The key ideas

Algorithms to Live By: The Computer Science of Human Decisions

Brian Christian, Tom Griffiths, 2016

You have to take the apartment today, and you have no idea whether it beats the others. Computers face problems like this every day, and many of them have an exact answer. Brian Christian and Tom Griffiths carry solutions from computer science into everyday life: when to stop searching, what to keep close at hand, how to start your day and how to stop blaming yourself for a bad outcome.

Who it is for

For anyone choosing an apartment, a job or a place to eat tonight who keeps wondering whether it is too soon to stop looking. And for anyone curious about what computers can teach people.

Even the best strategy sometimes yields bad results. That is why computer scientists take care to distinguish between process and outcome.

The apartment you have to take today

San Francisco, apartment hunting. Good places here are gone within hours: a crowd turns up for the viewing, and the landlord rents to the first person ready to sign. You cannot sleep on it. If you walk away to see a couple more, someone else will take this one. But how do you know it is the best when you have nothing yet to compare it with?

The writer Brian Christian and the cognitive psychologist Tom Griffiths noticed that this problem has an exact answer. It was found not by psychologists but by mathematicians and people who work in computer science. A computer runs on algorithms. An algorithm is an exact sequence of steps. And a computer faces the same problems people do: when to stop searching, what to keep close at hand, what to do first and how to act when the future is unknown.

The point of the book is not to live like a machine. Many of our problems are simply hard, and there is no perfect answer to them, not even for a computer. Computer science shows how to act sensibly when perfection is out of reach, and how to stop blaming yourself for not finding it.

Nine such problems lie ahead. You will learn when to stop looking and when it pays to try something new. You will see why a pile of papers on a desk can be smarter than a tidy cabinet, and why long calculations sometimes lose to a simple rule. And at the end you will see how one polite phrase can quietly put a burden on someone else.

Look first, then choose

Mathematicians have the secretary problem: candidates arrive one at a time, a rejection is final, and you cannot go back. Stop too early, and you miss the best one, who has not shown up yet. Wait too long, and the best one is already gone. The solution is called the 37 percent rule: for the first third or so, only look, without choosing anyone. After that, take the first one who beats everyone you have seen so far. If you have a month to find an apartment, the first eleven days go to getting to know the market.

Michael Trick, then a graduate student and later a professor who studies the mathematics of decisions, applied the rule to love. He decided he was looking for a partner between the ages of 18 and 40 and worked out his stopping point: just past 26. He was 26 at the time. He met a woman who was better than everyone before her, and he proposed. She said no.

After his first wife died, Johannes Kepler courted eleven women, one after another. Then he went back to the fifth, Susanna Reuttinger, and she said yes. Both stories show where the rule is simpler than real life. In the classic problem you are never turned down, and you can never go back. If you get turned down half the time, then by the authors' math you should propose sooner, right after the first quarter of your search.

And one honest number. Even the best strategy finds the very best candidate only 37 percent of the time, and misses the rest. A bad outcome does not prove the search was badly run.

Read in Livria

About the author

Brian Christian studied computer science and philosophy and writes about how people and machines make decisions. Tom Griffiths, a cognitive psychologist, was a professor at the University of California, Berkeley when the book came out, studying how people solve hard problems.

All books by this author: Brian Christian, Tom Griffiths

Terms from the book

  • Stanford
  • prisoner's dilemma
  • Darwin
  • Bayes
  • Monte Carlo
  • Copernicus
  • Nash
  • overfitting
  • Algorithm
  • Kepler
  • Computer science: The science of how information is stored, transmitted and processed, including by computers.
  • Secretary problem: A mathematical problem about choosing the best of several candidates who arrive one at a time, with no way back to earlier ones. Its answer is known as the 37 percent rule.
  • Trick: Michael Trick, an American operations researcher. He applied the 37 percent rule to looking for a partner.
  • Reuttinger: Susanna Reuttinger, Kepler's second wife.
  • Multi-armed bandit: A computer science problem about choosing between slot machines with unknown payoffs. An image for choosing between the new and the familiar.
  • Carstensen: Laura Carstensen, an American psychologist at Stanford University. She studies how people's goals and relationships change with age.
  • Cache: A small, fast memory in a computer that holds what is likely to be needed soon.
  • Noguchi: Yukio Noguchi, a Japanese economist. He invented a filing system in which a folder goes back to the left end of the box after use.
  • Ramscar: Michael Ramscar, a researcher of language and memory. He suggested that memory slows with age partly because the amount of knowledge grows.
  • Mars Pathfinder: An American spacecraft that landed on Mars in 1997 and delivered a small rover there.
  • Priority inversion: A failure in which an important task waits for a less important one to finish. It happened on Mars Pathfinder.
  • Thrashing: A state in which a computer spends almost all its time switching between tasks instead of working on them.
  • Knuth: Donald Knuth, an American computer scientist, author of the multivolume work The Art of Computer Programming.
  • Gott: J. Richard Gott, an American astrophysicist at Princeton, known for predicting how long things will last using the Copernican Principle.
  • Tenenbaum: Joshua Tenenbaum, an American cognitive psychologist at the Massachusetts Institute of Technology, Griffiths's research partner.
  • Kidd: Celeste Kidd, an American psychologist. She showed that waiting in the marshmallow test depends on whether adults keep their promises.
  • Marshmallow test: A famous experiment: a child can eat a treat right away or wait and get two.
  • Grossman: Dave Grossman, a former American soldier who taught psychology at the West Point military academy, author of books on how people behave in combat.
  • Ulam: Stanislaw Ulam (1909 to 1984), a Polish-born American mathematician, one of the creators of the Monte Carlo method.
  • Los Alamos: A laboratory in New Mexico where the atomic bomb was built during the Second World War.
  • ALOHAnet: One of the first computer networks to use radio, built in Hawaii in the early 1970s.
  • Exponential backoff: A rule by which the wait before the next attempt grows after each failure, usually doubling.
  • Bavelas: Janet Bavelas, a Canadian psychologist who studies how a listener shapes the storyteller.
  • Price of anarchy: A measure of how much worse the outcome is when everyone acts for themselves, compared with the best shared outcome.
  • Travis CI: A German company offering a service for testing software code. It replaced unlimited vacation with a mandatory minimum.
  • Computational kindness: The authors' term for the habit of taking on part of another person's mental work, for example by proposing a specific meeting time right away.

All terms

Similar books