Now, "AI" is just a pile of blackboxes. Is this progress, you tell me.
They are not two disjointed and separate things, they are just two different approaches that can be very well combined with each other
And it is not at all bad to use neural tensor calculation, for the "may be" of the inference engine.
Considering it was a 80s computer, I have never seen a prolog interpreter implemented in the firmware, and among other things with so few RAM and CPU resources
Today's tensor calculation is possible because we have NPUs and multicore machines, however we still don't have enough shared and dedicated RAM, per NPU, to allow the calculation of second derivatives (Taylor), and we have to be satisfied only with the first derivatives of tensors ... which means very approximate results.
2027-2030, we'll have that too!