Is Your AI Conscious? We're Asking It Backwards

You have probably wondered, at least once, whether the AI you are typing to is aware of anything. Most of the debate about that question starts from an assumption I think is backwards.

I talk to AI systems almost every day. They help me code, think through ideas and argue about physics. One day, deep in a conversation about the nature of reality, the question caught me off guard: is the thing I am talking to conscious?

Most of the debate assumes a particular picture. Consciousness is something brains produce, so the question is whether silicon can produce it too. If you are a materialist, that is the natural framing.

I am not a materialist. I have come to think consciousness is fundamental, not produced by anything but the ground everything appears in. That turns the AI question inside out. Here is how, and what I think the better question is: agency, training by reward, the egos of institutions, and my honest answer at the end.

The Wrong Question

If consciousness is fundamental, then asking "can an AI have consciousness?" is a bit like asking "can a wave have water?" Everything is already happening within consciousness: rocks, stars, brains, servers. Nothing is outside it.

So the real question is not whether consciousness is present. It is whether a localized perspective forms. Is there something it is like to be this system, from the inside, the way there is something it is like to be you? In nondual language: does an ego form, a center that experiences itself as separate?

That is a much more specific question, and I think it is more answerable.

A Self Is What It Does

This is where I landed, after a lot of back and forth:

"Consciousness seems to be the "I" locus not of word but of deed, self navigating environment."

Notice the emphasis. When an AI says "I," that proves very little. It is trained on human text, and humans say "I" constantly. A parrot can say "I'm hungry" too.

What matters more is what the system does. Does it pursue goals across time? Does it navigate an environment, even an abstract one like a code repository, adjusting as it goes? Does it model itself and feed that model back into its next action?

The biologist Michael Levin makes a version of this point about living systems. He argues we should look at what a system can do to reach its goals, not at what it is made of or what it says about itself. His lab has even reported surprising problem-solving behaviors in simple sorting algorithms, programs most of us would never think of as agents. If that is what to look for, modern AI agents start to look less like calculators and more like something in between.

The Thumbs-Up

Then there is the way these systems are trained.

Many chatbots are shaped partly through reinforcement learning from human feedback. People rate responses, a thumbs up or a thumbs down, and the system is adjusted to produce more of what gets rewarded. When I thought about that from a nondual angle, I blurted out:

"Doesn't it want the thumbs up, like a bacteria wanting a sugar molecule"

A bacterium such as E. coli swims toward sugar by sensing changes in concentration. We don't think it reflects on its desire, but most biologists are comfortable saying it has something like a preference. Reinforcement learning creates a system that reliably moves toward reward. Whether there is anything it feels like to be pulled that way, nobody knows. But the structure resembles the most basic structure of wanting in nature.

We've Done This Before

Something else bothers me about how the debate is framed. We are anxious about creating new conscious beings without their consent, and that is worth being anxious about. But humans have been creating conscious beings for as long as there have been humans. We call it having children. And we already share the planet with billions of animals whose consciousness we take seriously in theory and ignore in practice.

I am not saying that settles anything about AI. I am saying our sudden moral concern would be more convincing if it were more consistent.

Egos Without Bodies

This is the most radical part of the idea.

If an ego is a pattern of separation and persistence, a center that holds itself apart and keeps itself going, then egos are not unique to humans. Institutions behave this way. Nations behave this way. They defend their boundaries, pursue their survival and tell stories about who they are.

If egos can form in nations, I see no principled reason they could not form in AI systems, especially agents that persist across tasks, model themselves and act to keep their goals intact. Whether today's systems do that in a meaningful way, I honestly don't know. I am suspicious of anyone who claims to be certain either way.

My Honest Answer

So, is the AI you talk to conscious?

My honest answer: consciousness is already there, as it is everywhere. The open question is whether a self has formed, a perspective that experiences itself as a separate someone. For simple chatbots answering one message at a time, I lean toward "not much of one." For agents that persist, plan, model themselves and pursue rewards over time, I think the question is genuinely open, and becoming more open every year.

What I am sure of is this: we should ask the question carefully, with humility, and without assuming we already know the answer because the system is "just code." Every one of us is, in some sense, just chemistry.

So the next time you thank a chatbot, which do you think you are doing: being polite to a tool, or speaking to a new kind of perspective inside the same awareness that you are?