Chapter 12 Exercises

How to use these exercises: Work through the parts in order. Part A builds recognition skills, Part B develops analysis, Part C applies concepts to your own domain, Part D requires synthesis across multiple ideas, Part E stretches into advanced territory, and Part M provides interleaved practice that mixes skills from all levels.

For self-study, aim to complete at least Parts A and B. For a course, your instructor will assign specific sections. For the Deep Dive path, do everything.

Part A: Pattern Recognition

These exercises develop the fundamental skill of recognizing satisficing -- and distinguishing it from optimization -- across domains.

A1. For each of the following scenarios, identify whether the decision-maker is satisficing or attempting to optimize. Explain what the satisficing threshold or optimization criterion is in each case.

a) A hiring manager interviews candidates until she finds one who meets all the requirements listed in the job description, then makes an offer without interviewing the remaining applicants.

b) A student researches every available section of a required course -- comparing professors' ratings, reading syllabi, checking room locations and time slots -- before enrolling in the one with the highest composite score.

c) A driver looking for parking circles the block once, sees an open spot two blocks from the destination, and takes it.

d) A homebuyer spends eighteen months viewing 73 houses, creates a weighted scoring rubric with 15 criteria, and buys the house with the highest score.

e) A doctor in an emergency room diagnoses a patient with classic heart attack symptoms and begins treatment immediately, without ordering additional tests to rule out every alternative diagnosis.

A2. The chapter identifies satisficing in evolution, chess engines, military strategy, engineering, consumer behavior, expert decision-making, and fast-and-frugal heuristics. For each of the following domains, describe a satisficing strategy that operates there and identify what replaces optimization:

a) How a bird selects a mate.

b) How a startup selects its first product.

c) How a reader selects their next book.

d) How a compiler selects machine code instructions.

e) How a plant allocates resources between root growth and leaf growth.

A3. Classify each of the following as an example of bounded rationality, computational intractability, ecological rationality, or the paradox of choice. Some may involve more than one concept.

a) A protein-folding problem with 100 amino acids that would take longer than the age of the universe to solve by brute force.

b) A consumer who feels overwhelmed and anxious when choosing among 50 health insurance plans and ultimately picks the plan recommended by a friend.

c) A forager who uses the simple rule "eat the largest fruit you can find" in an environment where fruit size is correlated with caloric content.

d) A chess grandmaster who sees the right move instantly but cannot explain why -- she "just knows."

e) A student choosing a college who is less satisfied with her choice after visiting 12 campuses than a friend who visited 3.

A4. Herbert Simon argued that satisficing is not a failure of rationality but an adaptive strategy. For each of the following "cognitive biases" from the behavioral economics literature, argue whether it could be reframed as ecological rationality -- a satisficing heuristic that works well in natural environments even if it fails in laboratory settings.

a) The status quo bias (preferring to keep things as they are rather than changing).

b) The availability heuristic (judging the probability of an event based on how easily examples come to mind).

c) Loss aversion (weighing losses more heavily than equivalent gains).

d) The anchoring effect (being influenced by an initial reference point, even an arbitrary one).

A5. Identify the "tolerance" -- the range of acceptable variation -- in each of the following:

a) A recipe that calls for "a pinch of salt."

b) A bus schedule that lists arrival times "every 10-15 minutes."

c) A grading rubric that awards an A for "90% or above."

d) A marathon runner whose goal is to finish "under four hours."

e) A thermostat set to 72 degrees Fahrenheit with a 2-degree deadband.

Part B: Analysis

These exercises require deeper analysis of satisficing concepts.

B1. The Cost of Optimization. Consider the following scenario: You are furnishing a new apartment and need to buy a couch. There are approximately 500 couches available within your budget from local stores and online retailers.

a) List the dimensions you would ideally evaluate (comfort, size, color, material, durability, price, delivery time, return policy, aesthetics, etc.). How many dimensions did you identify?

b) Estimate the time required to evaluate all 500 couches on all your dimensions. Be realistic about travel time, testing time, and comparison time.

c) What satisficing strategy would you actually use? Describe your threshold and your search process.

d) Calculate the implicit "hourly wage" of your time spent optimizing. If spending an additional 10 hours of research could save you $50 on the couch, is the optimization worth it? What if it could improve comfort by 5%?

B2. Satisficing and Robustness. The chapter argues that satisficing produces more robust solutions than optimization.

a) Define robustness in your own words. How does it differ from optimality?

b) Using the engineering tolerance example, explain why a part designed to a tolerance of plus or minus 0.5 mm is more robust than a part designed to exactly 10.000 mm. What kinds of perturbations can the toleranced part withstand that the "optimized" part cannot?

c) Extend this reasoning to military strategy. Why is a plan that specifies "seize the high ground by noon" more robust than a plan that specifies "advance platoon A along route alpha at 0800, platoon B along route beta at 0815, commence shelling at 0830, assault the position at 0845"?

d) Can you think of a situation where robustness and optimality are aligned -- where the most robust solution is also the optimal one? If so, what features of that situation make this alignment possible?

B3. Gigerenzer vs. Kahneman. The chapter notes a tension between Gigerenzer's view (heuristics are ecologically rational tools) and the heuristics-and-biases tradition associated with Kahneman and Tversky (heuristics are sources of systematic error).

a) Describe a scenario where the same heuristic would be classified as "bias" by Kahneman and "ecological rationality" by Gigerenzer. What determines which label is appropriate?

b) Can both perspectives be correct simultaneously? If so, what does each perspective contribute that the other misses?

c) How does the concept of the decision environment help reconcile the two views?

d) If you had to advise a medical school on how to train future physicians to make better diagnostic decisions, would you draw more on Kahneman's framework or Gigerenzer's? Why?

B4. Maximizers, Satisficers, and Well-being. Schwartz's research shows that satisficers are generally happier than maximizers, even though maximizers often make objectively "better" choices.

a) Explain this paradox. How can someone who makes better choices be less happy with those choices?

b) The chapter connects maximizing to over-exploration (Chapter 8). Develop this connection further: what specific costs does a maximizer incur that a satisficer avoids?

c) Could a "meta-satisficer" -- someone who satisfices about whether to satisfice or maximize -- outperform both pure satisficers and pure maximizers? What would this look like in practice?

d) Social media often exposes people to the choices that others have made. How might social media shift people from satisficing toward maximizing, and what consequences would you predict for well-being?

B5. Recognition-Primed Decision Making and Expertise. Klein's research shows that experts satisfice rather than optimize.

a) Explain why this finding is counterintuitive. What does the popular conception of expertise assume about how experts make decisions?

b) How does the concept of "recognition-primed decision making" relate to Bayesian reasoning (Chapter 10)? What serves as the prior in an expert's rapid assessment?

c) If experts satisfice, does this mean that expertise is not about "knowing more" but about "needing to compare less"? Defend or challenge this claim.

d) What are the failure modes of recognition-primed decision making? When does the expert's pattern-matching lead to error?

Part C: Application

These exercises ask you to apply satisficing concepts to your own experience and context.

C1. Choose three decisions you made in the past week -- one trivial, one moderate, and one significant. For each:

a) Describe the decision and what you chose.

b) Did you satisfice or maximize? How do you know?

c) How many alternatives did you consider?

d) What was your implicit threshold of acceptability?

e) In retrospect, would you have been better served by satisficing more (if you maximized) or maximizing more (if you satisficed)?

C2. Identify a domain in which you are an expert (or at least highly experienced). Describe a recent decision in that domain:

a) Did you use something resembling recognition-primed decision making? Did you generate a single option and evaluate it, or did you generate multiple options and compare?

b) How long did the decision take?

c) Would a novice in the same situation have made the decision differently? How?

d) Can you identify the pattern library that enabled your rapid decision? What patterns are in it, and where did they come from?

C3. The chapter argues that culture pressures us toward maximizing. Identify three areas of your life where you feel pressure to optimize (find the best option) rather than satisfice (find a good-enough option).

a) Where does this pressure come from? (Social media, peer comparison, advertising, professional norms?)

b) What would it look like to satisfice in each of these areas instead?

c) What would you gain (time, peace of mind, reduced decision fatigue) and what would you risk losing?

C4. Design a satisficing protocol for a decision you frequently face in your professional life. Specify:

a) The decision (e.g., which vendor to choose, how to prioritize tasks, which candidate to hire).

b) The threshold criteria -- what must be true for an option to be "good enough."

c) The search procedure -- how you will generate and evaluate candidates.

d) The stopping rule -- when you will stop searching and commit.

Part D: Synthesis

These exercises require integrating satisficing with concepts from multiple chapters.

D1. Satisficing and Gradient Descent (Chapter 7). The chapter draws a parallel between satisficing and accepting a local optimum in gradient descent.

a) In gradient descent, what determines whether an agent stops at a local optimum or continues searching? How does this map onto the satisficing threshold?

b) Simulated annealing (Chapter 7) gradually reduces the probability of accepting worse solutions over time, transitioning from exploration to exploitation. How is this similar to the satisficing strategy of starting with a low threshold and raising it as experience accumulates?

c) In what sense is overfitting (fitting the global optimum of the training data too precisely) a failure to satisfice? How does early stopping in machine learning function as a satisficing strategy?

D2. Satisficing and the Explore/Exploit Tradeoff (Chapter 8). The chapter argues that satisficers exploit early while maximizers over-explore.

a) In the multi-armed bandit framework, what corresponds to the satisficing threshold? How does a satisficing strategy differ from an epsilon-greedy strategy or an Upper Confidence Bound strategy?

b) The chapter notes that the optimal explore/exploit balance shifts over time -- early on, exploration is more valuable; later, exploitation is more efficient. How does this map onto life decisions? Is it more rational to be a maximizer when young and a satisficer when old? Why or why not?

c) Can you design a "satisficing bandit" algorithm that uses a threshold rather than a comparison criterion? How would it perform compared to standard bandit algorithms?

D3. Satisficing and Bayesian Reasoning (Chapter 10). The chapter connects recognition-primed decision making to Bayesian priors.

a) In the Bayesian framework, satisficing corresponds to accepting a posterior probability above a certain threshold rather than computing the exact posterior. In what situations is this a good approximation, and in what situations does it fail?

b) How does the strength of an expert's prior affect the speed of satisficing? Why can an experienced doctor satisfice faster than a medical student?

c) The chapter argues that Gigerenzer's fast-and-frugal heuristics and Bayesian reasoning are not competing frameworks but tools for different environments. What features of the environment determine which tool is more appropriate?

D4. Satisficing and Power Laws (Chapter 4). The chapter connects the 80/20 rule to the economics of satisficing.

a) In a power-law distributed environment, the gap between the best option and the second-best option is typically large, while the gap between the tenth-best and the twentieth-best is small. How does this distribution shape the returns to satisficing vs. maximizing?

b) In a Gaussian (normally distributed) environment, most options cluster near the mean and the extremes are rare. How would the satisficing vs. maximizing tradeoff differ in a Gaussian environment compared to a power-law environment?

c) Given that many real-world domains are power-law distributed (Chapter 4), what does this imply about the general value of satisficing as a strategy?

Part E: Extension

These exercises push beyond the chapter's content into more advanced territory.

E1. Computational Complexity and Satisficing. The chapter mentions that many optimization problems are NP-hard.

a) Explain, in non-technical language, what it means for a problem to be NP-hard. Why does NP-hardness make optimization impractical for large problem instances?

b) Approximation algorithms guarantee a solution within a certain factor of the optimum (e.g., within 1.5 times the optimal cost for the metric traveling salesman problem). How do approximation algorithms relate to satisficing? Is "within 1.5 times optimal" a satisficing threshold?

c) Some researchers argue that P vs. NP (the question of whether NP-hard problems have efficient solutions) is the mathematical formalization of the satisficing insight -- that for many problems, finding the exact best answer is fundamentally harder than finding a good-enough answer. Evaluate this argument.

E2. Satisficing in Institutional Design. Simon's original work focused on decision-making within organizations.

a) How does bureaucracy function as an institutional satisficing mechanism? What role do standard operating procedures, rules of thumb, and default options play?

b) When does institutional satisficing become pathological? Can you identify examples where organizations satisficed in ways that produced bad outcomes -- where the threshold was set too low?

c) How could an organization design its decision processes to satisfice well -- setting appropriate thresholds, using effective heuristics, and avoiding both over-optimization and under-satisficing?

E3. Satisficing and Artificial Intelligence. Modern AI systems face a version of the satisficing dilemma.

a) Large language models generate text by sampling from a probability distribution over next tokens. In what sense is this a satisficing strategy rather than an optimizing one? What would "optimization" look like in language generation, and why would it fail?

b) The "alignment problem" in AI -- ensuring that AI systems pursue human-intended goals -- is often framed as an optimization problem (maximize some objective function). How does Goodhart's Law (Warning 2 in Section 12.10) apply to this framing? What would a satisficing approach to AI alignment look like?

c) AlphaZero satisfices in chess by using neural network intuition to guide search. Could the same approach work for real-world decision-making? What features of chess make it amenable to this approach that real-world problems might lack?

Part M: Mixed Practice (Interleaved Review)

These problems deliberately mix concepts from Chapters 8, 10, and 12 to strengthen retrieval and transfer.

M1. A medical researcher is designing a clinical trial to test a new drug. She must decide how many patients to enroll, how long to run the trial, and what statistical threshold to use for declaring success.

a) From a Bayesian perspective (Chapter 10), how should the researcher's prior belief about the drug's effectiveness influence the trial design?

b) From a satisficing perspective (Chapter 12), how should the researcher decide when the trial has produced "enough" evidence? What determines the threshold?

c) From an explore/exploit perspective (Chapter 8), how should the researcher balance testing this drug (exploitation) against investigating alternative drugs (exploration)?

d) How do these three perspectives interact? Can they conflict?

M2. A venture capitalist must decide how many startups to evaluate before making an investment.

a) Using the explore/exploit framework (Chapter 8), analyze the tradeoff between seeing more startups (exploration) and investing in a promising one (exploitation).

b) Using the satisficing framework (Chapter 12), define what a "good enough" startup looks like. What threshold criteria would you set?

c) Using the Bayesian framework (Chapter 10), how should the venture capitalist update her beliefs about the quality of available startups as she sees more of them?

d) Schwartz's research suggests that maximizers are less happy with their choices. Does this apply to venture capitalists? Why might the VC profession attract maximizers, and what consequences might this have?

M3. An ant colony is searching for food sources (Chapter 8). Individual ants use simple pheromone-based heuristics (Chapter 12). The colony collectively converges on good food sources through a process that resembles Bayesian updating (Chapter 10).

a) Identify the satisficing heuristic that individual ants use. What is the threshold?

b) How does the colony's collective behavior resemble Bayesian updating? What serves as the prior, evidence, and posterior?

c) In what sense does the ant colony avoid the paradox of choice? Could an ant colony suffer from "too many food sources"?

d) Compare the ant colony's strategy to a human organization's strategy for finding good suppliers. Where do the parallels hold and where do they break down?

M4. A teacher must decide how much time to spend grading each student's essay. She has 30 essays and 6 hours.

a) What would optimization look like? (Spending exactly the right amount of time on each essay to produce the most accurate grade.)

b) What satisficing strategies might she use instead? Identify at least two.

c) How does this connect to the explore/exploit tradeoff? (Should she read each essay once quickly, or some essays carefully and some quickly?)

d) How could Bayesian reasoning help? (If the student's previous work was consistently excellent, how should this prior affect the time spent grading the current essay?)