> We should evaluate AGI the same way we evaluate human general intelligence, which also manifests with dizzying variety. This is key. All humans don’t underst

> We should evaluate AGI the same way we evaluate human general intelligence, which also manifests with dizzying variety. This is key. All humans don’t understand everything—and the way we assess human conceptual understanding is through tests. This implies that we need a lot more tests for AI. I’d like to see a big investment in “Machine Psychology.”