I want to point out something about part of the discussion in the article about AGI, specifically that Gary Marcus quote. He said that "if a person made these e

I want to point out something about part of the discussion in the article about AGI, specifically that Gary Marcus quote. He said that "if a person made these errors, you'd think there was something very wrong with them," and therefore we're nowhere near AGI. I disagree, not with his premise, but with his conclusion; and I think it's a distinction of definition. To me, "human-level intelligence" doesn't mean "human-like behavior" at all. I think to measure intelligence is simply to measure the degree of generalization and synthesizing of data into new information. If an AI can generalize and synthesize info as well as a person, I'd call it human-level intelligence, even if it thinks very differently from humans, and therefore its mistakes are much "weirder" and "dumber" than the average human mistake. I think it's a bit of a biased fallacy to think intelligence should be measured by how human a system's behavior is in every domain.