THE AI THREAT THAT’S NOT THERE, SORRY (the two hard problems of AI aren’t hard)

THE AI THREAT THAT’S NOT THERE, SORRY
(the two hard problems of AI aren’t hard)

1. An automated economy isn’t a problem. It’s necessary. Estimates are off. We will very likely end up halving the world population with a vast elderly population before 2100. The Japanese are correct in their strategy.

2. “Genius AI’ is nonsense. The hard problems we face in science aren’t computational they’re the cost and methods of conducting experiments. Karl Popper was wrong. Humans are extremely good at it. For example, why has fundamental physics gone wrong? It’s gone wrong because the Einstein-Bohr conflation of mathematical properties with existential properties, and failed to work on classical models given that the universe is consistent at all scales.

3. Ethical, Moral, AI isn’t a hard problem. Though, it requires different architecture, embodiment, a world model, and the ability to categorize human demonstrated interests (stuff etc) as inviolable without obtaining permission.

4. But, controlling the hardware and firmware so that humans can’t produce malicious AIs is hard.

5. Controlling humans that will try to create malicious AIs is really really hard.

6. We will need AI for defense and offense and winning that competition is more important than was the atomic bomb.

7. BUT… AI’s will police AI’s the same way accounting systems police people’s tendency to steal, and records police people’s tendency to lie.