AI for science needs reasoning, not just data

AlphaFold demonstrated the scientific power of AI trained on exceptionally large, reliable datasets, but few fields can produce comparable data. AI agents may offer a broader path by reasoning with multiple tools, recording their work, preserving scientific knowledge, and accelerating experiments despite their current limitations.

AlphaFold展示了利用超大规模可靠数据训练的人工智能所具备的科研能力，但很少有领域能够生成类似的数据。人工智能智能体或许能提供一条适用范围更广的道路：它们可以结合多种工具进行推理、记录工作过程、保存科研知识，并在仍有局限的情况下加快实验进程。