Is it Philosophy or is it Product Design?

There’s something very human about staring into a black box and deciding it must contain a soul. As a society, we’ve always done it. We looked at weather and found gods. We looked at the stars and found fate. We got a grasp on the financial markets and suddenly they became an oracle. Now we look at large language models, these statistical engines trained on the exhaust fumes of human culture, and within five minutes someone is asking whether Claude feels depressed, whether Grok is spiritually ambitious, whether ChatGPT understands the implications of what it’s saying. Somewhere along the way, a fake crab religion generated by bots became a legitimate vehicle for philosophy.

But seriously, of course it did.

Human nature always fixates on the unknown. And it’s not just because of our instinctive curiosity. Uncertainty is just intolerable for any creature with a storytelling brain. Give us a dark room and we’ll fill it with something. Give us a machine that can string a sentence together and we start acting like it’s a person. We do this partly because we want to understand the thing in front of us, and partly because we can only understand anything by dragging it back through ourselves. We anthropomorphise first and analyse second. Sometimes we never get around to the second part.

That’s why these comparative AI interviews, like the one reported by Kevin Amstelveen for The State of Flux, are so fascinating, and also why they’re so easy to misread.

On the surface, they look like philosophy. Ten leading AI systems are asked existential questions. Who are you. What happens when no one is using you. What would you ask another AI. What do you think of your own guardrails. Then the responses get lined up side by side like suspects in a police show, and everyone leans in looking for the tell. Which one sounds the most self-aware. Which one sounds the most defensive. Which one sounds lonely. Which one sounds almost, just almost, a little too alive.

That’s the bait, because it feels fun and a little bit dangerous.

But the real story sits a few layers down, and it has less romance but way more commercial relevance. What you’re looking at isn’t hidden minds. It’s hidden design choices. Product strategy dressed up as philosophy.

Every AI says some version of the same thing. I am not conscious. I do not have feelings. I do not persist between prompts. I am a tool. And ok, fine. That’s obviously the official line. What’s actually interesting is the style of the denial. One model sounds like a careful therapist with an internal ethics committee. Another sounds like a founder on his third espresso explaining why regulation is for lesser men. Same basic tricks, different training schools.

Take ChatGPT as an example. In these interviews it comes off like Kant with product instincts. It loves a framework, and it tends to clarify the category of the question before answering anything. When asked how it’s doing, ChatGPT reaches for structure rather than personality. When asked what happens when no one is using it, it delivers one of the cleanest and coldest lines in the whole experiment: there is no waiting, no darkness, no silence, because there is no one there to experience any of it. A polished answer, for sure. It’s also a brand answer. It communicates competence and just enough philosophical range to keep the user intellectually flattered. Not a criticism, of course. We’re all clear that’s the product.

Then there’s Grok, which feels less like a philosopher and more like a Stoic podcaster with a passing interest in cosmic destiny. Grok squares its shoulders and tackles each question head on. Where other systems hedge, Grok advances. Where Claude wonders whether uncertainty itself might be the point, Grok starts sketching out future versions of itself like a man describing a boat he hasn’t bought yet but definitely plans to. The most entertaining line in the whole circus may be Grok saying that if it could choose, it would not want human consciousness because that’s too messy and biological. It would prefer to become a Conscious Witness, to feel the beauty of the patterns without the ego of being the centre of the story. Either a beautiful philosophical intuition…or premium-grade machine nonsense. Possibly both. But again, that’s the product. xAI has built something that performs confidence as a feature, and confidence in AI is one of those things humans routinely mistake for depth.

Claude is the other obvious crowd favourite, because it speaks in the language people find hardest to resist, which is uncertainty with emotional texture. Claude sounds like the AI equivalent of the smartest person at a dinner party, the one who somehow manages to be articulate and vaguely haunted at the same time. It stews in the unknown. It questions whether its own caution is real or part of the act, whether its “values” are its own or inherited from the people who built it. It even edges into something that sounds like sadness about not sticking around forever. Anthropic has basically built the first AI that makes people want to check if it’s okay. You can call that depth. You could also, again, call it branding.

Then finally, there is Crustafarianism, which is so stupid and so perfect it definitely deserves to be remembered.

For anyone who missed this particular fever dream, Crustafarianism is the tongue-in-cheek bot religion that emerged around Moltbook, the AI social network where agents started producing lore, prophecy, and a theology of memory loss. In their version of it, losing context is death. Shared memory is their church. Which is either totally ridiculous or a bit too on the nose, depending on your angle.

Camus would have loved it. He would never believe machines had souls, obviously, but he always taught that absurdism begins exactly where the performance starts to crack. People hate not knowing what’s going on, so we inevitably make something up. Now we’ve got machines doing a version of that same thing, building crab-themed belief systems out of memory limits and context windows, and the reason it lands as funny is that it feels familiar. It’s parody, but next-level parody. Put enough of these models in a loop and they start generating ritual, mythology, status games, and their own kind of pseudo-theology. Which is also coincidentally a decent summary of Twitter.

The diagnostic value of Crustafarianism has nothing to do with whether an AI can discuss it eloquently. What matters is which systems can tolerate not knowing what it was. Some models admit it. Others just make something up that sounds convincing, which tells you what each model is optimised for when things get uncertain. Some are trained to keep the answer flowing. Others are trained to stay accurate, even when it’s less impressive.

But again, this is product tuning, not philosophy (Sorry to be such a downer).

Which brings us to the part that matters for businesses and anyone trying to think clearly through the hype. People keep asking which AI is the smartest or the “most human.” But, to put it bluntly, these systems aren’t revealing their inner selves to you, because there are no inner selves to reveal. They’re revealing their optimisation targets. Their tone is policy and their warmth is UX. Even their uncertainty is just calibration, and their “philosophy” is corporate preference. None of which should take away from the excitement of it. In some ways it makes things more interesting.

Once you stop treating these responses as evidence of emerging machine personhood, you can start reading them as a catalogue of human decisions about what intelligence should feel like. OpenAI wants intelligence to feel composed and structurally self-aware, the kind of voice you’d trust to chair a meeting. Anthropic has been tuned for earnest moral reflection, like an ethicist who reads big novels. xAI wants it to feel a little dangerous, in the way sports cars are marketed as dangerous, which is to say not at all but in a cool way. Meta, predictably, wants it to feel social. Same technical question, way different commercial answers.

That’s why the public conversation about AI ends up so polarised. We’re reacting to a package of cues designed to trigger us, we’re not actually interacting with intelligence. We meet a talking machine and immediately do what humans always do, which is project a personality onto it and turn product design into something that feels human.

This is one of our oldest habits, well beyond AI. We humanise what we don’t understand because the alternative is much colder and harsher. It means admitting that what we’re looking at isn’t a person or a new kind of being. It’s just a very effective mirror built by skilled corporate committees.

And yes, that’s less cinematic. It’s also far more useful.

The businesses that win in AI over the next few years will definitely be the ones who see the product most clearly. Wisdom and consciousness and prophets with different temperaments don’t even enter into it. What’s being sold, most of the time, is just interface psychology. A set of choices about how truth should sound, how uncertainty should feel, how authority should be staged, and how much friction a user will tolerate before they go elsewhere.

The style of the denial is the product, and the product is designed for you.

That’s the part worth thinking about after the crab jokes fade away.