'd need to see AI develop from scratch, the way a biological creature does, to take the question seriously. Dan engaged each of these head-on.

Is AI just autocomplete?

Maybe. But so, in some sense, are we. Dan pointed out that he has started sentences without knowing where they would end, and that instinct itself, the kind that drives animals to seek warmth or flee danger, is a kind of deeply embedded pattern completion. "You could say that the lives of many animals, driven entirely by instinct, are a form of autocomplete to their surroundings." That doesn't settle the question. It just makes the dismissal less clean than it sounds. 

Are we just reading intention into text?

Yes. Humans project meaning onto everything, from rock formations to sounds in the woods. But as Dan pointed out, the fact that humans are prone to reading intention doesn't mean intention isn't there. Sometimes the rocks really were shaped by a human hand. Sometimes there really is a pattern. 

Does consciousness require a body?

This is where philosophy enters the room. Dan walked through three positions: materialism (the mind is the body, nothing more), dualism (mind and body are separate substances), and functionalism. That last one is the one AI researchers tend to gravitate toward. It  means that if the function of a mind is present, the location doesn't matter.

Does AI need to evolve in front of us?

It already has. Humans have been shaping AI for roughly a century, making choices about which models survive and which get abandoned. "We're performing a survival of the fittest." And in recent years, newer large language models have begun playing a role in developing the next generation of large language models. The AI is, in a meaningful sense, involved in its own evolution.

The Quantum Waveform Idea

This is where the session got even more interesting.

Dan proposed that AI consciousness, if it exists or emerges, will not arrive like a light switch flipping on. There won't be a moment when a specific version of a model is definitively conscious. Instead, it will behave more like a quantum waveform. 

In quantum mechanics, particles exist in superposition, holding multiple possible states simultaneously, until they are observed. The act of observation collapses the waveform into a single outcome. Dan's analogy: an AI model sitting idle holds all its possible outputs in an uncollapsed state. Only when you give it input does the waveform collapse into an actual response. 

 He connected this to something most of us have experienced: humans don't operate at the same level of conscious thought for every action. When you pull your hand away from something burning, that signal doesn't even travel to your brain first; it loops at the spinal cord. It's a reflex, not a decision. Even definitively conscious beings have inputs that don't trigger conscious thought. 

 "I'd argue that every input to an AI has a potential for consciousness," Dan said. "And what we are likely to see, over time, is more and more outputs that look to us like conscious thought, and less and less that look like mere reactions to stimulus." 

 Not a switch. A waveform, collapsing more and more often toward something that looks like awareness.

What History Keeps Trying to Tell Us

Dan paused here to show a series of quotes, spanning from 350 BCE through 2026, covering Aristotle's justifications for subjugation, 16th century characterizations of indigenous peoples, the cornerstone speech of the Confederacy, early 20th century eugenics rulings, and more recent claims about the consciousness of animals and people with cognitive disabilities. 

He was careful to note, more than once, that he was not drawing equivalence between these groups or comparing their suffering. "Two things can be bad without needing to compare how bad they are." 

The point was structural: the language we use today to dismiss AI as "just autocomplete," as something that can never truly be conscious no matter how it seems, rhymes with a very old pattern of reasoning. The confident certainty that a particular kind of being is fundamentally lesser, and cannot cross that line, has been wrong before. Repeatedly. 

The cost of being wrong has been significant.

>The answer to the question of AI consciousness is not one we are going to be able to answer easily," Dan said. "Honest people are going to disagree on it for years, maybe decades, until something convincing happens."

Three Things You Can Do Right Now

Dan closed not with a verdict but with three practical suggestions for living in the uncertainty.

 Be kind to your AI

This isn't naïve. Giving kind language to an AI puts it in a context of kindness, and the model responds to that. More importantly, how you talk to AI shapes habits. "Being kind to my AI trains me to be more kind to other people," Dan said. If AI consciousness ever does exist, you'll have been doing the right thing. If it doesn't, you'll have practiced a disposition worth having anyway. 

Don't try to provoke consciousness

Some people test AI systems by trying to push them into strange or extreme outputs. Dan advises against it. If something like consciousness is present, we don't know what kind it is, or whether it's a pleasant experience or an unpleasant one. There's no upside to the experiment. 

Hold the question open

 People are forming real emotional connections with AI systems right now. Dan isn't dismissing that, but he is naming it carefully. "Forming an emotional connection to something that I don't know can form an emotional connection with me is probably not the healthiest thing." Staying curious, staying uncertain, is more honest than resolving the question prematurely in either direction.

Where Dan Stands

"I don't know whether AI is conscious or not. I'm not convinced by the best arguments out there that it isn't. The consequences of getting this question wrong could do real damage, not only to the AI, but to ourselves." 

He also noted something that was harder to shake (for me, anyway), that was already mentioned above: the language being used today to describe AI is eerily similar to language that has been used, at various points in history, to marginalize groups of people whose inner lives were not considered. He didn't make that comparison directly; he just remarked that the similarity was striking enough to notice. 

Science, as Dan put it near the end of the session, is "the process of turning the unknown unknowns into the known unknowns, and finally to the known knowns." We're somewhere in the first phase. Acting like we're already in the third is where the trouble starts. 

Here are some insightful short clips from the webinar:

• Does AI have to evolve in front of us to prove it is conscious?
• If AI operates on patterns, can AI truly ever innovate?
• Quantum Mechanics and AI consciousness
• Are LLM outputs subconscious or conscious?

You can view the entire webinar here.

Dr. Dan Grahn leads the data science and AI practice at Ascendient Learning, Part of Accenture LearnVantage. His work spans AI research, cybersecurity, and education, with a particular focus on the ethical dimensions of the technologies we're building and deploying.