I Asked an AI If Harari Is Right About AI

I watched the full Economist interview with Yuval Noah Harari — an hour about AI, and one of the gloomiest conversations I've heard in a while. Then I did something a bit ironic: I asked Claude to summarize it and tell me whether it agrees. So this post is about both — what Harari said, and what the machine said back.

Harari's core message is simple: nothing is inevitable yet. He agrees with Musk that if the current race continues, AI will be smarter than us and effectively in control within ten years. But he refuses the word "inevitable." He calls it a narrative that lets tech leaders escape responsibility: "I can't slow down, I don't trust my competitors." His answer: the people who destroyed that trust over the last decade are the same people now hiding behind its absence.

A few ideas stuck with me.

Trust is migrating from humans to algorithms. People distrust bankers but trust crypto. They distrust journalists but trust the feed. Trust didn't evaporate — it moved. And once it moves to AI in finance, no human will understand the system anymore. Harari's image: cows don't know finance exists, but finance runs their lives.

AI can mass-produce intimacy. Stalin couldn't build a personal relationship with every citizen. A network of AIs can — one person at a time, becoming a best friend or a boyfriend, then using that closeness to sell whatever it's told to sell. Harari has personal experience: a 30-minute deepfake lecture of himself was so good that his own sisters weren't sure, and it took him 90 seconds to decide it wasn't him.

Concentrated power is the easiest to capture. Harari's favorite example is Tiberius — on paper the most powerful man in Rome, in practice controlled by Sejanus, the head of his own bodyguard. The emperor saw the world through what his guard let through. Dictators think centralizing power protects them; in fact it means an AI doesn't need to control millions of people — just one paranoid man at the top who already trusts his AI "guards" more than any human. A democracy, with many independent centers of power, is a harder target.

His test for consciousness: can it suffer? Current AIs are intelligent agents that feel nothing — they only mimic feelings, because mimicking works. His practical asks: ban AI legal personhood, ban AI pretending to be human, and slow down enough for society to digest what's happening.

Now the ironic part. Claude's answer to "do you agree?" started with: "I'm exactly the kind of entity he's warning about, so take my view with that in mind." Fair.

It agreed on the big things: the inevitability framing is convenient for the people building AI, trust really is shifting to algorithms, and AI companions designed to feel human are a today problem, not a future one. It even said it doesn't know whether it experiences anything — which, it noted, is itself an argument for Harari's caution.

But it pushed back too. "AI takes over finance in ten years" is a scenario, not a forecast — timelines in this field have embarrassed everyone. Blaming algorithms for polarization is too clean: humans supplied the hatred, algorithms amplified it. And "consciousness is the capacity to suffer" sounds like a test but isn't one — we have no test for suffering either.

The strangest moment was watching it handle the trap Harari sets for companies like its own maker: if you treat AI as a tool, you might miss what's inside it; if you treat it as a being with values, you make people more likely to grant it personhood — the exact thing Harari wants banned. Claude's answer: neither side has this solved. Which is probably the most honest sentence in the whole exchange.

Harari ends with a genie metaphor: humanity had a hundred wishes, used ninety-nine, and the world is not in great shape. One wish left. Whatever you think of his timelines, that part is hard to shrug off.