Lighthill 报告重创英国 AI

1973 年，英国应用数学家 James Lighthill 受科学研究委员会委托，评估英国 AI 研究的状况。他交出的报告毫不客气：除了极少数专门领域，AI 研究的进展远未达到承诺；至于「通用人工智能」，他认为遥遥无期。这份报告很快成为英国政府削减 AI 经费的依据，剑桥、爱丁堡等研究重镇规模收缩，英国 AI 进入漫长的十年低迷。

Lighthill 报告不是凭空而来的。到 1970 年代初，AI 界的几次豪赌都已经显出疲态：机器翻译的「大爆炸式发展」承诺落空，通用问题求解器表现平平，感知机的局限刚被系统论证。当一个领域长期靠「未来会更厉害」维持投入，外部审视迟早会来，Lighthill 只是那个把事实写下来的人。

报告的分量在于它的身份——不是外行人的嘲讽，而是权威应用数学家的冷静评估。它给出的结论「通用智能不可行」虽然过于悲观（后来被证明是错的），但在当时极具说服力。随之而来的经费收缩像多米诺骨牌一样倒向整个英国 AI 生态。

几乎同时，大洋彼岸的美国也在收缩。DARPA 在 1974 年前后削减了多数 AI 项目经费，两个事件合在一起，构成了 AI 史上第一次真正意义上的寒冬。研究方向的钟摆从「通用智能的狂热」荡回「小而具体的应用」，专家系统等务实路线反而在寒潮中活了下来。

回看 Lighthill 报告，它的教训至今适用：AI 的兴衰往往不是技术的失败，而是承诺与现实的落差被周期性地清算。每一次热潮都会引来一份「Lighthill 报告」式的审视，而真正能穿越寒冬的，永远是那些少说大话、把能力做实的路线。

In 1973 British applied mathematician James Lighthill was commissioned by the Science Research Council to assess the state of UK AI research. His report was blunt: outside a few narrow domains, progress had fallen far short of promises; as for general artificial intelligence, he considered it out of reach. The report quickly became the basis for government funding cuts. Teams at Cambridge, Edinburgh, and other hubs shrank, and British AI entered a decade-long slump.

The Lighthill report did not come from nowhere. By the early 1970s several of AI's big bets had gone soft: the promised "general advance" in machine translation had failed to materialize, general problem solvers were underwhelming, and the perceptron's limits had just been systematically documented. When a field sustains itself on "the future will be better", external scrutiny eventually arrives—Lighthill was simply the one who wrote the facts down.

What gave the report its weight was its author's standing. This was not an outsider's mockery but a sober assessment by a respected applied mathematician. His conclusion that general intelligence was infeasible was too pessimistic—later proven wrong—but at the time it carried enormous conviction. The funding cuts rippled through the entire UK AI ecosystem.

Almost simultaneously, the United States was contracting too. DARPA slashed most AI project funding around 1974, and the two events together produced the first true AI winter. The pendulum swung from "grandiose general intelligence" back to "small, concrete applications", and pragmatic lines like expert systems survived the cold.

Looking back at the Lighthill report, its lesson still applies: AI's booms and busts are rarely about technological failure but about the gap between promises and reality being periodically settled. Every hype cycle summons its own "Lighthill report", and what survives the winter is always the route that promises less and delivers more.