Kevin Kelly：AI是五十年的一夜成名，智能如同早期电力一样神秘Kevin Kelly: AI Is a 50-Year Overnight Success, Intelligence as Mysterious as Early Electricity
The Takeaway：我们对智能的无知程度，堪比牛顿时代对电力的无知，而智能更可能是化合物而非元素。
Wired联合创始人、长期未来学家Kevin Kelly指出，AI是典型的“五十年一夜成名”。他早在1987年就见过Jaron Lanier的VR演示，当时设备昂贵到数百万美元，如今虽便宜到100美元，但体验进步远没有数量级那么大，VR仍在等待自己的LLM时刻。他强调，预测容易，做出正确预测却很难：他曾对VR过度乐观，对eBay和比特币早期用途判断失误。
Kelly认为我们现在对智能的理解，就像早期科学家对电力的理解一样浅薄。牛顿对电力的想法是错的，当时有无数理论（包括生物现象说）。他怀疑智能不是单一元素，而是由多种认知元素组成的“化合物”，就像盐不是元素而是化合物。Marvin Minsky的“心智社会”和当今mixture of experts已在印证这一点。人类智能只是可能性空间中的边缘特例，未来AI将填充出成百上千种不同思维形式。
他分享了用o1 Pro做“观众只有一人”的创作实验：让AI想象达芬奇、马丁·路德与哥伦布在雪夜酒店对话，进而协作建城、写小说、做封面与营销，最终得到十部小说的史诗。乐趣在于共同创造过程本身，而非分享。他预测绝大多数生成式内容（每天数千万张图）最终都会是“观众只有一人”的自我表达与娱乐，更接近周日画画或写日记，而非职业。
Kelly还提醒，机器人落地会比预期慢得多，因为物理与生物约束（能耗、重量、眼睛聚焦）远比纯软件难解决。他主张同时保持前沿与历史视角：读AI的同时读电力发明史与Long Now的万年尺度。
“AI is a fifty year overnight success.”
Wired联合创始人、长期未来学家Kevin Kelly指出，AI是典型的“五十年一夜成名”。他早在1987年就见过Jaron Lanier的VR演示，当时设备昂贵到数百万美元，如今虽便宜到100美元，但体验进步远没有数量级那么大，VR仍在等待自己的LLM时刻。他强调，预测容易，做出正确预测却很难：他曾对VR过度乐观，对eBay和比特币早期用途判断失误。
Kelly认为我们现在对智能的理解，就像早期科学家对电力的理解一样浅薄。牛顿对电力的想法是错的，当时有无数理论（包括生物现象说）。他怀疑智能不是单一元素，而是由多种认知元素组成的“化合物”，就像盐不是元素而是化合物。Marvin Minsky的“心智社会”和当今mixture of experts已在印证这一点。人类智能只是可能性空间中的边缘特例，未来AI将填充出成百上千种不同思维形式。
他分享了用o1 Pro做“观众只有一人”的创作实验：让AI想象达芬奇、马丁·路德与哥伦布在雪夜酒店对话，进而协作建城、写小说、做封面与营销，最终得到十部小说的史诗。乐趣在于共同创造过程本身，而非分享。他预测绝大多数生成式内容（每天数千万张图）最终都会是“观众只有一人”的自我表达与娱乐，更接近周日画画或写日记，而非职业。
Kelly还提醒，机器人落地会比预期慢得多，因为物理与生物约束（能耗、重量、眼睛聚焦）远比纯软件难解决。他主张同时保持前沿与历史视角：读AI的同时读电力发明史与Long Now的万年尺度。
“AI is a fifty year overnight success.”
The Takeaway: We are as ignorant of intelligence as Newton-era scientists were of electricity, and intelligence is more likely a compound than an element.
Wired co-founder and long-time futurist Kevin Kelly calls AI a classic “50-year overnight success.” He first saw Jaron Lanier’s VR in 1987 when the gear cost millions; today it costs $100, yet the experience has not improved by orders of magnitude. VR is still waiting for its LLM moment. Predictions are easy; true ones are hard. He was too bullish on VR and too dismissive of eBay and early Bitcoin use cases.
Kelly argues our understanding of intelligence today resembles early theories of electricity. Newton was simply wrong about it; countless theories (including biological ones) competed. He suspects intelligence is not a single element but a compound of cognitive elements, the way salt is a compound. Marvin Minsky’s “society of mind” and today’s mixture-of-experts architectures already point this way. Human intelligence is an edge case in a high-dimensional space of possible minds; future AIs will populate that space with many different forms of thinking.
He describes an “audience of one” creative experiment with o1 Pro: imagining da Vinci, Martin Luther and Columbus stuck in a snowy hotel, then collaborating on a science-and-freedom city in the New World, generating novels, covers and marketing materials. The joy was in the co-creation itself. He predicts the vast majority of generative output (tens of millions of images daily) will ultimately serve an audience of one—closer to Sunday painting or journaling than to a career.
Robots will arrive far slower than expected because of physical and biological constraints (power, weight, ocular focus). Kelly balances frontier time with historical reading and Long Now’s 10,000-year view.
“AI is a fifty year overnight success.”
查看原文 →
Wired co-founder and long-time futurist Kevin Kelly calls AI a classic “50-year overnight success.” He first saw Jaron Lanier’s VR in 1987 when the gear cost millions; today it costs $100, yet the experience has not improved by orders of magnitude. VR is still waiting for its LLM moment. Predictions are easy; true ones are hard. He was too bullish on VR and too dismissive of eBay and early Bitcoin use cases.
Kelly argues our understanding of intelligence today resembles early theories of electricity. Newton was simply wrong about it; countless theories (including biological ones) competed. He suspects intelligence is not a single element but a compound of cognitive elements, the way salt is a compound. Marvin Minsky’s “society of mind” and today’s mixture-of-experts architectures already point this way. Human intelligence is an edge case in a high-dimensional space of possible minds; future AIs will populate that space with many different forms of thinking.
He describes an “audience of one” creative experiment with o1 Pro: imagining da Vinci, Martin Luther and Columbus stuck in a snowy hotel, then collaborating on a science-and-freedom city in the New World, generating novels, covers and marketing materials. The joy was in the co-creation itself. He predicts the vast majority of generative output (tens of millions of images daily) will ultimately serve an audience of one—closer to Sunday painting or journaling than to a career.
Robots will arrive far slower than expected because of physical and biological constraints (power, weight, ocular focus). Kelly balances frontier time with historical reading and Long Now’s 10,000-year view.
“AI is a fifty year overnight success.”