I do think humans are easy to fool but so far only in extremely constrained scenarios. E.g., limited to chat bots. But that’s not what I’m talking about here. To fool people in the manner that’s being discussed, robuts would need to fool people with buddy language and on a subconscious level. So much more of how we communicate is beyond reading words on a screen. Consider the uncanny valley that is still prevalent with deep fakes. Getting better for sure, but that’s still constrained to premise video. To be able to do that in person in real time is a much larger problem.

The other examples appear to me as magnifications of mental illness instead of generalizable scenarios, possibly including the Google employee when you consider his background.

Hm. To me, reaching uncanny valley means they're probably quite close in form and behaviour. But it may well be the case that the remaining 1% (or whatever the percentage is) of the details may be harder than the first 99%.

I can see your point. But to me, the uncanny valley isn’t evidence that we’re super close, it’s evidence that we don’t know how to fully define the problem because there are elements we know are not wore right but can’t say why. Because we can’t fully define it means we aren’t close to solving it.

I certainly agree that we don't know what we're missing. My best guess timeline for this is very vague for this exact reason.