E-Book, Englisch, 315 Seiten
Reihe: PublishDrive
Cottrell Behavioral Economics in the Agentic AI Era
1. Auflage 2026
ISBN: 978-1-972752-87-6
Verlag: PublishDrive
Format: EPUB
Kopierschutz: 0 - No protection
A Plain English Guide to Agentic Decision Science, Choice Architecture, and How Algorithms Shape Human Choice
E-Book, Englisch, 315 Seiten
Reihe: PublishDrive
ISBN: 978-1-972752-87-6
Verlag: PublishDrive
Format: EPUB
Kopierschutz: 0 - No protection
AI systems are now shaping your choices about what to buy, watch, and believe, often without you noticing. This book explains modern decision science and choice architecture in plain English, so you can understand and protect your own decisions.
Behavioral Economics is a plain-English guide to what decision science has learned in the years since the classic work of Kahneman and Tversky - and to what daily choice looks like in a world where a map, a shopping site, a calendar tool, an inbox filter, and a chatbot are all quietly reshaping the menu before you ever see it. Written for the curious professional, the manager, the clinician, the product designer, the policy analyst, and any reader who lives inside AI-mediated decisions, the book replaces the tired 'your brain is broken, learn these tricks' framing with a more useful one: decisions happen inside environments, and modern research now studies those environments as carefully as it studies the mind. Recurring guides anchor every chapter - Vera, a 47-year-old chief resident in a Chicago emergency department who decides before certainty arrives; Marcus, whose work sits between recommendation engines and human judgment; and additional characters whose settings make certain shortcuts sensible and others dangerous. Chapters walk through why decision science matters now, what came after Kahneman (heuristics, biases, and the modern extensions that separate noise from bias), the neuroscience of a choice, how AI recommendations and conversational systems rewire the shape of a decision, choice architecture in the AI age, nudges that worked and ones that backfired, why committees get it wrong, and how to design better decisions in ordinary work. The recurring discipline: decision science is a collection of methods and findings, not a single doctrine - some results have been repeated across settings, some are useful but conditional, and some famous claims weakened when larger studies tried to reproduce them. This book says which is which. Not investment, legal, medical, or professional advice.
Autoren/Hrsg.
Weitere Infos & Material
Why Decision Science Matters Now
At 2:07 a.m., the emergency department in Chicago has the particular brightness of a place that refuses to admit it is night. Vera, the 47-year-old chief resident, stands beside a curtain pulled halfway around a chest-pain patient. The monitor shows a steady pulse. The first electrocardiogram, the EKG, is not normal enough to dismiss and not clear enough to settle anything. A nurse is waiting for an order. A family member is asking whether this is a heart attack. On Vera’s workstation, the triage system gives a soft chime and recommends elevated cardiac risk; consider rapid escalation.
Vera has seen alerts that mattered and alerts that did not. The patient’s story, medications, vital signs, and earlier tracing all compete for attention. She knows the system saw patterns across many cases. She also knows it cannot feel the thin, tense silence around a bed or notice every ambiguity in a hurried history. She does not treat the recommendation as a verdict. She treats it as one more piece of evidence, then asks what pushed it upward, what might be missing, and what harm follows if she is wrong in either direction. Within minutes she chooses a cautious path, documents why, and moves to the next decision waiting down the hall.
That moment is not a story about a heroic human defeating a machine, or an infallible machine correcting a human. It is a story about the new normal. A recommendation is now present inside a decision that used to be described as an individual act of judgment. The central question of this book begins there: what happens when advice systems become part of the loop through which people see, interpret, and choose?
The New Room Where Choices Happen
For most of human history, people made choices among options that were physically nearby, socially available, or remembered. The environment mattered. A store put one product at eye level. A form placed one answer first. A colleague spoke before another colleague. These were forms of choice architecture: arranging a decision to affect behavior without formally removing options. They mattered, but usually locally—one clerk, one room, one form, one meeting.
Digital systems changed the scale. A search engine can rank a universe of pages in a fraction of a second. A recommendation service can assemble a different storefront for each user. A work platform can surface a suggested task, a default time, or an automatically written summary. Large language models add something new. They do not merely rank items. They converse, reframe questions, propose plans, and make an answer feel tailored to the person asking.
The practical effect is not that choice has disappeared. Instead, the visible choice set is increasingly produced before a person starts comparing. When the first three results answer a question, the tenth result may as well be in another town. When a chatbot gives two plausible actions, the person may not think to ask for a third. When a calendar assistant offers the next open slot, the offered time becomes a quiet default. The decision begins earlier than the click.
This is why familiar advice such as “slow down and think” is not enough. Slowing down can help, especially when stakes are high, and evidence is available. But it does not tell you whether the right information was shown, whether the alternatives were framed honestly, whether a model is calibrated for this setting, or whether a user has a realistic chance to question the tool. Better decision design looks at the whole loop, not only at the person at its center.
Vera’s triage alert illustrates the point. The alert may reduce one kind of error by reminding a tired clinician of a risk pattern. It may also create anchoring, the tendency to rely too heavily on the first number or label that appears. It may increase vigilance on a busy night and cause alert fatigue the next. Its value depends on the task, its accuracy, the way it is displayed, the workflow around it, and the clinician’s ability to inspect it. There is no single answer called “AI is good at decisions.”
A useful mental model is that decision quality has several ingredients. There is the evidence available. There is the method used to weigh it. There are the constraints of time, attention, and emotion. The environment makes some actions easy and others burdensome. And now there is often an automated adviser. If any one ingredient changes, the same person can make a different choice. That is not a character defect. It is a feature of situated human reasoning.
An Inflection Point, Not a Fad
Why call the years from 2024 through 2026 an inflection point? Not because human psychology suddenly changed. Loss aversion didn't arrive with chatbots, and people didn't become suggestible only after software learned to write. The inflection is environmental. Generative systems moved from a specialist tool to an ordinary interface for searching, drafting, planning, explaining, and deciding. Their advice is not tucked away in a report. It appears in the same conversational space where people form questions. The strength of that shift varies by domain. In consumer search and drafting, the change is already visible in measured user behavior; in regulated clinical, legal, and financial decisions, it remains uneven and conditional on institutional adoption, evidence, and oversight.
That matters because language does more than carry information. It sets a frame. A question phrased as “Which option should I choose?” invites a recommendation. A question phrased as “What tradeoffs should I inspect?” invites comparison. A response that begins with a confident answer can anchor a reader before the caveats arrive. A response that lists uncertainties can leave a reader feeling lost, even when the evidence is good. Conversational systems have a new power to shape not just answers but the structure of deliberation.
Research on large language models added a second reason to pay attention. Studies comparing models with human behavior on classic reasoning tasks found overlaps, gaps, and sharp failures that were more interesting than the simple claim that a model is smart or not smart. In some experiments, models show patterns resembling familiar human errors. In others, a small wording change produces a striking change in the response. In still others, a system produces a correct answer with an explanation that should not be treated as a transparent record of its internal process.
Work by researchers such as Eric Schulz and Marcel Binz, whose 2023 PNAS paper "Using cognitive psychology to understand GPT-3" is a foundational reference, helped make this comparison more disciplined. Instead of treating a model's answer as a magic performance, researchers ask what pattern of choices it produces across a controlled set of problems. Does it generalize? Does it change its response when the framing changes? Does it rely on a shallow cue? Does it show an apparent strategy only because the prompt makes that strategy easy to imitate? These questions bring cognitive science methods into contact with systems people increasingly use as advisers.
This is not merely a laboratory topic. A manager can ask an assistant to summarize candidate interviews. A clinician can encounter a risk score. A customer can receive a conversational recommendation about a product. A policy analyst can use a system to organize public comments. In each case, a tool may save genuine time. In each case, the person also faces a new task: deciding how much weight to give an output that is fluent, fast, and sometimes persuasively wrong.
The old mental picture placed technology outside the decision-maker. A calculator was a tool. A database was a tool. A spreadsheet was a tool. Those labels remain true, but they are incomplete for systems that propose a next sentence, a next diagnosis, a next purchase, or a next action. When a tool suggests rather than merely retrieves, it becomes part of the choice architecture. It deserves the same design and evidence standards we would apply to any other influence on a consequential decision.
The inflection is also institutional. Organizations are building AI into workflows before they have agreed on standards for verification, logging, escalation, and appeal. A person may be told to use a system but not told when to override it. A product team may optimize a recommendation for clicks without agreeing on what a satisfied user looks like a month later. A public agency may automate a screening step without a clear plan for explaining it to affected people. Decision science is useful here because it asks practical questions before philosophical ones.
A Suggestion Is an Intervention
A recommendation feels harmless because it lets a person say no. Freedom matters, but it is not the only fact that matters. A suggestion can alter attention, confidence, memory, and the perceived range of alternatives. It can make one action feel normal and another action feel unusual. It can make a decision feel already made. These effects can be useful or harmful, which is why “it was only a suggestion” is too weak a defense and too crude a criticism.
Consider a streaming screen that starts playing a new program without a clear pause. The system has not forced anyone to watch. Still, it has made continued watching easier than stopping, and it has turned a fresh choice into a default path. Consider a subscription page that features a bright button to join and hides cancellation behind several ambiguous menus. Again, the person technically retains a choice. The architecture has raised the effort required to exercise it.
The same logic applies in professional settings. A risk score hidden until after a clinician records an initial assessment influences...




