AI又要颠覆数学？陶哲轩紧急发声：停止造神

Core Viewpoint - The article discusses the exaggerated claims regarding AI's ability to solve complex mathematical problems, particularly in relation to Erdős problems, and emphasizes the need for a more nuanced understanding of AI's contributions in mathematics [1][2]. Group 1: AI's Capabilities in Mathematics - AI's achievements in solving certain mathematical problems are often overstated, leading to misconceptions that AI can independently innovate or replace human mathematicians [2][4]. - The difficulty level of problems solved by AI varies significantly, making direct comparisons misleading; some problems are much easier than others, which can skew perceptions of AI's capabilities [2][3]. - Many problems labeled as "unsolved" may have been previously addressed in literature, leading to potential misattributions of "first solutions" to AI [3][10]. Group 2: Evaluation of AI Contributions - AI's contributions can be categorized into several types, including generating complete or partial solutions, conducting literature reviews, and formalizing proofs [6][12]. - Specific examples illustrate that AI has successfully provided solutions for certain problems, but these often require validation against existing literature to confirm their novelty [8][10]. - The process of formalizing AI-generated proofs can introduce risks, such as the potential for misinterpretation or the introduction of unverified axioms [4][12]. Group 3: The Role of Human Mathematicians - Human mathematicians remain essential for formulating deep questions, creating new concepts, and integrating results into the broader knowledge network of mathematics [12]. - The future of mathematics may involve a collaborative relationship where humans guide AI in exploring mathematical landscapes, rather than AI acting as an independent entity [12].