It’s possible. Even the researchers at OpenAI are, as far as I know, still divided on the question whether true artificial intelligence can be arrived at purely by analyzing language alone.

At the same time, who would have thought that the current results would be possible with a language - only approach? One can even see similarities between GPT-3 and a child. namely, when you ask a very long question, the model sometimes seems to “lose focus“, so you sometimes have to repeat your initial question again. That’s pretty much the same thing that sometimes happens when you talk to a coworker :-)

Also, one could argue does a Tesla FSD computer does not learn driving as a human would. Because the computer itself he’s not afraid of getting crushed in an accident, for instance. and still, it learns to drive!

The jury is still out. But it is very much possible that we will be able to arrive at a super intelligence more capable and human intelligence by merely letting the computer learn from written language. and I think this would be as close as you could get to a proof of Skinners ideas.

All the capabilities that GPT-3 has today are the result of applying what Skinner wrote in “Verbal Behavior“. I have yet to see a technology based on the ideas of Chomsky that is even just remotely as impressive or useful.