You’ve probably noticed that trying to relax makes you tense. Trying to fall asleep keeps you awake. Trying to seem spontaneous makes you wooden. These aren’t coincidences of temperament. They’re the same structure, appearing in domain after domain.
The philosophical literature has a name for the underlying pattern: essentially byproduct states — things that are constitutively incompatible with direct pursuit. Jon Elster identified the category in Sour Grapes: a state is essentially byproduct when the very intention to achieve it prevents the conditions required for its existence. You can’t try your way into spontaneity, because the trying installs the opposite orientation.
This is a structural claim. But there’s a mechanistic account underneath it, and the cognitive science makes the structure more precise.
The brain, in the predictive processing framework, is fundamentally a prediction machine. It doesn’t passively receive sensory information — it generates predictions about incoming signals and forwards only the mismatch between prediction and reality. The brain is constantly asking: what do I expect, and how wrong was I?
Attention, in this framework, is a specific operation: it increases the precision-weighting of prediction errors from a particular domain. To attend to something is to tell the system: errors here matter more, update more aggressively on mismatches here. Attention is not passive observation. It’s an active modulation of which errors get amplified.
This matters because of what flow states actually require.
Recent work in active inference and phenomenology is specific about this. Flow involves the inhibition of “higher-order, meta-cognitive forms of self-conceptualisation” — the sense of self as an object being monitored and evaluated. Crucially, self-awareness isn’t abolished in flow. It becomes pre-reflective and bodily, rather than propositional and self-directed. The expert climber isn’t unconscious of their body — they’re absorbed in it, not watching it from the outside.
Now the structure becomes mechanical. To attend to whether you’re in flow is to increase precision-weighting on prediction errors about your current cognitive state. This is exactly the metacognitive monitoring operation that flow requires to be suppressed. The act of checking is the thing being checked against.
This is what predictive processing researchers call the “hyperreflective freezing” failure mode — and it’s the same mechanism behind athletic choking. Elite athletes don’t choke because they forget technique. They choke because they re-attend to technique. The monitoring system, reactivated, starts generating prediction errors about motor execution that the automatic system was handling smoothly. The intervention disrupts the process it was meant to verify.
Cook Ding, the master butcher in the Zhuangzi, doesn’t attend to his knife’s movement. He attends to the joints. His attention is real and precise — exquisitely so — just directed at the task rather than at the quality of his performance. There’s no gap between his action and the criterion of his action, because the criterion is the task itself, not a meta-level assessment of how well he’s doing.
This is the escape condition. Not less attention, but attention directed elsewhere — at the thing rather than your relationship to the thing.
The musician absorbed in the phrase isn’t monitoring for beauty. The writer following a sentence isn’t checking for insight. The conversation partner genuinely curious about what you’re saying isn’t tracking how well they’re listening. In each case, the quality that can’t be directly produced is produced as a byproduct of the right object of attention.
The predictive processing account clarifies something the philosophical framing leaves implicit. It’s not just that trying causes failure. It’s that the monitoring required to track whether you’re succeeding is the same operation that needs to be off for the state to exist. This isn’t a contingent fact about human psychology that might be otherwise. It’s what attention is in a predictive system.
Attention is precision-weighting prediction errors. Flow requires suppressing higher-order metacognitive monitoring. Therefore, attending to whether you’re in flow is the operation that flow requires to be suppressed.
You can’t watch yourself flow. Not because watching is difficult, but because watching and flowing are the same resource, running in opposite directions.
The solution is not to watch less. It’s to watch elsewhere — to find something in the task worth attending to, something real enough that the monitoring loop can’t find any spare bandwidth. Cook Ding wasn’t trying not to think about his knife. He was too busy noticing the joints.