Algorithms to Live By: The Computer Science of Human Decisions
by Christian, Brian
Published: April 19, 2016
Read: January 30, 2018
Review
As a CS major, this book offered some great connections between some ideas I had learned in class and life. While oftentimes the advice gleaned was fairly obvious or too simplified to implement, the insights provided a strong mathematical support for some common sense ideas as well as a new framework for analyzing our own lives. The book was well written with a good depth of research, yet the explanations of CS topics were concise and clear. This book showcased that efficient algorithms and game theory aren't just for computers. A must read for CS majors, and still an approachable good read for everyone else.
Notes
Dating, apartment hunt, search for parking
Humans face some of the hardest forms of the problems computers face. Face uncertainty and limited resources often.
Optimal Stopping
Secretaries problem, interview applications can only decide whether to hire or continue looking at candidates and only judgement is comparison between applications, zero info. Look then leap after 37% of pool to maximize chances of riding the one, 37% of finding them. Rejection and recall decrease and increase the look period. If you have some objective measure where you know the percentiles, then you just wait for a threshold. Burglar problem, risk it all for one more heist, optimal stop is roughly the chance of success divided by chance of failure.
Life is an optimal Stopping Problem
Every moment that pass when to speak when to lean in for the kiss. Is one that you can't get back. There is always a cost of search the time of your life.
EXPLORE EXPLOIT
Firms index solve explore expolate questiom. Calculate gittens index for each machine separately and choose the best one. Unknown casino machines have a high 7/10 value encouraging exploration. Explore early though you will be disappointed. Win stay is good, lose-shift not always ideal. Upper Confidence Bound, choose arm that could reasonably perform best in the future. People tend to over explore leaving something tht is clearly optimal.
But like a child who explores or an old man with the same routine explotating, you must think about the future and the potential reward as a new good thing can pay over many times and probabilities of good change over time so exploring being restless is a natural response on our restless world.
Sorting
Soccer has a lot of noise in hierarchy. Monkeys have a hierarchy in their mind and fight only when the hierarchy is unclear, i.e new members. Humane to limit group sizes? Notice marathons and benchmark, solve computational problems being efficient without head to head trusses. Gdp good measure despite all it's stuff. More logical beings have to tussle less.
Caching
Stacking papers on desk is actually pretty efficient, like least recently used policy.
People think memory is suboptimal. But perhaps memory is limitless, but search time is limited so we do least recently used in our minds. The learning curve.
Scheduling
Shortest Job First, Earliest Deadline, Throw out biggest item if can't finish. Interrupt coalescing
Bayes
Laplace's Law: w success in n attempts means expectation of success is w+1/n+2. Bayes. Just multiply probabilites to combine prexisting beliefs with observed evidence. Prior differ based on the distrubution. Power law, normal, or uniform. Power law means few megas far above average so expect those that have to get more. For normal, with no information a phenomem will go on ad long as it has gone on since you are most likely to find it in the middle. People's priors are pretty accurate and rich, we can mine it by asking people their expectation based on a given data point.
Overfitting
How hard should we think?
OVerfitting->CrossValidation->RegularizationMetrics can bastardize processes.More information and complexity can lead to worseee solutionsWhen the information is uncertain like in stocks, general ideas of diversification are strong. Bigger gap between what you can measure and what matters. Evolution resist change making the few changes that get through essential and robustHave a weight towards history can avoid the boom and bust of fads
It can be rational to think less and draw with a broad brush instead if a pen makes details you focus on. Heuristics exist for a reason and can be computationally ideal and efficient.
Relaxation
Solve simpler problems first. Discrete numbers to cover area are with something like firestations so it covers everything. Having it be continuous then rounding can be easier.
Randomness
MonteCarlo, Hill Climbing. Metropolis problem, best heat it up then cool it down to solve optimization problems. Pick a random solution then randomly make a change if the dice is above 2 and always take a superior change. Then cool it down more only if 3 or more. Etc until you reach local maximum. Creativity and fitness of systems depend on randomness where gpod ideas are preserved. Creative people are a hot melting pot of ideas where bonds and associations can be easily created or destroyed.increasing randomness can move your life and habits from local maxims. Front load randomness and sometimes take a bad idea, but always take a good one.
Buffering
Mm, yea, uhhuh actually serve a great purpose in speech encouraging a story on. We are always connected because instead of dropping packets we buffer communications. To be answered later
Game Theory
Exponential backoff in relationships. Try to reschedule after one week then 2 etc instead of just abruptly giving up. Sake in court case punishment
I know you know what I know, recurison. Can play normally, but bait opponent into thinking you are playing st a higher level. Often trying to play levels just makes you make an inferior play.
Did you know that for any 2 player game a nash equilibrium exists? Finding Nash equilibrium is intractable. The fact that it's hard to find means rational agents probably won't do it. Prisoner dileema. Price of anarchy can be quantified. 4/3 Price of packet routing meaning 33% more inefficient. In theory you could increase the penalties to any level and selfish agents will still choose it.
Unlimited vacation prisoner dilemma, want to take as much vacation as possible yet less vacation then peers to be seen as more valuable. Nash equilibrium is 0. Can be seen in who opens earliest on black Friday. Reverse game theory, mechanism design, what rewards would encourage best behavior. Could worsen every outcome, but by changing game make everyone better off. Cant just pay 1k to people who tske vactaion, vhange the game by making compulsory vacation. Vindictive actions like scaring reviews are irrational but good for society.
Emotion
Morality is here instinct in humans and emotions are the mechanism design in a species. Emotions are involuntarily so cooperation doesn't need enforcement. Someone who responds irrationally to theft will have their stuff stolen less.
Love changes the payoffs making long term relationship and those long term commitments worth while if leaving will being massive pain.
Auction
Vickrey auction, a blind auction where the highest bidder pays the 2nd highest bid price. Honesty is the best policy and the game ensures that. Overtime, the auctioneer won't actually lose money
Cognitive Kindness
Cognitive Kindness. Computers and us both have limited space and time so we are comrades and can learn. If we follow the best process it won't always work. Reduce the cognitive load of others, don't ask what they want, reduce the choices. When are you free vs when ate you free on Wednesday? Humans face some of the hardest problems with incomplete information, assumptions, and trade off speed and accuracy as is rational.
"Intuitively, we think that rational decision-making means exhaustively enumerating our options, weighing each one carefully, and then selecting the best. But in practice, when the clock is ticking, few aspects of decision-making (or of thinking more generally) are as important as this one: when to stop."
"What’s more, being able to fall involuntarily in love makes you, in turn, a more attractive partner to have. Your capacity for heartbreak, for sleeping with the emotional fishes, is the very quality that makes you such a trusty accomplice."