Humans do engage with backwards reasoning. But they are also capable of checking that the constructed justification does then imply the conclusion.

ChatGPT apparently is not doing this on the basis of these examples.

Yep, I agree. I just think GPT is closer to human thinking than we appreciate, while still lacking critical components.

We can think logically and symbolically. But we mostly don't, and the parts where we don't are where most of the cognitive work happens.

Yeah, half the cryptics I solve is by looking at synonyms of the defition path and trying to backsolve the riddle part.