3 ways to make better decisions -- by thinking like a computer | Tom Griffiths

Episode Summary

Executive Summary: Tom Griffiths argues that many everyday decisions—from finding housing to choosing restaurants or organizing paperwork—have computational structures that computer science can solve well. Using optimal stopping, explore-exploit trade-offs, and memory-optimization principles, he shows that rational decision-making often means accepting uncertainty, settling for good-enough outcomes, and following the best process rather than chasing perfect results.

Main Topics: Optimal stopping and the 37% rule (Priority: 5/5): Griffiths explains that the best strategy for choosing a home is to spend 37% of the search period observing options, then choose the first later option better than all seen so far. Explore-exploit trade-off in everyday life (Priority: 5/5): He frames repeated choices like restaurants, music, and social time as a balance between trying new options and sticking with known good ones. Applying computer science to human decisions (Priority: 4/5): The talk argues that difficult personal decisions can be understood through computational models, and that human behavior often mirrors efficient algorithms. Memory management as a model for organization (Priority: 4/5): Computer memory systems illustrate why recently used items should be most accessible, offering a framework for organizing wardrobes, desks, and files. Good-enough solutions and limited capacity (Priority: 5/5): He emphasizes that many problems are too hard for exhaustive optimization, so people should use approximations, simplifications, and accept imperfect outcomes. Emotional relief through process-based rationality (Priority: 4/5): Griffiths suggests that understanding algorithmic limits can reduce self-blame and help people feel more comfortable with uncertainty and compromise.

Key Arguments: The 37% rule is an optimal stopping strategy that maximizes the chance of choosing the best option in a finite search. The explore-exploit trade-off explains why sometimes novelty is rational and other times sticking with known options is better. Babies’ tendency to try everything is not irrational; it fits the logic of exploration when future opportunities are abundant. Older adults repeatedly choosing familiar restaurants are often behaving optimally because exploitation becomes more valuable when the future is shorter. The least-recently-used principle from computer memory is a useful model for deciding what to keep accessible in wardrobes and offices. Messy piles and simple filing systems can be efficient because they preserve access to the most recently used information. For many real-world problems, the best strategy is not perfection but a practical process that acknowledges constraints and uncertainty. Rationality should be judged by process, not guaranteed outcomes; even optimal strategies can fail much of the time.

Data Points: Optimal stopping proportion: 37% - Share of a search period to spend observing before making a choice, used in the housing example. Example search duration: 11 days - If searching for a month, 37% is approximately 11 days to set a standard before deciding. Talk recording: TEDxSydney 2017 - The event where Tom Griffiths delivered the talk live. Explore-exploit history: Over the last 60 years - Computer scientists’ progress in studying the explore-exploit trade-off.

Pivotal Quotes: "37%." — Tom Griffiths: The central rule presented for maximizing the chance of finding the best home. "You can't control outcomes, just processes." — Tom Griffiths: Conclusion of the talk, summarizing his view of rational decision-making. "These aren't the concessions that we make when we can't be rational." — Tom Griffiths: He explains that taking chances or settling for good-enough options can be rational, not a failure of rationality.

Implications: Listeners can use simple algorithmic heuristics to make better choices under uncertainty, reduce decision fatigue, and accept that rationality often means following an efficient process rather than chasing perfect outcomes.