>A far more interesting question, in my view, is when will algorithms be able to do any productive task that a human can do, at a competent level.

That's what I would personally define AGI as. Any task a human can do, at least intellectually but possibly physically as well but to me that's more robotics than intelligence. It may require a physical body to achieve true AGI.

I agree with your statement about consciousness, that's why I excluded it from the definition.

But, I somewhat disagree about GATO. But only slightly, and that's more to the point of my post. I don't know exactly what to define Proto-AGI as and how many general tasks it must be equal or greater to a human at. But I'd definitely define full AGI as capable in all human intellectual tasks at a level equal or greater to humans.

So GATO might be Proto-AGI today by definition. It's general, it's definitely not narrow. But I'm trying to say 2023 will be when we get a general AI that is able to meet or surpass human ability across most/many intellectual tasks. I think memory and reinforcement learning will be the key to achieving something that's basically AGI next year, but we'll probably move goal posts as it gets closer.