注意力是关系，不是状态
Attention as relation, not state
EEG and the philosophy of attention 的延续。
神经反馈产业兜售一个诱人的故事：注意力是一种脑状态，这个状态有可测量的特征（alpha/theta 比、SMR、P300 潜伏期），通过对这些特征做闭环，大脑就能学会更长时间地停留在那里。
我认为这个故事对了一半，而它没说的另一半才是真正危险的部分。
它对的地方
确实存在可被复现的、与持续注意相关的脑电特征。SMR 上调研究（Egner & Gruzelier, 2003-2004）以及针对 ADHD 的 theta/beta 比方案，已经积累了几十年数据。某些事确实在发生。
它错的地方
注意力不在头里。注意力是主体与世界之间的关系——什么算作”专注”取决于任务、可供性、主体的目标，以及环境的时间结构。一个 alpha 曲线平坦的禅修者和一个正在缠斗中的战斗机飞行员都在”专注”，但他们专注的对象几乎正交。
当我们训练人产生某种特定的脑电特征时，我们就是在训练他们产生那种特征。原本与该特征共现的与世界的接触是否会一并被带回来，是一个经验问题，而证据远比营销文案弱。
我想把这条线推向何处
- 给”注意力”做一个分类法——持续 / 选择 / 分配 / 执行 / 定向——并搞清楚到底哪些脑电特征对应哪一种。
- 闭环训练的后果：我们是不是在教受试者”骗自己的指标”，就像学生学会应付标准化考试？
- 一个奖励信号以任务为根而非以特征为根的研究设计。这样特征就只是一种工具，而不是判据。
参见：Predictive processing 101，“状态”也许根本就是错的本体论。
A continuation of EEG and the philosophy of attention.
The neurofeedback industry sells a tempting story: attention is a brain state, that state has measurable signatures (alpha/theta ratios, SMR, P300 latency), and by closing the loop on those signatures the brain learns to dwell there longer.
I think this story is half right and dangerous because of the half it leaves out.
What’s right
There are real, replicable EEG correlates of sustained attention. SMR-uptraining studies (Egner & Gruzelier, 2003-2004) and theta/beta ratio protocols for ADHD have decades of data. Something is happening.
What’s wrong
Attention isn’t in the head. Attention is a relation between an agent and a world — what counts as “attentive” depends on the task, the affordances, the agent’s goals, and the temporal structure of the environment. A meditator with a flat alpha curve and a fighter pilot mid-merge are both “attending” but to almost orthogonal things.
When we train people to produce a specific EEG signature, we are training them to produce that signature. Whether the world-engagement that originally co-occurred with that signature comes along for the ride is an empirical question, and the evidence is weaker than the marketing copy.
Where I want to take this
- A taxonomy of “attention” — sustained / selective / divided / executive / orienting — and which EEG signatures actually map to which.
- Consequences for closed-loop training: are we teaching subjects to hack their own metric, the way students learn to hack standardised tests?
- A research design where the reward signal is task-grounded, not signature-grounded. The signature is then merely an instrument, not the criterion.
See also: Predictive processing 101, for why “states” might be the wrong ontology entirely.
💬展开评论 / Show comments