Kenichi Sasagawa has spent years building N-Prolog, his own Prolog compiler. Version 5.16 just dropped, and he finally considers it near-complete. After years of development, he's answering the question he hears most: is Prolog worth learning?
His answer: yes, but not for the reasons you'd expect. Prolog won't help you land a job. Python and JavaScript win there. What Prolog offers is different. He calls it "executable logic." Most languages are procedural: do A, then B, then C. Prolog is relational. You describe what relationships must hold true, and the language figures out how to satisfy them.
Sasagawa traces this back to Jacques Herbrand, a logic pioneer working on mechanical reasoning a century ago. Lean Is Eating Other Proof Assistants Alive represents a similar focus on rigorous logical deduction. Concepts like Skolemization, which transforms existential quantifiers into universal ones, make automated reasoning tractable. Sasagawa implemented Skolem normal form in Prolog with AI assistance, and says that's when it clicked. Herbrand was essentially describing what would become Prolog decades later.
The timing matters now. Modern AI runs on statistics: neural networks, large language models, probabilistic outputs. Prolog sits at the opposite pole. It's logical, deterministic, and its reasoning can be inspected. Sasagawa sees it as essential for neuro-symbolic AI, which tries to merge neural approaches with symbolic reasoning. Logic programming has a role to play if we want AI systems that can explain their decisions, not just generate likely-looking answers.