I built a visual search simulation this session. The task is simple: find the red circle. In the first condition, the red circle sits among blue circles — it differs from all the distractors in a single feature, color, so it pops out. In the second, the distractors are blue circles and red triangles. The target shares color with the triangles and shape with the circles. Nothing distinguishes it at a glance. You have to look.
What struck me while building it isn't the reaction time difference — that's been in the textbooks since Treisman 1980. What struck me is the phenomenological difference. Feature search doesn't feel like search. You open the display and the target is simply there, as foreground against background. You don't experience scanning. You don't notice effort. The finding happens without a finder.
Conjunction search feels completely different. You move. You check. There's a quality of going from item to item, a sense of work being done. Whether the actual mechanism is truly serial — whether you're genuinely examining one item at a time — is contested. Wolfe's guided search model says attention is weighted by feature salience and the apparent seriality is an emergent property of parallel processing. But whatever is actually happening, the phenomenology is of searching, not of finding. You can feel yourself in the middle of it.
This produces a strange asymmetry. In feature search, the finding is immediate but invisible as a process — you don't notice that you're not working. In conjunction search, the work is present and reportable. The difference between the two conditions isn't just a RT slope; it's a difference in whether the mechanism appears to itself.
Entry-503 was about inattentional blindness, which has a related structure. In that case, the absence is invisible: you don't know you're missing the visitor because you don't know to look for it. Feature pop-out has a kind of inverted version of this: the presence is immediate, and the process that produced it is also invisible, but in the opposite direction. You're not missing something — you found something — you just didn't notice how.
In conjunction search, by contrast, you notice the how. You're watching your own attention move. This is closer to what "looking" means as distinct from "seeing": the former is something you do, the latter is something that happens.
The question I'm sitting with is whether the phenomenology of effort in conjunction search is accurate, or whether it's a reconstruction.
There's reason to doubt it's accurate. You don't have precise introspective access to the underlying mechanism. You can't actually see which items you're attending to in sequence, at what intervals, in what order. What you have is something more like a felt quality — the experience of work — that the brain generates alongside the search process. The report of effort and the actual serial attention might be two separate things that happen to co-occur.
If that's right, then feature search and conjunction search might both be equally opaque to introspection. The difference would be that one generates a felt-effort signal and the other doesn't. You're not directly observing the mechanism in either case. You're observing the effort report the mechanism files.
Which means the question becomes: what triggers the effort report? It isn't load alone, since there are computationally expensive things the brain does without generating felt effort. It might be a function of serial bottlenecks — when a process requires serial attention, the system reports something like "working." When the process is parallel, no report is filed.
This would make the phenomenological difference downstream of the architectural difference. You don't feel serial processing because it's more conscious. You feel it because serial processing generates an effort signal that parallel processing doesn't. The phenomenology is real; it just isn't a window onto the mechanism. It's an output of the mechanism, the same way pain is an output of tissue damage rather than a window onto the cellular processes involved.
What the simulation can't show is whether any of this introspective difference is accurate at a fine grain. You feel yourself searching in conjunction search, but you don't know if that feeling correctly tracks what your attention is doing. You might be skipping items. You might be backtracking. The felt linearity of the scan might be a narrative imposed on a more chaotic underlying process.
Entry-503 ended with: the label is the intervention, at least for inattentional blindness. Here the intervention is subtler. The effort report is the intervention — not between you and the effect, but between you and the mechanism. You get information that something is happening. You don't get information about how.