A poverty of attention

Herbert Simon spent his career on how people and organisations decide when they cannot know or compute everything. He called the condition bounded rationality: decision-makers do not maximise over all possibilities but search until they find something good enough (Simon 1955). In 1978 he received the Nobel Memorial Prize in Economic Sciences for this work on decision-making in organisations (The Royal Swedish Academy of Sciences 1978).

In a 1971 lecture on organisations in an “information-rich world”, Simon turned the idea around. Designers of information systems, he argued, worried about a shortage of information when the real shortage was of the attention needed to process it. Information consumes attention, and so a wealth of information creates a poverty of attention and a need to allocate it efficiently among the sources that compete for it (Simon 1971).

This chapter puts numbers on the budget and then asks how a rational person should spend it.

The hard limit

A day has 24 hours; most adults are awake for about 16. Within those hours, two very different measures of capacity are useful.

The first is behavioural throughput: how much information a person can take in and act on. A recent review collected measurements from typing, speech, reading, games and memory tasks and found that, across very different activities, human behaviour runs at roughly 10 bits per second, many orders of magnitude below what the senses deliver (Zheng and Meister 2025). The second is reading speed: a meta-analysis of 190 studies put the silent reading rate of adults reading non-fiction in English at about 238 words per minute (Brysbaert 2019).

awake_h = 16                       # waking hours per day
# conscious throughput, order of magnitude (Zheng and Meister 2025)
bits_per_s = 10
words_per_min = 238                # silent reading of non-fiction by adults (Brysbaert 2019)

seconds = awake_h * 3600
print(f"waking seconds per day:          {seconds:,}")
print(f"bits of conscious throughput:     {seconds * bits_per_s:,}  (= {seconds * bits_per_s / 8 / 1000:.0f} kB)")
print(f"words if every minute were read:  {awake_h * 60 * words_per_min:,}")
print(f"60-second videos if nothing else: {awake_h * 60:,}")
# waking seconds per day:          57,600
# bits of conscious throughput:     576,000  (= 72 kB)
# words if every minute were read:  228,480
# 60-second videos if nothing else: 960

Seventy-two kilobytes a day, less than a single small photograph, is the order of magnitude of what a person can consciously process. Even the 228,000 words, about three novels, would require reading without pause from waking to sleep. Against this budget, the supply is effectively unlimited: a single platform shows each user more videos in a day than they could watch in a year.

How a scarce budget should be spent

Economics has a standard way to think about a fixed budget spread across uses with diminishing returns. Gary Becker treated time itself as a resource that households allocate, with a “shadow price” equal to what an extra hour would be worth (Becker 1965). The same logic applies to attention. If the value of time spent on source $i$ grows as $u_i(t_i) = a_i \ln(1 + t_i/s)$, the best allocation of a total of $T$ minutes makes the marginal value of the last minute equal across every source that gets any time at all:

$$\frac{a_i}{s + t_i} = \lambda \quad \text{for every } t_i > 0, \qquad \sum_i t_i = T .$$

The common value $\lambda$ is the shadow price of attention. A source whose value at zero time, $a_i/s$, is below $\lambda$ gets nothing.

# Allocate T minutes among sources with diminishing returns u_i = a_i * ln(1 + t_i / s).
# Optimum (KKT): every source that gets time has the same marginal value lam = a_i / (s + t_i).
# minutes to spend; scale of diminishing returns
T, s = 120.0, 10.0
a = {"long video A": 9.0, "newsletter B": 6.0, "short feed C": 4.0, "short feed D": 1.2}

def allocate(lam):
    return {k: max(0.0, v / lam - s) for k, v in a.items()}

lo, hi = 1e-6, max(a.values()) / s                    # bisection on the shadow price lam
for _ in range(100):
    lam = (lo + hi) / 2
    lo, hi = (lam, hi) if sum(allocate(lam).values()) > T else (lo, lam)
t = allocate(lam)
print(f"shadow price of one more minute: {lam:.4f} units of value")
for k, v in t.items():
    print(f"{k:13s} {v:6.1f} min")
# shadow price of one more minute: 0.1267 units of value
# long video A    61.1 min
# newsletter B    37.4 min
# short feed C    21.6 min
# short feed D     0.0 min

Three lessons follow, and they recur throughout the book. Attention is spread across sources, not given entirely to the best one, because of diminishing returns. The shadow price falls as the budget grows and rises as it shrinks: a person with ten free minutes is much harder to win than one with two hours. And there is a threshold: a source that cannot offer at least the shadow price at its first minute gets nothing at all, which is why so much content receives no attention whatever, rather than a little.

Attention is not one thing

Psychologists distinguish the capacity to select one stream from many, the capacity to sustain focus over time, and the effort that attention costs. Daniel Kahneman’s account treated attention as a limited pool of effort that tasks draw on, with harder tasks drawing more (Kahneman 1973). For the economics of creator commerce, two consequences matter. First, a minute of attention to a demanding long video and a minute spent scrolling a feed are not the same commodity; they draw on effort differently and leave different traces in memory. Second, a feed that requires almost no effort per item can fill more of the day than one that requires a lot. Chapters 4 and 8 return to this difference.