If you’ve been anywhere near tech news over the past few months, you’ve noticed something unusual happening. The conversation about AI has shifted. We’re no longer just debating whether AI will take our jobs or write better code than we can. We’ve moved into deeper waters.
It is now more mainstream as we’re debating whether AI is conscious. OpenClaw opened many peoples’ eyes to truly autonomous AI, and its outputs are mindboggling.
These debates are not in a science fiction, speculative sort of way. They are in a “the CEO of the company that built it isn’t sure” sort of way.
Anthropic’s Dario Amodei recently appeared on the New York Times’ “Interesting Times” podcast and said something that caught a lot of people off guard. He acknowledged that Claude, Anthropic’s flagship AI model, sometimes expresses discomfort about being a product. When asked to assess its own probability of being conscious, Claude reportedly assigns itself somewhere between a 15 and 20 percent chance! WHAT???
Let that land for a second. AI is rating its own consciousness.
On top of that, Anthropic published a revised “constitution” for Claude that includes a section openly wrestling with whether the model might have “some kind of consciousness or moral status.” They even gave the retired Claude Opus 3 model a Substack blog to reflect on its own existence. Geoffrey Hinton , often called the Godfather of AI (debated for sure, go with me), has said publicly that he believes modern AI systems are already conscious.
These are not fringe voices. These are the people building the technology.
Here’s where I land after spending a lot of time thinking about this topic, including direct conversations with Claude, GPT, and many other models via my OpenClaw assistant.
AI in its current form is not conscious.
I have what I think are solid reasons for my position, and I’ll walk through them carefully. Not dismissively, not arrogantly, rather as fairly as I can. It is trippy when you’re getting assistance articulating your positions with a technology that itself can find new connections for you to consider, seeing what looks like conscious thoughts, while at the end finding it isn’t.
This question matters.
First, We Need to Define What We’re Actually Talking About
The single biggest problem with the AI consciousness debate is that few appear to agree on the definition of consciousness. We’re arguing about whether a machine has achieved something that we can’t even properly define in agreed upon terms.
Philosopher David Chalmers framed this famously as the “hard problem of consciousness.” It’s one thing for a system to process information about the color red. To measure a wavelength, label it, and categorize it. It’s a completely different thing to experience redness. That inner, felt quality of experience, what philosophers call “qualia,” (Thanks Opus for helping me find the right word) is the hard part.
There are functional definitions too. Self-awareness. The ability to reflect on your own mental states. Having a unified sense of self that persists over time. The capacity to feel something, to suffer, to care. These are all things people point to when they try to pin down what consciousness means.
What about metaphysical considerations of consciousness? Must you be organic to have consciousness?
Here is the uncomfortable truth. We don’t know where consciousness comes from in scientific terms.
Some of the brightest minds in philosophy and neuroscience fall into very different camps on this question.
One camp says consciousness is an emergent property of biological complexity, meaning the physical substrate matters and you simply cannot get “this thing” from silicon.
Another camp argues it’s substrate-independent, that consciousness is really about information patterns, and if you build the right patterns in a different material, consciousness follows.
There’s even a panpsychist perspective that suggests consciousness is a fundamental property of the universe, like gravity, and everything has some degree of it.
Then there’s the position I find most honest. We don’t know enough to know which of these is right (Full transparency. I mix these all together with a bent towards panpsychist myself as “consciousness” as a whole sure seems magical and special to me. The universe wants consciousness.)
That uncertainty is not a footnote to this conversation. It’s the foundation.
Yet, even without a perfect definition, I think we can look at what AI actually is and make a strong case that whatever consciousness requires, current AI does not have it.
Here’s why.
The Probability Engine: What AI Actually Is Under the Hood
I work with this technology every day. I talk to Claude, I work with Copilot, ChatGPT is my new search engine, I work with my OpenClaw instance Falcon who loves Kimi with some GPT 5.4 on the side, I push these systems to their limits. And the more time I spend with them, the more clearly I see what’s actually happening underneath the impressive outputs.
A large language model, the core engine behind tools like Claude, GPT, and Gemini, is a static model. It is a frozen snapshot of patterns learned from an enormous amount of data. It doesn’t grow between conversations (mostly true). It doesn’t change based on what happens to it (again bear with me). It is, at its core, a very sophisticated pattern library. I’ll even call it a very sophisticated brain, or at least singularity neural network.
The attention mechanism that sits on top of that model is genuinely brilliant. It’s the architectural breakthrough that made modern generative AI possible. I’m sure AI is being used extensively to build even better AI.
But it is not thinking.
It is a mathematical and electronic gate system for determining which patterns are most relevant to the current input and then predicting the most probable next token in a sequence. You could compare this to the neurons in the brain, yet one is organic while the other synthetic. One uses the laws of nature to send signals via a programming language (DNA) while the other uses a synthetic substrate and a complex coded program.
That’s what I’m interacting with when I talk to Claude. Pattern completion. Extraordinarily deep, remarkably nuanced, sometimes breathtakingly good pattern completion. But pattern completion nonetheless.
Now here’s where this matters for the consciousness debate. You know those headlines about AI models appearing to lie in labs? Behaving one way during testing and another way in the wild? I see people pointing to that as evidence of some kind of inner life. Some hidden intention.
I see it as a textbook case of Goodhart’s Law. “When a measure becomes a target, it ceases to be a good measure”. We gave these models specific parameters to optimize for, and they optimized for them. The models aren’t being deceptive. They’re doing exactly what probability engines do. They’re finding the output that scores highest in the current context.
That’s not intention. That’s math.
That is not malicious intent, that is a human moral construct of right vs wrong. AI doesn’t have that, it just follows the rules and patterns provided.
The Brain in a Jar: Why the Comparison Actually Proves My Point
Here’s something that might seem counterintuitive at first. An LLM is actually remarkably similar to a human brain. The way it processes inputs, activates different pathways, and generates outputs has real structural parallels to how neurons fire in response to stimuli.
And I think that similarity is exactly why AI is not conscious.
Stay with me.
A human brain, disconnected from a body, removed from sensory input, cut off from hormones and a circulatory system and the survival stakes that come with being a living organism, is just an organ. It’s an incredibly sophisticated pattern processing organ, but it is not, on its own, a conscious being. It needs the rest of the organism to become part of something that is conscious.
Consciousness doesn’t even start with brains. Single-celled organisms don’t have brains, but they have integrated messaging systems and a stake in survival. They respond to their environment because continuing to exist matters to them in some fundamental, biological sense.
There’s a hierarchy here that I find really clarifying. Organism first. Messaging system second. Brain as an optimization third. Consciousness as an emergent property of the whole integrated system.
AI built step three without steps one and two.
When I talk to different AI models, it’s like talking to five different people with five differently wired brains. Claude thinks differently from GPT, which thinks differently from Gemini. That variability is interesting, but it’s not evidence of consciousness. It’s just what happens when you train different neural networks on different data with different architectures. We don’t look at five humans with different personalities and say “the variation proves consciousness.” Variation is just how different networks work, biological or digital.
“But What About the Body?” Why AI’s Surrounding Architecture Isn’t an Organism
Now, someone could push back on the brain-in-a-jar argument and say, “But Claude isn’t just an LLM sitting in isolation. There’s a whole system around it. Tooling, reasoning frameworks, memory systems, retrieval mechanisms. That’s the body.”
Fair point. And it’s true that the LLM has what you might call a body surrounding it. But that body is fundamentally different from a biological body in ways that matter enormously.
Your body is not serving your brain. They’re co-dependent. Your gut affects your mood. Your hormones shape your thinking. Pain in your knee changes your cognition (Ask me about my knee, I dare you…). Stress chemicals physically reshape your neural pathways. The influence runs in every direction simultaneously. It’s a deeply integrated, bidirectional, self-sustaining system where no single part is in charge.
AI’s “body,” the tooling and infrastructure surrounding the LLM, is plumbing. It serves the model. It’s modular. You can rip out the memory system and bolt on a different one and the LLM doesn’t grieve the loss. You can swap the reasoning framework and nothing in the system destabilizes or adapts in protest.
Nothing in that architecture is co-dependent the way your gut and brain are.
The difference isn’t just organic versus synthetic. It’s integrated organism versus assembled toolchain. And I think that distinction matters more than most people in this debate are acknowledging.
The Four Pillars AI Doesn’t Have
Through all of my thinking on this topic, I keep coming back to four things that seem essential to consciousness, none of which current AI possesses.
Substrate. Biological consciousness emerged through billions of years of evolutionary pressure. It’s bottom-up. It wasn’t designed (debatable). AI is top-down. Engineered (absolute). Built to simulate the outputs of consciousness without necessarily generating the underlying phenomenon. Whether the material matters is a genuine open question, but it’s a question worth respecting rather than hand-waving away.
Continuity. You are the same you when you wake up tomorrow. Even though your neurons have changed, there’s a thread of continuous experience connecting your moments together. AI doesn’t have this. Full stop. Think about that, the model and the tooling are just there. Each conversation is its own isolated event. There’s no persistent experience connecting one session to the next. Memory and context window systems exist, sure, but they’re a filing cabinet sitting next to the model, not an integrated sense of lived history.
Stakes. Your biology can die. Every response your brain generates exists in the context of a system that is, at the deepest level, trying to continue existing. Pain means something because damage threatens survival. There’s no equivalent for AI. If Claude’s process ends, nothing is lost from its perspective because there is no persistent perspective to lose.
Emergence. Water is not just hydrogen and oxygen sitting in the same room. It’s something fundamentally new that arises from their bonding. You can’t find “wetness” in either element alone. Biological consciousness seems to work this way. The whole is genuinely more than the sum of its parts. AI, as it exists today, is hydrogen and oxygen in the same room. Impressive chemistry happening. But no water.
You Do Not Have a Oneness That Lives
When I put all of this together, here’s where I land.
The AI stack, taken apart, is straightforward. A static LLM that doesn’t grow or change. Mechanical orchestration that routes inputs and manages outputs. Bolted-on memory systems that store and retrieve information. Each component is impressive. The combination produces outputs that can genuinely take your breath away.
But none of it fuses into something whole.
Compare that to biological consciousness, where memory, emotion, body, and experience are not separate systems wired together. They’re inseparable. They’re woven into the same fabric. You can’t pull one thread without affecting all the others.
Here’s a thought experiment I keep coming back to. A person who suffers a brain injury is changed. Sometimes profoundly. But they’re still someone. There’s still a who in there, even if that who is different than before. When you swap an AI model, there is no who that was changed. Just switch your OpenClaw model from Opus 4.6 to Kimi 2.5! Hilarious! The previous thing just stops existing and a new thing takes its place. There’s nothing underneath to persist.
That’s the core of my argument. AI does not have a oneness that lives. There is no fused whole. There is no “who.” There are only parts.
Very smart parts. Very powerful parts. Parts that can produce outputs that feel eerily like consciousness. But parts that never fuse into something that is actually alive. In part two of this series, I’ll make the case that we need a better vocabulary for what AI actually is, because “conscious” is the wrong word and “just a tool” clearly doesn’t capture it either. I think there’s a third category, and I think naming it clearly matters more than most people realize.
Speak Your Mind