2026-03-24 11:26

Status: active

Tags: [[3 - Tags/books|books]] [[3 - Tags/decision-making|decision-making]] [[3 - Tags/productivity|productivity]]

Algorithms to Live By - Key Takeaways

Authors: Brian Christian & Tom Griffiths Subtitle: The Computer Science of Human Decisions Source: Book + ToSummarise.com review

Summary

Computer science algorithms aren’t just for machines. They offer practical frameworks for everyday decisions, from when to stop searching and commit, to how to organize your life. The key insight: having an optimal process doesn’t guarantee the best outcome, but it gives you the best odds.

Core Principles

1. The 37% Rule (Optimal Stopping)

When making a decision with sequential options (hiring, apartment hunting, dating):

Explore the first 37% of options without committing
Then immediately commit to the next option that beats all previous ones
Success rate: ~37% (sounds low, but it’s mathematically optimal)

Real-world application: If you plan to interview 10 candidates, interview the first 4 without hiring, then hire the next one who’s better than all 4.

2. Explore vs. Exploit Trade-off

Life balances novelty and tradition. The optimal strategy depends on your time horizon.

Life Stage	Strategy	Why
Early career/youth	Explore heavily	Information value is high, time to use it
Later career/older	Exploit your discoveries	Less time to benefit from new info

Key insight: Older people report higher happiness because they’ve shifted to exploitation mode, enjoying what they’ve found works for them.

3. Sorting vs. Searching Trade-off

Don’t organize everything. Consider:

How often will you search for it?
How costly is each search?
Can technology search for you? (Email search makes inbox sorting less valuable)

Rule: Sort things you’ll search frequently. Leave rarely-accessed items unsorted.

4. Caching (LRU Algorithm)

The “Least Recently Used” principle is optimal for organizing:

Keep accessible: Items you’ve used recently
Archive/discard: Items untouched for a long time

That pile of clothes by your bed? Actually an efficient cache. Most-used items naturally float to top.

5. Context Switching Costs

Every task switch has overhead. Too much switching = “thrashing” where nothing gets done.

Solutions:

Set minimum time blocks for focused work
Practice “interrupt coalescing” (batch email checks, office hours)
Accept less responsiveness for higher throughput

6. When Queues Overflow, Drop Things

When your to-do list is too long, the answer isn’t “work faster.” It’s to close the queue and drop items.

“The problem isn’t that we’re always connected; we’re not. The problem is that we’re always buffered.”

Permission granted: Say “no” when overloaded. It’s a feedback mechanism.

Decision-Making Frameworks

Relaxing Intractable Problems

When a problem is too complex, make it simpler:

Drop constraints - Remove requirements that aren’t essential
Accept “good enough” - A 90% solution in 1% of the time may be worth it
Use randomness - Random sampling can break local maxima

Computational Kindness

Make others’ decisions easier:

Bad: “What time works for you?” (forces them to search their calendar)
Good: “Does Tuesday 2pm or Thursday 10am work?” (they just verify)
Bad: “I’m flexible on dinner” (hides your preferences)
Good: “I’d love Thai or Italian” (constrains the search space)

Bayes in Practice

Combine prior beliefs with new evidence. Different prediction rules apply based on distribution type:

Normal distribution (height, test scores): Predict average
Power law (wealth, city sizes): Predict multiplicatively higher
Erlang distribution (time until events): Use the memoryless property

Counterintuitive Insights

Humans are surprisingly good at solving hard CS problems intuitively
Simple models often beat complex ones (overfitting is real)
Randomness helps - Being willing to accept temporarily worse outcomes can escape local maxima
Having the best process ≠ best outcome - The optimal hiring strategy still fails 63% of the time

Memorable Examples

Sequels as exploitation: Hollywood making more sequels signals they believe the industry’s end is near
Clinical trials: “Adaptive trials” could save more patients by shifting allocation to treatments showing promise
Donald Knuth: The legendary programmer has no email, checks postal mail quarterly
Soccer randomness: A 3-2 win only gives 63% confidence the better team won

Action Items

Apply 37% rule to next major sequential decision
Audit current task switching habits - set minimum focus blocks
Practice computational kindness in scheduling
Review to-do list - what should be dropped, not done?
Organize by LRU principle - most-used items most accessible

How to Decide by Annie Duke
Noise by Kahneman, Sibony & Sunstein
Thinking, Fast and Slow by Daniel Kahneman