Behavioral Economics

Learning Objectives

By completing this book, you will be able to:

1. Explain why the rational agent model fails as a description of human behavior and what this implies for the design of systems that depend on people's decisions.

2. Apply Kahneman and Tversky's Prospect Theory to predict how people will evaluate gains, losses, and probabilities in real decision situations.

3. Identify the systematic behavioral biases—the endowment effect, mental accounting, status quo bias—and recognize how they distort the formulation and execution of intentions.

4. Design choice architectures that use defaults, simplification, and feedback to align real behavior with declared intention.

5. Analyze the intention-behavior gap as a predictable and structural phenomenon, not as a failure of individual willpower.

6. Apply the principles of behavioral economics to the design of AI interactions that anticipate and compensate for the systematic deviations of human behavior.

1. From Homo Economicus to the Real Human

For more than two centuries, economics built its models on an extraordinary assumption: that human beings are rational agents who maximize their expected utility with consistent and stable preferences. This hypothetical being—Homo economicus—calculates costs and benefits with precision, is never influenced by how options are presented, and always chooses what suits them best in the long run. It is a mathematically elegant model. And it is systematically false as a description of how people really decide.

“An executive declares in Monday's meeting: ‘This quarter I'm going to prioritize strategic planning.’ By Friday he has spent five days answering urgent matters. He did not change his mind—his preferences are not inconsistent in the abstract. What happens is that each morning, the immediate urgency activates his System 1, which overweights the present and discounts the future. His declared intention was sincere. His behavior follows a different logic.”

Behavioral economics was born as an empirical rebellion against this model. It did not propose replacing economics with psychology, but rather enriching economic models with what cognitive psychology had already discovered: that human decisions are predictably irrational. Not random—predictable. The errors have structure, pattern, and direction. And it is precisely that predictability that makes interventions possible.

Richard Thaler, in Misbehaving, documents how the economics profession resisted for decades the evidence that humans do not behave as the standard model predicts. Economists knew that people make mistakes, but they argued that these cancelled out in the aggregate, or that market competition eliminated them. Both arguments turned out to be empirically incorrect: biases are systematic, not random, and they persist even in competitive markets and among experienced professionals.

“Doctors—professionals trained in evidence-based reasoning—display the same behavioral biases as anyone else when making clinical decisions under time pressure. They overweight recent cases (availability), they seek to confirm their initial diagnosis (confirmation), and they are reluctant to change a treatment they have already begun (status quo). Professional experience does not eliminate the biases; sometimes it reinforces them.”

For Intention Architecture, this lesson is foundational: a system that models its users as rational agents will systematically fail to predict their behavior. If you design an interface, a prompt, or a policy assuming that people will choose what suits them best when presented with the correct information, you are making the Homo economicus error. The necessary information is not sufficient. How it is presented, when it is presented, and what is required of the user in order to act on it determines whether intention is translated into behavior or remains an aspiration.

1.1 From Anomaly to Pattern

What we today call behavioral economics began as a collection of “anomalies”: behaviors that did not fit the standard theory. Economists treated them as laboratory curiosities. Kahneman and Tversky demonstrated that they were not anomalies but the norm—that the theoretical model, not human behavior, was the deviation.

Three fundamental findings demolished the classical model. First, preferences are not stable: they depend on the reference point and on the frame in which they are presented. Second, people do not evaluate probabilities correctly: they overweight small ones and underweight large ones. Third, losses hurt more than equivalent gains please: the value function is asymmetric. These three empirical facts are the basis of Prospect Theory, which you will study in the next section.

“A restaurant offers a menu with three wine options: one at 20 euros, another at 35, and another at 60. Most people choose the 35—the middle one. If the restaurant changes the options to 35, 60, and 90 euros, most choose the 60. The preferences did not change—the context did. The rational agent would choose the same wine regardless of the alternatives. The real human chooses as a function of the set of options available.”

2. Prospect Theory: How We Really Evaluate

Prospect Theory, published by Daniel Kahneman and Amos Tversky in 1979, is probably psychology's most important contribution to twentieth-century economics. It replaced expected utility theory—which says how people ought to decide—with a descriptive model of how they really decide. It is neither a normative model (how to decide better) nor a prescriptive one (what you ought to do), but a descriptive one: this is how your mind works when it evaluates options under risk.

The theory has three pillars that transform how we understand human decisions, and each one has direct implications for Intention Architecture.

2.1 Reference Points and Loss Aversion

The first pillar of Prospect Theory is that people evaluate outcomes not in absolute terms but relative to a reference point. You do not calculate how much you have, but how much you gain or lose relative to where you are. This reference point is typically the status quo—your current situation—but it can be manipulated by the frame, by expectations, or by social comparison.

“An employee who expected a 10% raise and receives 7% experiences a loss, even though objectively he gained 7%. His reference point was not his current salary but the expected raise. Meanwhile, a colleague who expected 3% and receives 5% experiences a gain of 2%. Both receive raises. Only one is satisfied. The objective reality is identical for the practical effects on their pockets; the subjective experience is radically different.”

The second discovery is loss aversion: losses weigh approximately twice as much as equivalent gains. Losing 100 euros produces more pain than the joy of gaining 100 euros. The value function has an asymmetric S shape: it is concave for gains (each additional euro is worth a little less) and convex for losses (each euro lost hurts a little more), with a steeper slope in the territory of losses.

This means people are risk-averse when it comes to gains (they prefer to win 80 euros for certain over an 80% probability of winning 100) and risk-seeking when it comes to losses (they prefer an 80% probability of losing 100 over losing 80 euros for certain). The same individual is conservative and risk-taking, depending solely on whether the frame presents the situation as a gain or as a loss.

“A finance director who has accumulated losses on a project tends to double down—investing more resources instead of cutting losses. It is not that he is irrational; his brain is in the territory of losses, where Prospect Theory predicts exactly this risk-seeking behavior. The sunk cost fallacy you studied in Module I is not only a cognitive bias: it is a direct consequence of the asymmetric value function.”

2.2 Non-Linear Weighting of Probabilities

The third pillar is that people do not process probabilities linearly. They overweight low-probability events (which is why they buy lottery tickets and insurance against improbable catastrophes) and underweight medium-to-high probabilities. There is a certainty effect: a certain outcome carries a disproportionately greater psychological weight than a highly probable one. The difference between 99% and 100% is psychologically enormous; the difference between 45% and 46% is imperceptible.

“When an AI system says ‘I have 95% confidence in this answer,’ the user processes that 5% of uncertainty very differently depending on the context. In a casual search, the 95% is treated as certainty. In a medical diagnosis, the 5% margin of error is magnified until it dominates the experience. The same objective probability produces radically different reactions depending on what is at stake and how it is presented.”

The combination of loss aversion and non-linear weighting of probabilities produces a pattern of four attitudes toward risk that Kahneman calls the “fourfold pattern”: risk aversion in the face of probable gains (accept the sure deal), risk seeking in the face of probable losses (gamble to avoid the certain loss), risk seeking in the face of improbable gains (buy lottery tickets), and risk aversion in the face of improbable losses (buy insurance). This pattern explains behaviors that classical theory considered contradictory.

3. Systematic Behavioral Biases

Prospect Theory provides the general framework; behavioral biases are its specific manifestations in everyday life. They are not random defects but structured patterns emerging from the cognitive architecture you studied in Module I. What Module I explained as biases of information processing, behavioral economics connects with measurable and designable economic consequences.

3.1 The Endowment Effect

The endowment effect is the tendency to value the things we already possess more than those we do not possess, even when the possession is trivial or recent. Thaler documented it in classic experiments: students who are given a coffee mug ask, on average, twice as much money to sell it as other students would be willing to pay to buy it. The same mug, valued by the same type of person, is worth twice as much when you have it as when you do not.

“A user who has spent months using a mediocre software tool resists switching to an objectively superior alternative. It is not that he cannot see the advantages—it is that the endowment effect turns the features of the current tool into possessions he would lose, while the features of the alternative are potential gains. And loss aversion says that losses weigh more. Result: the user stays with the mediocre option, not from lack of information but from an asymmetry in how his brain values what he has versus what he could have.”

The endowment effect is a direct consequence of loss aversion: when you possess something, parting with it is coded as a loss; when you do not possess it, acquiring it is coded as a gain. Since losses weigh more, the willingness to sell is always greater than the willingness to buy. This has enormous implications for Intention Architecture: any change you ask of the user involves losing something (their current configuration, their way of working, their familiar workflow), and that perceived loss will weigh more than the objective gain.

3.2 Mental Accounting

Mental accounting, a central concept of Thaler's, describes how people organize, evaluate, and track their financial activity in separate mental accounts, violating the economic principle of the fungibility of money. A euro is not a euro: a euro of salary is not spent the same way as a euro from an unexpected bonus; a euro ‘saved’ in an emergency fund is not touched even if you have credit card debt at 20% interest.

“A person who swore to save 500 euros a month for her retirement fund refuses to cancel three digital subscriptions of 15 euros each that she does not use. In rational terms, cancelling the subscriptions is an instant saving of 45 euros a month at no real cost. But in the user's mental accounting, retirement saving sits in a long-term mental account, while the subscriptions sit in the mental account of everyday expenses. They are sealed compartments that do not communicate.”

Mental accounting produces three effects that are particularly relevant to Intention Architecture. The first is the asymmetric treatment of costs and benefits in different accounts. The second is the evaluation of transactions not in terms of net result but according to how they are grouped: people prefer to gain 50 twice rather than 100 once (segregating gains), and prefer to lose 100 once rather than 50 twice (integrating losses). The third is the closing of accounts: closing a mental account in the negative—selling a stock at a loss—produces a pain that prevents optimal decisions.

3.3 Status Quo Bias

Status quo bias is the disproportionate preference for the current state of things, regardless of whether it is optimal. It combines the endowment effect (valuing more what one has), loss aversion (fearing what is lost more than desiring what is gained), and ambiguity aversion (the known seems less risky than the unknown, even when it is not). The result is a powerful bias toward inaction: when the default option is not to change, most people do not change.

“When countries change their organ donation system from opt-in (you must actively register to be a donor) to opt-out (you are a donor unless you exclude yourself), donation rates go from 15–20% to 85–99%. The preferences of the population did not change. What changed was the default. Status quo bias did the rest. This is perhaps the most powerful demonstration that the design of the decision environment can matter more than the preferences of the individual.”

4. Choice Architecture

If behavioral biases are predictable, then the decision environment can be designed to compensate for them. This is the central thesis of Nudge, the book by Richard Thaler and Cass Sunstein that introduced the concept of choice architecture: the deliberate design of the context in which people make decisions, with the aim of facilitating better outcomes without restricting freedom of choice.

The fundamental idea is that there is no such thing as a ‘neutral design.’ Every presentation of options has a design, and that design influences the choice. The order in which options appear, which is the default option, how each alternative is described, how much information is presented, and how much effort it takes to change the selection—all of this shapes the decision. The choice architect is the person who configures that context, consciously or not. If you do not design it, someone or something is designing it for you.

“A company cafeteria that places the salads at eye level and the fries on a lower shelf does not prohibit fries—it merely changes the choice architecture. Visual accessibility is a nudge. Employees can still choose fries, but the design of the environment favors the healthier option. The same thing happens when an AI system presents options: which appears first, which has more detail, which is preselected—all of it is choice architecture.”

4.1 The Power of Defaults

Of all the tools of choice architecture, the most powerful is the default—the option that applies if the user does nothing. Defaults are extraordinarily sticky: most people accept the default, regardless of what it is. This is not due to preference but to a combination of status quo bias, the cognitive effort of evaluating alternatives, and the implicit signal that the default is the ‘recommended’ option.

Thaler and Sunstein documented the power of defaults across multiple domains: retirement savings plans (automatic enrollment multiplied participation), organ donation (opt-out defaults raised rates to 85–99%), and privacy settings (most people never change the default settings of their applications). In each case, the option chosen as the default dominates final behavior, regardless of declared preferences.

“When an AI system generates a response and offers a preselected ‘accept’ button and a small ‘edit’ link, the default is to accept. The choice architecture is pushing the user toward passive acceptance. If the objective is for the user to review the response critically, the design must invert the architecture: make review the path of least friction.”

4.2 Simplification, Feedback, and Friction

Beyond defaults, choice architecture operates through three additional mechanisms. Simplification reduces the cognitive load: the easier the options are to understand, the better the decision. Feedback informs about consequences: people make better decisions when they receive timely information about the results of their choices. And friction—the difficulty of executing an action—can be used in both directions: reduce it to facilitate good behaviors, increase it to discourage bad ones.

“An AI system that wants the user to formulate their requests better can use productive friction: instead of accepting any prompt and generating the best possible answer, it can ask a clarifying question before responding. That friction—a second of pause and reflection—improves the quality of the intention formulated. But it must be calibrated friction: too much and the user abandons; too little and the user never improves their formulation.”

Thaler proposed the acronym NUDGES to summarize the design principles: iNcentives, Understand mappings, Defaults, Give feedback, Expect error, and Structure complex choices. Each principle addresses a specific cognitive limitation documented by psychology and offers a design tool for compensating for it.

5. The Intention-Behavior Gap

Perhaps the most relevant phenomenon in all of behavioral economics for Intention Architecture is the intention-behavior gap: the documented fact that people systematically do not do what they say they want to do. Not for lack of information, not for lack of declared motivation, not for lack of access to the means. But because human cognitive architecture produces a predictable disconnection between the self that plans and the self that executes.

“Eighty-eight percent of New Year's resolutions fail before February. Not because people did not want to change—they wanted it enough to declare it publicly. They fail because the self of January, full of abstract motivation, made commitments that the self of February, confronted with the concrete reality of the effort, cannot or will not sustain. They are two agents with different utility functions negotiating inside the same person.”

5.1 The Present Self vs. the Future Self

Temporal discounting—the tendency to prefer small immediate rewards over large future ones—is one of the most robust findings in behavioral economics. But the most useful formulation is not that of discounting: it is that of the ‘multiple selves.’ Thaler proposes that within each person there is a planner (the future self, who sets long-term goals) and a doer (the present self, who makes decisions moment to moment). The conflict between the two explains the intention-behavior gap.

The planner self operates with System 2: reflective, future-oriented, capable of weighing distant consequences. The doer self operates with System 1: impulsive, present-oriented, dominated by immediate gratification. When a person formulates an intention—‘I want to save more,’ ‘I want to exercise,’ ‘I want to use AI more strategically’—it is the planner speaking. When the moment of execution arrives, the doer decides. And the doer never read the planner's memo.

“A professional resolves to set aside 30 minutes every morning to formulate his requests to the AI precisely before sending them. It is a perfectly formulated intention. But every morning, on opening his computer, he sees 47 urgent emails. The doer self—activated by urgency, operating with System 1—starts answering emails immediately. The 30 minutes of formulation never happen. The intention was sincere; the environment was not designed to support it.”

5.2 Commitment Devices

If the intention-behavior gap is predictable, it is also designable. Commitment devices are strategies the planner self implements to restrict the options of the doer self, reducing the probability of deviation. Ulysses ordered his sailors to bind him to the mast before passing the sirens: the original example of a commitment device.

Modern commitment devices take many forms: the automatic savings transfer that executes on payday (before the doer self can spend it), the alarm that locks the phone after 11 PM, the social contract of going to the gym with a partner. All share a structure: they are decisions made by the planner, in a moment of clarity, that restrict the options of the doer in the moment of temptation.

“A company configures its AI system so that, before executing an ambiguous request, it requires the user to complete a brief clarification form. The user—the planner self who designed the policy—knows that clarification improves results. The user—the doer self who wants the answer now—curses the form. But clarification is a commitment device: a productive friction that the planner imposes on the doer to protect the quality of the intention.”

6. Behavioral Economics and Intelligent Systems

Everything above converges on a practical question for the intention architect: how do you design AI interactions that anticipate and compensate for the user's behavioral deviations? The answer is not to correct the user—it is to design the system so that it works with real humans, not with imaginary rational agents.

When a user interacts with an AI system, they bring with them all the behavioral baggage we have studied: reference points that bias their evaluation of the options, loss aversion that magnifies the risk of change, mental accounting that fragments their reasoning, status quo bias that anchors them to their initial formulation, and temporal discounting that prioritizes the fast answer over the correct one. A system that ignores these factors is a system that systematically disappoints.

“A user asks an AI system: ‘Give me a quick marketing strategy for this launch.’ The adjective quick is the doer self speaking—it wants the answer now, not the better answer. If the system obeys literally, it produces a superficial response that confirms the user's temporal bias. If the system applies behavioral economics, it can respond: ‘I can give you three ideas in 30 seconds. I can also give you a more robust strategy if you give me 3 minutes to ask some questions. Which do you prefer?’ This preserves freedom of choice while making the trade-off visible.”

Five design principles emerge from applying behavioral economics to interaction with AI:

1. Do not assume rationality. Design for users with an active System 1, fragmented attention, and loss aversion. If your system requires the user to read, evaluate, and compare complex options, most will not do it—they will choose the first reasonable option or the default.

2. Design the defaults deliberately. Every response has an implicit default: the first option, the suggested format, the level of detail presented. Choose those defaults based on what produces better results for the user, not on what is easier to generate.

3. Use friction productively. Not all ease is good. Sometimes a second of pause—a clarifying question, a reformulation of the problem—radically improves the quality of the result. The friction must be calibrated: enough to activate System 2 without being so much that the user abandons.

4. Make the intention-behavior gap visible. When you detect an inconsistency between what the user says they want and what their behavior suggests, point it out respectfully. ‘You say you want a deep analysis, but your questions suggest you need a quick answer. Which is the real priority?’

5. Offer commitment devices. Allow the user to configure restrictions for their future self: ‘I want you to always ask me for clarification before answering ambiguous requests,’ ‘I want you to show me the trade-offs before giving me the answer,’ ‘I want you to remind me of the second-order effects.’

7. Practical Application: The Case of the Wellness Platform

HealthPath is a digital corporate wellness platform that integrates AI to personalize health programs for employees. Its value proposition is clear: use artificial intelligence to adapt recommendations for exercise, nutrition, and stress management to each user. The technology works. The algorithms are sophisticated. The data are accurate. And yet, after six months, only 12% of enrolled employees are still using the platform actively. The other 88% abandoned it.

The product team diagnoses the problem as one of engagement and proposes a solution: gamification, more frequent push notifications, and a system of points redeemable for rewards. Let us apply the complete behavioral analysis before accepting that diagnosis.

7.1 Behavioral Diagnosis

The first error is one of framing (Module II): the team diagnosed an engagement problem when the real problem is one of behavioral design. The platform was designed assuming a rational user who, on receiving personalized information about their health, will act accordingly. But we know that information does not produce behavior—choice architecture does.

Bias by bias: Status quo bias operates in favor of not using the platform—the default is the employee's existing routine, and the platform requires active effort to change it. Loss aversion magnifies the immediate costs (time, effort, discomfort) while minimizing the future benefits (health, wellbeing). Mental accounting separates ‘work time’ from ‘wellness time,’ and the former always wins. Temporal discounting makes the benefits—which materialize in weeks or months—psychologically invisible against the costs, which are paid now.

“The average employee receives the HealthPath notification at 9:15 in the morning, exactly when he is going through his email. The doer self, activated by the day's urgencies, dismisses the notification in less than a second. The planner self—who enrolled in the platform with enthusiasm—takes no part in that decision. The intention to exercise remains intact; the execution never occurs.”

7.2 A Redesign Based on Behavioral Economics

A redesign informed by behavioral economics would address the problem in a radically different way:

Change the default. Instead of requiring the employee to open the app and choose an activity, the system pre-schedules a 5-minute micro-activity in their work calendar. The default shifts from ‘do nothing’ to ‘do something minimal.’ The employee can cancel, but now cancelling requires an active action.

Reduce initial friction. Instead of asking the user to complete a 15-minute profile before starting, the platform begins with a single question: ‘What would you like to improve: energy, sleep, or stress?’ The complete profile is built gradually.

Reframe as loss avoidance. Instead of promising future health gains, the platform shows what the employee has already achieved and what they would lose by stopping: ‘You're on a 12-day streak. If you stop today, you lose the accumulated progress.’ Loss aversion works in favor of the desired behavior.

Create social commitment devices. Pair employees into wellness duos in which one person's inactivity affects the other's streak. Light social pressure—combined with reciprocity (Cialdini)—produces behavioral consistency that individual motivation does not achieve.

Offer immediate feedback. Close the temporal gap between action and reward: after each micro-activity, show an immediate measurable impact (‘your resting heart rate improved 2 bpm this week’) instead of promising long-term benefits.

7.3 Derived Design Principles

Principle 1: Design for the doer self, not for the planner self. The planner enrolls; the doer decides each day whether to participate. The design must work for the doer.

Principle 2: The default is the architect's most important decision. Most users will accept the default. Design the default that produces the best result for them.

Principle 3: Information is not intervention. Knowing what is good for you does not produce behavior. Only the architecture of the environment translates knowledge into action.

Principle 4: Every behavioral intervention has second-order effects. Gamification can produce extrinsic motivation that displaces intrinsic motivation (Module I). Nudges can infantilize the user. Commitment devices can generate resentment. Apply the systems thinking of Module III.

Principle 5: Transparency is the ethical line. Every nudge must be able to withstand being made visible if the user asks how the system works. A nudge that would not survive transparency is probably manipulation, not architecture.

8. Glossary of Key Terms

9. Final Reflection and Self-Assessment

Before moving on to the next topic (Incentives), verify that you can apply behavioral economics as a tool of analysis and design, not merely as vocabulary.

10. Bibliography and Recommended Readings

Essential Readings

• Thaler, R. (2015). Misbehaving: The Making of Behavioral Economics. W.W. Norton.

• Kahneman, D. & Tversky, A. (1979). “Prospect Theory: An Analysis of Decision under Risk.” Econometrica, 47(2), 263–292.

• Thaler, R. & Sunstein, C. (2008). Nudge: Improving Decisions About Health, Wealth, and Happiness. Yale University Press.

Complementary Readings

• Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux. [Chapters 25–34 on Prospect Theory and behavioral economics]

• Ariely, D. (2008). Predictably Irrational. HarperCollins.

• Sunstein, C. (2016). The Ethics of Influence: Government in the Age of Behavioral Science. Cambridge University Press.

• Gigerenzer, G. (2014). Risk Savvy: How to Make Good Decisions. Viking. [A critical contrast with the biases approach]

Next Topic: The next book of this module addresses Incentives: the rewards and penalties that translate analysis into action. You will study with Cialdini, Thaler, and Sunstein how extrinsic incentives can displace intrinsic motivation (connecting with Module I), how social incentives frequently outweigh material ones, how temporal discounting sabotages long-term incentives, and how perverse incentives produce exactly the opposite of what they sought—the cobra effect you already studied in Module III, transformed into a design principle.