The bottleneck's location shifts with task type; there is no single fixed filtering stage

What it is

Early and late selection theory argued for a long time over exactly where filtering happens. Later research's answer was that the question itself was framed wrong: the bottleneck's location is not fixed at one stage — it moves between early and late depending on the nature of the current task. This view is called flexible bottleneck theory, and its most influential specific version is perceptual load theory: whether the bottleneck sits early or late depends on how much of the perceptual resource the current task consumes.

This is not a diplomatic compromise splitting the difference between the two earlier theories — it offers a unified framework that explains both camps' evidence at once. The experimental findings behind early selection theory and those behind late selection theory are both genuine; they simply correspond to performance under different perceptual load conditions.

Why it happens

The core logic of perceptual load theory is: perceptual analysis itself consumes resources, and this resource is automatically committed, as fully as possible, to whatever perceptual processing the current task demands, until it runs out. When the primary task's perceptual load is low (finding a target in a sparse, simple visual scene, say), perceptual resources go unused, and the surplus automatically spills over, incidentally processing irrelevant information that didn't need attention — at that point irrelevant information can be processed to a relatively deep level, looking more like late selection. When the primary task's perceptual load is high (a crowded visual scene requiring fine discrimination), all perceptual resources are exhausted by the primary task, and irrelevant information gets excluded at the early stage simply because the resource has run out, never reaching the semantic level at all — looking more like early selection.

In other words, the phenomena each of early and late selection theory describes are both real; they are not two mutually exclusive truths, but two visible expressions of the same flexible mechanism under different load conditions. This mechanism shows the bottleneck's location was never a pre-fixed switch setting in the system — it cuts off wherever the resource happens to run out, and it moves in real time with the task.

Studying it

The standard paradigm is a perceptual load manipulation: participants perform a primary perceptual discrimination task whose own perceptual load is manipulated (e.g. the number of simultaneously presented distractors, or the similarity between target and background), while a completely irrelevant probe stimulus is placed at the edge of the scene. The test is whether that irrelevant stimulus gets processed — for example, whether it produces a priming effect or a reaction-time cost.

Common independent variables: the primary task's perceptual load level (low vs. high), and the spatial/semantic relationship between the irrelevant stimulus and the primary task. Common dependent variables: the amount of interference the irrelevant stimulus causes (RT cost, error rate) as a function of primary-task load — a curve showing clear interference under low load and vanishing interference under high load is the central piece of evidence for perceptual load theory.

In interface research, this logic is commonly used to explain a counterintuitive phenomenon: the simpler and "cleaner" an interface, the more easily users are actually distracted by irrelevant elements (ads, pop-ups); when the interface itself is information-dense and the task itself is effortful, the same irrelevant elements are less likely to cause interference — a direct manifestation of perceptual load determining bottleneck location.

A methodological caution: "perceptual load" here refers specifically to load at the perceptual discrimination stage, a distinct construct from "cognitive load" involving active memory or decision-making. If manipulating perceptual load inadvertently raises cognitive load at the same time, the source of the resulting effect becomes confounded.

Where it stops holding

• This theory only answers where a perceptual-type bottleneck moves to; it does not address whether attentional resources form a single, homogeneous pool. Whether the resource pool needs to be subdivided by sensory channel into multiple independent pools is a separate theoretical question that flexible bottleneck theory makes no claim about and does not elaborate here.
• Load manipulation must be at the perceptual level and kept distinct from semantic/memory load. If an experimental design confounds the two kinds of load, the resulting curve cannot be attributed to perceptual load theory directly.
• This theory has been most thoroughly validated in the visual domain. Evidence is comparatively limited on whether the same load-dependent rule for the bottleneck holds in auditory or cross-channel settings.
• "Complete exclusion of interference" under high perceptual load is a statistical tendency, not an absolute guarantee — an extremely salient irrelevant stimulus can still break through this mechanism, the same limitation noted elsewhere: a task goal can suppress bottom-up capture but never eliminate it entirely.

Related

• Same group: A5.10.1 Limited processing resources mean simultaneously presented information necessarily involves trade-offs · A5.10.2 Early selection theory holds that filtering happens before semantic analysis · A5.10.3 Late selection theory holds that all channels are processed to the semantic level, with selection happening afterward · A5.10.4 Capacity allocation is dynamic — a harder sub-task takes a larger share at other tasks' expense
• Nearby: A5.12 Top-down attentional guidance · A9.10 Multiple resource theory (how resource pools split across sensory/response channels — a different question from the within-channel bottleneck movement discussed in this group)
• Search terms: flexible bottleneck · perceptual load theory · moveable bottleneck · Lavie