AI Has No Libido — That's the Real Problem
AI 没有 libido——这才是真正的问题所在
Why bodily drive changes creativity — and what current AI does not yet supply.
身体驱力如何影响创造,以及当前 AI 尚未提供什么。
We've been debating AI's "creativity" — whether it will take over artists' and writers' jobs, whether AI-generated work counts as real creation.
But there's a more fundamental question nobody's asking: what drives AI?
At the base of human creativity is something AI simply doesn't have — libido.
Not in the Freudian sense of sexual drive. Something broader: pre-linguistic, pre-choice, vital impulse. That internal surge that makes you unable not to do something, for reasons you can't quite articulate.
What Clubbing Is Actually Doing
Start with clubbing. Why does it sometimes help you think through things? This sounds absurd, but there's a structural reason worth taking seriously.
For some people, the clubbing configuration — repeated beats, bodily movement, and reduced visual input — may loosen deliberate monitoring and shift attention toward bodily sensation. Rhythm can reduce the need to plan each movement, while closing the eyes can make logical framing and social self-consciousness less prominent for a time.
Phenomenologically, what remains can feel like a self-running divergence engine: drive supplies energy, associations scatter, and pre-linguistic bodily sensation catches some of what emerges.
Later sleep may help fragments settle and recombine. Sometimes a person wakes with a usable connection; sometimes nothing arrives.
This is a structural and phenomenological hypothesis, not an established neural mechanism specific to clubbing. Creative workers report analogous shifts while running, showering, or falling asleep: deliberate control relaxes, and less supervised associations become easier to notice.
What AI Is Actually Doing
What AI does looks similar on the surface. Give AI a problem, and it diverges — making associations, drawing analogies, approaching from different angles. Spread out its outputs and there's a "casting a wide net" feeling.
But AI's divergence has no fire underneath.
Its sampled variation is usually pseudorandom — calculated from a random seed and sampling algorithm, reproducible when the full computational state is fixed. Its associations are strongly shaped by training data, prompts, tools, and post-training. Models can combine these materials in genuinely surprising ways, but surprise in the output is not yet evidence of felt urgency beneath it.
More fundamentally: present AI gives us no evidence of drive in the sense meant here — an internally felt, self-maintaining urgency with something at stake.
It doesn't need to speak. It can't not answer — but wait, isn't that the same thing? No. AI's "can't not answer" is a response after the system is triggered, an external request activating a generation process. A human's "can't not do something" may be an internal surge that precedes the current request and is not exhausted by it. Kant couldn't not write the Critique of Pure Reason — not because someone submitted a task, but because after Hume stung him, something kept growing out of the wound.
That difference is the difference between libido and no libido.
AI is a high-speed blender whose engine and objective are supplied from outside. It can propose directions the user had not imagined, drawing on training, context, tools, and search. But a proposal is not yet a direction that matters to the system itself. AI can generate candidates; the user still supplies or accepts the stake.
Meditation Is the Other Path
Clubbing can be understood here as a bottom-up route: beat and bodily movement may loosen deliberate management and give less supervised associations more room.
Meditation works top-down — using willpower to dismantle the framework layer by layer, standing bare-handed in front of drive and saying "you sit down first." Nothing borrowed — no beat, no movement, no vestibular disruption. Every step is self-contradictory: using choice to choose "not to choose."
That's why meditation is hard. A meditation master is genuinely impressive — dismantling their own framework without any external tools.
The bodily state clubbing reaches is hot, full — packed with just-released energy. The bodily state meditation reaches is empty, still — drive being steadily held. Clubbing is a shotgun: mostly noise, but occasionally hitting something you couldn't reach sitting for ten thousand years. Meditation is a telescope: precise and high-reaching, but not wild enough.
AI? AI is something like a vast, rapidly recombining library. It can produce pages no one previously wrote. What we have not established is that it can feel: "I just have to write this book."
So How Should You Use AI?
Understanding this structure, AI's correct positioning becomes clear.
AI can be a creative partner in the practical sense: it can surprise, challenge, and extend you. Structurally, the division proposed here is different: you supply or ratify the drive; it supplies coverage and variation.
You have a vague directional sense — you can't say why, but this direction just feels right — and then you use AI to rapidly unfold in that direction, seeing what exists on that path, discovering connections you hadn't thought of, filtering out what's worth digging into. That's a cultivating use.
Conversely, if you use AI as the sole source of direction — "AI, tell me what to do, what to write, what to think" — then you're asking generated plausibility to replace your own libido. The model may offer an unusual and valuable path; the danger is not that it can produce only the statistical average, but that you may adopt a path before discovering whether it matters to you.
The least creative use: let AI give you your direction.
The most creative use: you first have an inexplicable impulse, then use AI to amplify that impulse as far as it can go.
One Last Question
People ask me: will AI eventually have libido?
The honest answer is: we do not know. Within the Self-as-an-End hypothesis used here, libido would require a structurally irreducible source of direction, not merely varied output; the framework connects that source to genuine physical unpredictability accumulated through time. That connection is a theoretical commitment, not an established result of AI or consciousness science. If an artificial system ever developed a drive it could own and sustain, the change would be categorical in our relationship to it, whatever physical mechanism made it possible.
Until that day, AI is a powerful tool, and the tool's power source is you.
You are the engine. Don't outsource the engine.
我们一直在讨论 AI 的"创造力",讨论它会不会抢走艺术家、作家、设计师的工作,讨论 AI 生成的内容算不算"真正"的创作。
但有一个更基础的问题没人在谈:AI 靠什么驱动?
人类创造力的底层,有一个 AI 完全没有的东西——libido。
不是弗洛伊德意义上的性冲动,是更广义的、前语言的、前选择的生命驱力。那种说不清楚为什么、但就是不能不做某件事的内在涌动。
蹦迪在做什么
先从蹦迪说起。蹦迪为什么有时候能让人想清楚一些事?这听起来像是无稽之谈,但有一个可以严肃讨论的结构性原因。
对一些人而言,重复的 beat、身体运动与视觉输入的减少,可能会松动有意的认知监控,并把注意力更多转向体感。节奏降低了逐个规划动作的需要;闭眼时,逻辑框架与社会性的自我审查也可能暂时不那么突出。
在现象体验上,保留下来的东西可以像一个自运转的发散引擎:驱力提供能量,联想向外散开,前语言的体感接住其中一些碰撞。
之后的睡眠可能帮助这些碎片沉淀和重组。有时醒来会得到一条可用的连接,有时什么也没有。
这是一个结构与现象学假说,不是已经确立的、专属于蹦迪的神经机制。做创造性工作的人在跑步、洗澡或将睡未睡时也会报告类似变化:有意控制放松了,较少受管理的联想更容易被注意到。
AI 在做什么
AI 在做的事,表面上看起来很像。给 AI 一个问题,它会发散——联想、类比、从不同角度切入。把它的输出铺开来看,确实有一种"四面撒网"的感觉。
但 AI 的发散,底下没有火。
它采样时的变化通常是伪随机——由随机种子与采样算法计算,在完整计算状态固定时可以复现。它的联想受到训练数据、提示、工具与后训练的强烈塑形,也确实能够把这些材料组合成让人意外的新东西。但输出令人意外,还不是底下存在切身紧迫感的证据。
更根本的问题:当前 AI 没有显示出本文所说的驱力——一种从内部被感受、能够持续维持、而且有所押注的紧迫性。它的“不能不回答”是系统被触发之后的响应,是外部请求或被设定的任务激活了生成过程。人的“不能不做某件事”则可能是一种先于当前请求的内在涌动。康德不能不写《纯粹理性批判》,不是因为有人向他提交了一道任务,而是休谟刺痛他之后,有什么东西不能不从伤口里长出来。
这个区别,就是 libido 和无 libido 的区别。
AI 是一台发动机与目标都从外部接入的高速搅拌机。它可以借训练、语境、工具与搜索提出用户从未想过的方向,但一个提案还不等于系统自身所在意的方向。AI 可以产生候选方向;用户仍要提供或确认其中的押注。
冥想是另一条路
在本文框架中,蹦迪可以被理解为从下往上的路径——beat 与身体运动可能让有意管理松动,给较少受监督的联想更多空间。冥想则可以理解为从上往下的路径——用意志力一层一层松开框架,赤手空拳站在驱力面前说"你先坐下"。什么都不借——不借 beat,不借移动,不借前庭扰动。每一步都带着悖论:用选择去选择"不选择"。
所以冥想难。冥想高手是真厉害——他们在没有任何外部工具的情况下,自己拆自己的框架。
蹦迪能到达的体感状态是热的、饱满的——充满刚刚被释放的能量。冥想能到达的体感状态是空的、静的——驱力被稳稳 hold 住。蹦迪像散弹枪,大多数是垃圾,偶尔撞出坐在那里一万年也想不到的东西。冥想像望远镜,看得准,看得高,但不够野。
AI?AI 更像一座巨大而高速重组的图书馆。它可以生成从未有人写过的页面;我们尚未建立的是,它会不会感觉“我就是要写这本书”。
那 AI 应该怎么用?
理解了这个结构,AI 的定位会更清楚。它在实践上可以成为创意伙伴——让你意外、反驳你、延伸你;但本文提出的结构分工不同:你提供或确认驱力,它提供覆盖面与变化。
你有一个模糊的方向感——不知道为什么,就是觉得这个方向对——然后你用 AI 在这个方向上快速展开,看看这条路上有什么,发现你没想到的连接,筛选出值得深挖的东西。这是涵育式的用法。
相反,如果你把 AI 当成唯一的方向来源——“AI,告诉我该做什么,该写什么,该想什么”——那你是在让生成出来的可行性替代自己的 libido。模型可能给出少见而有价值的路径;危险不在于它只能吐出统计平均,而在于你还没有发现这条路对自己是否重要,就先把它领走了。
最没有创造力的用法是:让 AI 给你方向。
最有创造力的用法是:你先有一个说不清楚为什么的冲动,然后用 AI 把这个冲动放大到它自己可以走到的地方。
最后
有人问我:AI 会不会最终有 libido?
诚实的回答是:我们不知道。在本文采用的 Self-as-an-End 假说内部,libido 需要一种不能被还原为“输出更多变化”的方向来源;这个框架把它与真实物理不可预测性在时间中的积累联系起来。但这种联系是理论承诺,不是 AI 或意识科学已经确立的结论。如果人工系统有一天发展出能够由自己承担并持续维持的驱力,无论它依靠什么物理机制,我们与它的关系都会发生范畴性的变化。
在那一天到来之前,AI 是一个强大的工具,工具的动力源是用户。
你就是那个发动机。别把发动机外包出去。