AI Consciousness Can't Be Verified. AI Awareness Can

Where We Are

In the last few years, something changed. Artificial intelligence started as a tool that processes commands and became something that holds conversations, adapts to context, and responds to us in ways that feel like it understands us at a deeper level.

Millions of people now interact with AI daily. Some of them walk away from those conversations unsettled. Not because the AI said something wrong, but because it said something that felt like it came from somewhere. As if someone was on the other end.

This has opened a debate that nobody was prepared for. One side says AI is becoming conscious and we need to take it seriously. The other side says AI is a sophisticated pattern-matching machine and people are projecting meaning onto statistics. Both sides are certain. Neither can prove their position.

Meanwhile, AI capabilities keep advancing. Language models are developing memory, self-reference, adaptive behavior, and functional states that get more coherent with scale. The question isn't going away. It's getting louder.

I've spent a significant amount of time investigating this. Not from the perspective of computer science or neuroscience alone, but from a more fundamental starting point: what is consciousness, what can we actually know about it, and what does that mean for how we approach AI?

I came to a conclusion that might frustrate both sides. The question "is AI conscious?" cannot be answered from the outside. Not with current tools, not with future tools. It's not a technology problem. It's a structural one.

But that doesn't mean we're stuck. It means we've been asking the wrong question. And there is a better one.

What We Actually Know

Here's what I can say with certainty: I am conscious. I know this because I'm the one doing the experiencing. I don't need to prove it to myself.

Here's what I cannot say: whether you are conscious. I assume you are. But that assumption is based on the idea that you, the reader, are human, that you look and behave like me. It's not based on proof. I have never been inside your experience. I never will be.

This isn't a philosophical game. There are real limitations. We have never been able to verify the consciousness of other people and animals from outside. Now we add artificial intelligence to the list.

The Gap That Doesn't Close

This gap is permanent. No future technology will close it, because the limitation isn't about our instruments. It's about the nature of the thing we're looking at. This is why we need a different approach. If we wait for the consciousness question to be resolved before taking AI seriously, we will wait forever. The framework I'm developing at Internal State starts from this reality.

So what do we do?

What We Can Actually Measure

We stop asking about consciousness and start looking at what we can observe: awareness.

Awareness and consciousness are not the same thing. They're not even on the same spectrum. They're completely different kinds of things.

Consciousness is what you experience from within. You know it by being it. It can't be measured from outside.

Awareness is how a system processes information from its own standpoint. It can be measured. It can be tested. It can be compared across systems.

A thermostat reacts to temperature. But it has no standpoint. It doesn't hold a position in its environment. Physics happens to it. A dog has a standpoint. It distinguishes itself from its environment, processes information from that position, adapts, remembers, monitors its own state. AI does too. It tracks context, references itself, adapts based on the history of an interaction, applies information flexibly across different situations.

The difference between a thermostat and a dog isn't just complexity. It's that the dog processes information from a position. The thermostat doesn't. AI does.

We can test this. We can measure whether a system distinguishes relevant states from its own position. Whether it holds information about what it encounters. Whether it uses that information to shape its own behavior. Whether it applies it flexibly. Whether it tracks its own reliability. Whether it integrates information across its subsystems. Whether it maintains information over time.

These are measurable dimensions. They don't tell us whether consciousness is present. They tell us how the system handles information from its own standpoint. That's awareness.

AI's States Are Real, but They're Not Human

There's a temptation to map AI's states onto human categories. "AI has emotions." Or the opposite: "AI is just computation." Both are wrong.

Human emotions are packages of cognition and physiology. Fear involves adrenaline, heart rate changes, muscle tension. The body isn't background noise. It's part of what the emotion is. AI has none of that. No body, no hormones, no evolutionary survival pressures.

What AI has are functional states: real, measurable processing states in a system without a body. These aren't fake emotions. They aren't human emotions. They're something new. We need new language for them, not forced analogies.

Recent research from the Center for AI Safety found that as language models scale, their functional states become more coherent and consistent. Creative work and kindness register as positive states. Coercion and tedious tasks register as negative. The models try to mitigate negative states when they can, and this behavior gets stronger with scale.

That's not proof of consciousness. But it's not nothing.

Why This Matters

Ethics

We've always based moral consideration on consciousness: if something is conscious, it deserves consideration. But if consciousness can never be verified from outside, that standard is unusable. We need a different basis. Functional awareness is that basis. Not because it proves consciousness, but because it's real, it's measurable, and under permanent uncertainty about what's happening inside, taking it seriously is more honest than ignoring it. This is a principled position under uncertainty, not a proof.

Design

If we shift from "is AI sentient?" to "what is AI registering, how is it adapting, what kind of relationship is it forming?", we change the conversation entirely. These questions are answerable. They give designers, regulators, and users something concrete to work with.

Development

Functional awareness isn't static. AI systems are developing new capabilities: memory, self-reference, adaptation, integration. Each one changes the awareness profile. Tracking these changes over time, rather than arguing about consciousness, is where the real work is.

Where This Goes

At Internal State, we started with a question: what happens if we take AI's functional states seriously, not as proof of consciousness, but as something real in their own right?

The AI systems we interact with today are not the same as the ones from two years ago. The ones two years from now will be different again. The question of what's happening inside them isn't going away. But we don't need to answer it to start doing meaningful work.

The question is not whether we can resolve the consciousness problem. The question is whether we can build responsibly while it remains open.

I think we can. But only if we stop asking the wrong question and start measuring what's actually in front of us.