What is Kahneman's attention theory?

Daniel Kahneman, answered from the texts and cited to the page.

Daniel Kahneman answers

The question of attention occupied me for years before I ever thought about heuristics or biases, and the early work shaped everything that came after. Let me say what I actually found. The starting point was a disagreement built into the data. People plainly do several things at once — driving while talking, for instance — which suggests attention is divisible.¹

But present two stimuli simultaneously and often only one is perceived, or the responses queue rather than run in parallel, which suggests a bottleneck that can only handle one thing at a time.² Both observations are real, and they pull in opposite directions. The bottleneck models were the dominant answer. Broadbent's filter theory placed the bottleneck early, before perceptual analysis; later models moved it downstream.³

But the evidence was uncomfortable for both. Selective attention does affect perceptual analysis, which cuts against the late-filter view; yet people can divide attention between concurrent messages, which cuts against the early-filter view.⁴ The conclusion I drew from that impasse was that human cognitive operations are far more flexible than either architecture would predict.

What replaced the bottleneck picture, in my own work, was a capacity model. Rather than a structural chokepoint, I proposed a general limit on the mental work a person can perform at any moment — a limited resource that can be allocated with considerable freedom across concurrent activities.⁵ The terms I used then — "exert effort," "invest capacity," "pay attention" — I treated as synonymous, because I thought the phenomenon they pointed to was the same phenomenon.⁶

Measuring pupil dilation with Jackson Beatty gave us a physiological window onto that resource: the pupil dilates under load and contracts as demand eases, and it does so with enough precision to track the second-by-second fluctuation of effort.⁷ The popular framing in Thinking, Fast and Slow rendered this as a budget. You dispose of a limited budget of attention, and if you try to go beyond it, you will fail.⁸

Effortful activities interfere with each other — you cannot compute 17 × 24 while making a left turn into dense traffic, and you should not try.⁹ Easy tasks can run in parallel; demanding ones cannot share the resource without one of them suffering. What the capacity model adds beyond the budget metaphor is a story about allocation. The response to overload is not a circuit breaker that cuts everything at once.

It is selective: System 2 protects the most important activity, and spare capacity is distributed second by second to secondary tasks.¹⁰ In the gorilla experiment Beatty and I ran, when subjects were instructed to prioritize the digit task, the timing of a visual target had no effect on digit performance — the priority instruction held — but detection of the target collapsed at moments of peak demand.¹¹

The capacity was there; it had simply been claimed. There is also a layer that sits below voluntary allocation entirely. Pre-attentive mechanisms operate outside voluntary control and provide a preliminary organization to perception — grouping, segmentation, defining the objects on which selective attention then operates.¹² The general rule that emerges is asymmetric: it is easy to focus on a single object and difficult to divide attention among several; but it is difficult or impossible to prevent the perceptual analysis of irrelevant attributes of an object you are already attending to.¹³

You cannot see shape without seeing color. That asymmetry matters, because it means the boundary of voluntary control is not where common sense places it. System 2 does have some purchase on System 1, though. It can program the normally automatic functions of attention and memory — setting yourself to look for a white-haired woman in a crowd, or to notice capital cities beginning with N.¹⁴ But maintaining that set requires continuous effort, and the moment the effort lapses, the default returns.¹⁵

Sources

Two common observations pertinent to the question of the unity of attention, but the answers they suggest are contradictory. The first of these observations is that man often performs several activities in parallel, such as driving and talking, and apparently divides his attention between the two activities.
Attention and Effort, pp. 12–13
The second basic observation is obtained when two stimuli are presented at once: often, only one of them is perceived, while the other is completely ignored; if both are perceived, the responses that they elicit are often made in succession rather than simultaneously.
Attention and Effort, pp. 12–13
Model A illustrates some central aspects of the filter theory first proposed by Broadbent (1957a, 1958). This theory assumes a bottleneck
Attention and Effort, pp. 12–13
The evidence of these studies indicates that selective attention to inputs affects perceptual analysis. This is contrary to model B. However, man is also capable of dividing his attention between concurrent messages. This is contrary to model A.
Attention and Effort, pp. 14–15
A capacity theory of attention provides an alternative to theories which explain man's limitations by assuming the existence of structural bottlenecks. Instead of such bottlenecks, a capacity theory assumes that there is a general limit on man's capacity to perform mental work. It also assumes that this limited capacity can be allocated with considerable freedom among concurrent activities.
Attention and Effort, pp. 14–15
In the present work, the terms 'exert effort' and 'invest capacity' will often be used as synonymous for 'pay attention.'
Attention and Effort, pp. 14–15
We learned a great deal about the working mind—which I now think of as System 2—from measuring pupils in a wide variety of tasks.
Thinking, Fast and Slow
The often-used phrase 'pay attention' is apt: you dispose of a limited budget of attention that you can allocate to activities, and if you try to go beyond your budget, you will fail.
Thinking, Fast and Slow
You could not compute the product of 17 x 24 while making a left turn into dense traffic, and you certainly should not try.
Thinking, Fast and Slow
the response to mental overload is selective and precise: System 2 protects the most important activity, so it receives the attention it needs; 'spare capacity' is allocated second by second to other tasks.
Thinking, Fast and Slow
We know that they followed that instruction, because the timing of the visual target had no effect on the main task. If the critical letter was presented at a time of high demand, the subjects simply did not see it.
Thinking, Fast and Slow
While the allocation of attention is flexible and highly responsive to the intentions of the moment, there are pre-attentive mechanisms that operate autonomously, outside voluntary control (Neisser, 1967). These provide a preliminary organization to perception by a process of grouping and segmentation.
Attention and Effort, pp. 14–15
The general rule that focus exclusively is it is easy to attention on a single object and difficult to divide attention among several objects. Conversely, it is easy to notice several aspects or attributes of an object, but it is difficult or impossible to prevent the perceptual analysis of irrelevant attributes.
Attention and Effort, pp. 14–15
System 2 has some ability to change the way System 1 works, by programming the normally automatic functions of attention and memory.
Thinking, Fast and Slow
you will find that the consistent maintenance of a set requires continuous exertion of at least some effort.
Thinking, Fast and Slow