‘World Models,’ an Old Idea in AI, Mount a Comeback

World models, a concept where AI systems build internal representations of their environment to evaluate predictions before real-world application, are making a comeback in AI research. Leading experts like LeCun, Hassabis, and Bengio consider them essential for truly intelligent AI systems, though researchers disagree on implementation details. While current AI uses disconnected "bags of heuristics" rather than coherent models, robust world models could reduce hallucinations, enable reliable reasoning, and improve AI interpretability.