AI 会受困于人类数据瓶颈吗

人工智能是否会被人类掌握的数据所禁锢

近期，Geoffrey Hinton 在某段视频中阐述了一项见解。

这确是他的原话

"If you look around for what AI systems don't have any problems with data limits, one example is AlphaGo, Alpha0, where they're playing games."

"They play against themselves… they can generate infinite amounts of data. They don't need any external data put in."

"Mathematics is kind of a closed system. You can make conjectures, see if you can prove them… with no external input of data."

"Those kinds of systems can run all the time and get smarter and smarter. And I believe that's going to happen with large language models eventually."

其核心含义在于

若观察哪些 AI 体系不受数据上限掣肘，AlphaGo 与 AlphaZero 便是典范——它们专精于博弈。通过自我对弈……能创造无穷无尽的数据，无需外部注入。数学在某种程度上属于封闭系统。人们可提出假设，尝试论证……无须外界数据投喂。此类系统能够持续运转，日益聪慧。我坚信大语言模型终也将步入此境。

最终结论是 AI 不会被人类数据彻底锁死

的确，这是一种 akin 于数学领域的发现、实验与验证逻辑，构成了闭环发展的必然趋势。

不过，就当下而言，绝大多数 AI 仍须倚仗人类数据，毕竟 AI 旨在服务人类，此环节无法省略亦难以跨越。正如企业推进数智化转型时，若略过数字化基石，人工智能便无法真正落地生根。

因此，我们应致力让 AI 持续运用人类数据，从而更好地服务于碳基生命。