Prolog was never good at the things they thought it would be at, like AI, which is better done by ML today, specifically often like you said, with NNs. But it turned out to be good for other things, and those use cases are still alive today, even though there are many competitors. Look at Tiobe index, Prolog's usage is constant just under 1 percent, and has been for decades. So it's good for something.