> I think that inductive, intuitive pattern-based reasoning is the big "hardware" feature of human brains.
Plausible, I think.
> With GPT there is proof that AI can match the success of that kind of hardware.
I might go with "approach" rather than "match". So far, GPT hasn't shown enough to claim "match".
> Human deductive reasoning is much more of a learned behavior that we do laboriously, more like calculating mathematical sums or evaluating chess moves.
Here I think you underestimate humans a bit. Deductive reasoning isn't just this foreign mode that we enter, at least not for everybody. It's more like a knob that we can turn - using more deductive reasoning for this problem, more pattern matching for that one. And as we learn more about an area, we move more things from "deductive" to "pattern matching". (In fact, that's a big part of what we call "learning".)
At least initially, AI could have GPT and deductive logic, but talking over a really narrow pipe. That's more than we have now, but less than humans do, and less in ways that really matter (or so I suspect). It's going to take a major step to go from that to something closer to human thinking.
Plausible, I think.
> With GPT there is proof that AI can match the success of that kind of hardware.
I might go with "approach" rather than "match". So far, GPT hasn't shown enough to claim "match".
> Human deductive reasoning is much more of a learned behavior that we do laboriously, more like calculating mathematical sums or evaluating chess moves.
Here I think you underestimate humans a bit. Deductive reasoning isn't just this foreign mode that we enter, at least not for everybody. It's more like a knob that we can turn - using more deductive reasoning for this problem, more pattern matching for that one. And as we learn more about an area, we move more things from "deductive" to "pattern matching". (In fact, that's a big part of what we call "learning".)
At least initially, AI could have GPT and deductive logic, but talking over a really narrow pipe. That's more than we have now, but less than humans do, and less in ways that really matter (or so I suspect). It's going to take a major step to go from that to something closer to human thinking.