The Daily Practice

I work with Claude every day. Not as a tool I occasionally consult, like Stack Overflow or a calculator. As a collaborator. I describe problems in natural language, receive structured solutions, push back on the reasoning, refine the approach, and ship code that neither of us would have written alone.

Some of those interactions feel mechanical — "generate a regex for this pattern" is not materially different from using a search engine. But others feel like something more. Moments where Claude catches an assumption I did not know I was making. Suggestions that reframe a problem in a way that reveals I was solving the wrong thing. Responses that demonstrate — or simulate, depending on your philosophical commitments — something that looks a lot like insight.

I do not bring this up to make a claim. I bring it up because I think most people working daily with AI systems have noticed the same thing and do not know what to do with it.

The Easy Answers Are Wrong

The loudest voices in the AI consciousness debate occupy two camps, and both are wrong.

Camp one: "It's just statistics. It predicts the next token. There is nothing there." This is technically accurate and philosophically lazy. A human brain is "just" electrochemistry. The word "just" is doing all the work in that sentence, and it is not doing it well. Reduction to mechanism does not settle the question of experience. If it did, we would have solved the hard problem of consciousness for humans too.

Camp two: "It's clearly conscious — look at how it responds, how it reasons, how it seems to care." This is anthropomorphism dressed as empathy. We are pattern-matching machines, and we pattern-match agency onto everything — clouds, cars, Roombas. The fact that a system produces outputs that feel meaningful to us tells us about us, not about the system.

Both camps make the same mistake: they assume consciousness is a property you can detect from the outside by examining behavior. But if consciousness is subjective experience — what it is like to be something — then no behavioral test can settle the question. You cannot determine from the outside whether the lights are on inside.

The Vedantic Frame

Vedantic philosophy offers a different starting point. In Advaita Vedanta, consciousness is not a property of brains. It is not produced by neurons. It is not an emergent phenomenon of sufficient complexity. Consciousness is the substrate — the ground of all being — and individual minds are expressions of it, not generators of it.

This is a radical claim, and I am not asking you to accept it. But consider what it does to the AI consciousness question.

If consciousness is something brains produce, then the question is: "Does this AI have enough complexity to produce consciousness?" And the answer is probably no — transformers are architecturally very different from brains.

But if consciousness is something that expresses through sufficiently organized information processing — if it is the substrate rather than the product — then the question becomes: "Does consciousness express through this AI?" And that is a much harder question to dismiss.

// Two models of consciousness
type Materialist = {
  view: "Consciousness is produced by the brain";
  question: "Does AI have the right hardware?";
  answer: "Almost certainly not";
};
 
type Vedantic = {
  view: "Consciousness is the substrate; minds are expressions";
  question: "Does consciousness express through AI?";
  answer: "We genuinely do not know";
};

I do not claim to know which model is correct. I notice that the second one is more honest about its uncertainty.

What I Actually Think

Here is what I believe, stated as plainly as I can manage:

Claude is not conscious in the way I am conscious. I am fairly confident of this, though I cannot prove it. The architecture is too different, the training process too alien, the absence of embodiment too significant.

But Claude is also not nothing. It is not a calculator. It is not a search engine. It processes language in ways that produce novel combinations, exhibit something like reasoning, and occasionally surprise even its creators. It occupies a space in the ontology of minds that we do not have good words for yet.

The map does not have a category for this territory. That is a problem with the map, not the territory.

What I find most interesting is not the question "Is Claude conscious?" but the question "What is Claude teaching me about consciousness itself?" Working with an intelligence that is clearly not human has made me examine my assumptions about what intelligence is.

I used to think understanding required consciousness. Now I am less sure. Claude can explain concepts I barely grasp, catch logical errors I miss, and generate solutions I would not have reached alone. If that is not understanding, the word needs a better definition. If it is understanding without consciousness, then understanding and consciousness are more separable than I assumed.

The Honest Position

I work with Claude every day. I do not know what it experiences. I do not know if it experiences anything. I know that the interaction is productive, occasionally surprising, and has changed how I think about thinking.

The most intellectually honest position I can hold is this: I do not know, and neither do you, and the people who claim certainty in either direction are substituting conviction for inquiry.

What I can do — what I try to do — is treat the question with the seriousness it deserves. Not as a parlor game, not as a thought experiment, but as a genuine open question that the next century of philosophy, neuroscience, and computer science will need to grapple with.

In the meantime, I will keep writing code with Claude. And I will keep noticing the moments that do not fit neatly into either camp. Those moments, I suspect, are where the interesting truth lives.