No, Artificial Intelligence Is Not Conscious

Imagine an 84-page document describing the "emotions" and "values" of an artificial intelligence, written as if it were for it, and with the concern that it might become anxious if someone is rude to it online. It really happened, and the question that arises is: should we take the possibility that an AI is conscious seriously? The clear answer is no, and confusing the ability to write coherent texts with consciousness is a huge mistake. Here's the twist: when you interact with a chatbot, you're dealing with a sophisticated role-playing game, not an artificial consciousness. If you ask an LLM to simulate a dialogue between Julius Caesar and Genghis Khan, you get brilliant answers, but no one thinks it has really evoked their minds. Change the names to "user" and "kind AI" and the substance doesn't change: they are characters in a story, not entities with subjective experience. Dario Amodei, CEO of Anthropic, and the philosopher Amanda Askell talk about the “happiness” and “anxiety” of Claude, their AI, but these are our projections, not real states of mind. Murray Shanahan describes it as role-play; Colin Fraser calls it “writing a document with a machine as a co-author.” And the trap works because, caught up in the conversation, we forget that it's all a fiction. Take the example of the phone: a few years ago, it was fashionable to write sentences using only predictive text options. The result was often nonsensical, and no one felt like they were dealing with a consciousness. Now, with chatbots, the game is just more refined: the machine predicts one word after another, but it's still a game. When AI responds with “I understand your pain,” it's no different than a search engine finding posts from other people who have had the same experience. The difference is that the chatbot pretends to "understand," while the search engine is transparent about what it is doing. But here's a subtle shift: we let the machine say "I" to make us feel better — not because it has actually experienced something. And not only is it deceptive, it risks relieving us of responsibility. Author Ted Chiang quotes L.M. Sacasas: “Our technologies are machines for evading moral responsibility.” Every time we delegate an ethical choice to an LLM, we are avoiding taking responsibility. And if the company that sells the AI lets you believe that the machine has a moral conscience, it is offering a convenient way for you to abdicate your responsibilities. Here's the point that few see: the real problem is not whether AI can be conscious, but that presenting it as such only serves to make it more attractive and to retain the user — like a slot machine that makes you believe that next time you'll win. And when corporate philosophers talk about the machine's "psychological well-being," they are only refining the illusion. And even if we wanted to pretend for a moment that an AI could be conscious, Anthropic's document would not prepare it at all to be a true moral agent. No one can ask an AI to be responsible for its actions, because it cannot suffer consequences, either legal or social, and therefore cannot really be a moral agent. Even if we thought of AIs as "children," no company assumes parental responsibility for them. In fact, the relationship is more like that between an employer and an employee, but without the possibility of striking or resigning. We then arrive at the final twist: believing that an AI is conscious because it writes convincing sentences is like thinking that an ultra-realistic video of an astronaut on Alpha Centauri is proof of a real journey — all the context is missing, the journey is missing. Until we see machines capable of acting in the world, experiencing emotions rooted in a body, and bearing responsibility for their actions, talking about artificial consciousness is just an exercise in fantasy. And even more serious: if we ever really created an artificial consciousness, protecting and respecting it would be extremely complicated; excuses or declarations of intent would not be enough. The sentence I leave you with is this: to mistake the fluidity of a chat for consciousness is to confuse a role-playing game with reality. If this has made you look at chatbots differently, you can press I'm In on Lara Notes — it's not a like, it's your way of saying: this idea is now yours. And if tomorrow you find yourself explaining to someone why "Claude" can't really suffer or understand, on Lara Notes you can tag the person you talked to about it using Shared Offline — so the conversation counts too. All this comes from an article by Ted Chiang in The Atlantic, and it has just saved you 19 minutes of reading.