The paper solves a famous open problem in mathematical logic that had been unsettled for decades. Medvedev's logic, proposed in the 1960s, is a formal system for reasoning about computational problems and their relative difficulty. The central question was whether there exists an algorithm that can determine, in finite time, whether any given logical statement is provable in this system. The answer, established here, is no: the logic is "undecidable," meaning no such algorithm can exist.
The proof works by connecting Medvedev's logic to a completely different mathematical puzzle called the tiling problem. A tiling problem asks whether a given set of tile shapes can be arranged to cover an infinite flat plane without gaps or overlaps. Some versions of this puzzle are known to be computationally unsolvable, and the authors build a careful translation showing that solving Medvedev's logic would require solving tiling problems first. Since the tiling problems are unsolvable, the logic must be too. A related system called Skvortsov's logic, which deals with infinite rather than finite problems, is shown to be undecidable by the same strategy using a slightly different variant of the tiling problem. As a bonus, the argument also proves that these two logics are genuinely different from each other, something not previously established.
Perhaps the most striking aspect of the paper is how the proof was found. The core ideas and technical arguments were generated by an AI system called ChatGPT Sol 5.6, and then the resulting proof was formally verified for correctness by another AI, Claude Opus 5, using a computer proof-checking language called Lean. The paper includes a detailed account of how human researchers directed and collaborated with these AI tools to produce a verified mathematical result, making it a notable example of AI-assisted mathematical discovery at the research frontier.