Cognition and Biases

Module 2 mapped the structures that connect things — nodes, edges, and the shape a network takes. This module turns to whatever sits at each node: the deciding mind. Over the next three modules the human being enters the book as an agent — any entity that sizes up a situation and chooses an action. The word is deliberate. Later, in agent-based modeling, an agent will be a simulated entity following a few explicit rules; here it is a person, with all the inconsistency that implies. This module takes the agent deciding alone, under its own limits; the next takes agents in strategic contact; the one after follows them into the environment now engineered to capture their attention.

Bounded Rationality and the Architecture of Thought

Start with something ordinary. A supermarket shelf reads “Soup — $0.99 each, limit 12 per customer.” You came for two, but your cart somehow holds five. The number twelve told you nothing about how much soup you need, yet it quietly moved your hand. No one deliberated; the choice was made for you by a feature of the situation you never consciously weighed. That small surrender — fast, automatic, steered by the framing around it — is the raw material of this module.

And now you are forewarned, which should protect you. Commit to a prediction before the first demo below: will your estimate of Germany’s population land near the truth — or near whatever arbitrary number the demo happens to show you first?

If the demos worked on you the way they work on most people, your population estimate leaned toward the anchor you were shown — people shown 120 million typically guess near 98 million, people shown 20 million near 55 million; the real figure is 84.5 million. You can now name what the soup shelf did to you: an anchor, an arbitrary number that pulls a judgment toward itself. Your disease-program choice flips for most readers when identical outcomes are reworded from lives saved to lives lost — a framing effect (in the classic gain wording, 72 percent choose the “safe” program; rewrite it as losses and the majority reverses). And if dramatic causes of death ranked too high in your ordering, that is availability: whatever comes easily to mind feels common. The unsettling part: you knew these demos were about bias, and it probably did not save you.

Why not? Until the 1970s, most scientists assumed that people generally behave rationally, and explained the exceptions away with strong emotions — fear, love, hate. Daniel Kahneman and Amos Tversky dismantled that picture: the errors are not emotional outbursts but systematic, predictable, and lawful enough to map. Kahneman’s map has two parts. System 1 is always on and cannot be switched off — you cannot choose not to understand a sentence spoken in your language; to keep it out you would have to cover your ears. System 2 is normally asleep and wakes only when called: startled awake — “What was that noise?” — or when you pull yourself together instead of shouting in anger. Its work is slow, deliberate, and genuinely tiring, which is why it delegates everything it can. In summary: System 1 does heuristic work — mental shortcuts that are fast, efficient, and systematically biased — while System 2 does systematic work. This is the practical content of what Herbert Simon called bounded rationality: real minds decide within hard limits of time, information, and computation. The less time there is, the more System 1 decides.

These demonstrations came from laboratories. Did they survive? Beginning around 2015, systematic attempts to reproduce landmark behavioral findings — the replication crisis — revealed that some effects were robust: anchoring, loss aversion (losses weigh heavier than equal gains), the power of defaults (the preset option tends to win). Others crumbled. Ego depletion, the idea that willpower is a finite resource, failed a major multi-lab replication in 2016. Social priming effects, such as the famous “walking slowly after reading elderly-related words” study, could not be reproduced. The Replication Scorecard in the explorer above shows the tally.

Try one more decision: which city has more inhabitants, San Diego or San Antonio? German students, asked this, overwhelmingly say San Diego — often the only one of the two they recognize — and they beat American students who know both cities. Choosing the option you recognize is a shortcut, and in an environment where recognition tracks the answer, the shortcut wins. Gerd Gigerenzer built a different lens from results like this: ecological rationality. The question is not whether a heuristic is biased, but whether the environment it evolved for still matches the environment it operates in.

Humans are not blank slates rationally processing information — they are bounded agents using contextual shortcuts. That raises two questions the rest of this part takes up in turn: what happens when several such minds must anticipate one another, and what happens when the environment those shortcuts operate in is redesigned by algorithms? The first is the subject of the next module — strategic interaction, where your best move depends on what someone else does.