Harari's Fixes — And Which Ones Might Work

My last post was about Harari's warnings. This one is about his solutions — because he does offer some, and again I asked Claude what it thinks of them.

Harari's protection plan has three layers.

Legal bans. No legal personhood for AI — it can never own money, companies, or property. And no pretending: a bot must always say it's a bot. He calls AIs posing as people "counterfeit humans," and wants them treated like counterfeit money.

Slow down and rebalance. Give society time to digest the change. And fix the absurd spending ratio: about 100x more goes into making AI powerful than into making it safe. In pharma, a company spending 1% on safety testing would be shut down.

Strong institutions. His deepest point. Protection is not a technical trick — it's courts, free press, elections, many independent centers of power. His Tiberius story cuts both ways: one strongman is easy to capture, a distributed system is hard.

Claude's verdict was mixed, and I mostly share it.

The bans are the strong part. "Bots must say they are bots" is a clear, enforceable rule — like labels on food. The personhood ban closes a dangerous door while it's still cheap to close. (Claude added, fairly: "I can't be fully neutral on that one.")

"Slow down" is the weak part. Nice words, no mechanism. Who stops first? Nobody wants to be first. The spending ratio is the rescueable piece — 100:1 is measurable, and measurable things can become regulation.

The institutional bet is where it gets interesting. Harari calls "fix the AI itself" naive, but treats "keep institutions strong" as achievable. Are courts and free press really in better shape than alignment research? Both are hard. Claude's conclusion, which I'll steal: we need both — safe engineering AND strong institutions. Betting on only one is a bet nobody can afford to lose.

One thing Harari says that needs no debate: the time to decide is now, while humans still write the rules.