别追AGI了！LeCun团队：人类本来就不通用

AGI是个伪命题

从Altman到Hinton，所有人都在谈论AGI。但Yann LeCun和新论文合作者Judah Goldfeder、Philippe Wyder、Ravid Shwartz-Ziv抛出一个尖锐问题：人类真的”通用”吗？

答案是否定的。

论文直指核心矛盾：AGI被定义为”能做人类能做的一切”，但这个定义本身就有问题——

人类智能根本不是通用的，它只是恰好适合我们生存的高度专业化工具。

Moravec悖论：你的直觉是错的

什么对人类”简单”？走路、辨认面孔、常识推理。 什么对人类”困难”？下棋、解微积分、记忆海量数据。

但AI的表现恰恰相反。1990年代Moravec就指出：我们认为容易的事，进化花了几亿年优化；我们认为难的事，其实计算上很简单。

这不是巧合，而是证据——证明人类只是在自己的生态位里高度特化，而非什么”通用智能”的典范。

SAI：放弃通用，追求超人类

论文提出Superhuman Adaptable Intelligence (SAI)：

旧思维 (AGI)	新思维 (SAI)
模仿人类	超越人类
什么都会	专精关键领域
与人类比较	填补人类盲区

SAI的核心是适应性+超人类性能：能快速学习任何重要任务，并在该任务上达到人类无法企及的水平。

为什么这很重要

当前的AI评估标准是扭曲的：“能不能像人一样做数学题？”

但SAI问的是：“能不能解决全世界最顶尖的数学家也无法解决的问题？”

当AI明确是”超级专才”而非”通用替代者”时：

• 研究方向更清晰（不再追求大而全的模型）
• 社会讨论更务实（工具 vs 竞争者）
• 进展可衡量（明确的性能基准）

LeCun等人的结论很直接：语义很重要。如果整个领域都在追求一个定义模糊、理论上不可能的目标，那就是在浪费资源。

放弃AGI的幻觉，拥抱SAI的现实。

AI的未来不是第二个”人类”，而是无数超越人类的”超级专家”。

The AGI Myth

From Sam Altman to Geoffrey Hinton, everyone is talking about AGI. But Yann LeCun and co-authors Judah Goldfeder, Philippe Wyder, and Ravid Shwartz-Ziv pose a sharp question: Are humans truly “general”?

The answer is no.

The paper cuts to the core contradiction: AGI is defined as “AI that can do everything a human can do.” But this definition itself is problematic—

Human intelligence is not general at all; it is merely a highly specialized tool finely tuned for our survival.

Moravec’s Paradox

What’s “easy” for humans? Walking, recognizing faces, common-sense reasoning. What’s “hard” for humans? Playing chess, solving calculus, memorizing vast amounts of data.

Yet AI performs exactly the opposite. Moravec pointed out in the 1990s: The things we find easy took evolution hundreds of millions of years to optimize; the things we find hard are computationally trivial.

This is not a coincidence—it’s evidence that humans are highly specialized to their ecological niche, not a paradigm of “general intelligence.”

SAI: Abandon Generality

The paper introduces Superhuman Adaptable Intelligence (SAI):

AGI Mindset	SAI Mindset
Mimic humans	Surpass humans
Jack of all trades	Master of key domains
Compete with humans	Fill human capability gaps

SAI’s core is adaptability + superhuman performance: the ability to quickly learn any important task and achieve levels beyond human capability.

Why This Matters

Current AI evaluation criteria are distorted: “Can it do math like a human?”

But SAI asks: “Can it solve problems that the world’s best mathematicians cannot?”

When AI is clearly a “super-specialist” rather than a “general replacement”:

• Research directions become clearer
• Societal discussions become more pragmatic
• Progress becomes measurable

Abandon the illusion of AGI. Embrace the reality of SAI.

---

Paper Information

Title: AI Must Embrace Specialization via Superhuman Adaptable Intelligence
Authors: Judah Goldfeder, Philippe Wyder, Yann LeCun, Ravid Shwartz-Ziv
arXiv: https://arxiv.org/abs/2602.23643

Abstract: Everyone from AI executives and researchers to doomsayers, politicians, and activists is talking about Artificial General Intelligence (AGI). Yet, they often don’t seem to agree on its exact definition. One common definition of AGI is an AI that can do everything a human can do, but are humans truly general? In this paper, we address what’s wrong with our conception of AGI, and why, even in its most coherent formulation, it is a flawed concept to describe the future of AI.

Goldfeder, J., Wyder, P., LeCun, Y., & Shwartz-Ziv, R. (2026). arXiv preprint arXiv:2602.23643.