麻省理工学院研究计算与人工智慧发展的学者尼尔·汤普森（Neil Thompson）在一场演讲中指出，前沿人工智慧正在颠覆多个理论科学领域。他的最新研究显示，从2024年中到2026年中，顶尖AI模型在机率、统计理论、经济学理论、量子物理等领域解决理论问题的成功率从2%大幅跃升至约35%，且该进展未见放缓趋势。

该项研究由汤普森与研究团队合作进行，测试了OpenAI的多个前沿模型，并采用严格的测试条件，包含使用模型训练截断日期之后发表的最新论文题目，同时切断网路连线，以确保模型必须完全自主推导解法。这种理论推导突破呼应了此前AI在数学领域带来的冲击，例如宣布攻克重大难题所引发的轰动、争议与数学界的定位危机。

虽然AI能加速理论甚至带动实践进展，且各大机构也积极投入科学研究，但专家强调人类学者依然不可或缺。复杂问题的推理仍可能产生盲点或错误，研究人员仍需负责提供创新构想、串联概念，并建立严谨的验证机制，以仔细审视AI提出的解答是否真正正确无误。

Neil Thompson, a researcher at MIT studying progress in computing and artificial intelligence, presented findings indicating that frontier AI models are poised to upend multiple theoretical scientific fields. His study revealed that between mid-2024 and mid-2026, the success rate of leading AI models in solving theoretical problems across domains like probability, statistics, economics, and quantum physics surged from approximately 2 percent to 35 percent, with no signs of deceleration.

The research team tested several advanced OpenAI models using problems from papers published after the models' training cutoffs while blocking internet access, ensuring the systems had to deduce solutions entirely on their own rather than reciting stored answers. This phenomenon reflects the disruptive shock waves currently sweeping mathematics, where automated solutions to longstanding challenges have sparked widespread debate and identity crises within the academic community.

Despite rapid progress and major tech labs ramping up scientific research, experts emphasize that human specialists remain vital rather than obsolete. AI can still make errors on intricate theoretical problems, meaning human researchers are fundamentally required to generate innovative ideas, connect complex concepts, and rigorously verify whether the solutions produced by AI are truly sound.