When should you stop looking for a better deal on a house? How much new music should you listen to vs. your classics? What’s the most efficient way to organize your closet? Which chores should you do first on Saturday?
These practical questions lack an intuitive answer. We experiment over time and act accordingly. There doesn’t seem to be a correct answer to any of them. Computer scientists have proved this wrong. In writing algorithms dictating computers’ actions, computer scientists have determined the most efficient ways to explore new concepts, sort large quantities of data, and organize a to-do list.
In their recent book, Algorithm’s to Live By, Brian Christian and Tom Griffiths explore these solutions and present them in a manner accessible to non-geeks. Find below a distillation of a few key points from their book. It’s worth exploring if you’re interested in this organizing your life.
There is a best time to look for something and a best time to stop looking. Consider looking for a house in a fast-paced housing market. The most efficient strategy is to determine the total number of houses it is reasonable to consider in your search. Say 100.
The optimal stopping point is 37% of the # of houses you are willing to consider. Computer science has proven the optimal strategy is to pick anything better than the first 37% of houses you consider. This will prevent you from falling into the trap of stopping too early or late. Look at the first 37 houses with no commitments, then on the 38th house take anything that is better than the first 37.
This is a simplified rule. The book details how adding new parameters, like the ability to backtrack to previous considerations, will modify the optimal strategy. If you have quantifiable parameter (such as average home price per sq. ft in your market), you should utilize this data to bolster your dataset and set an average price/sq.ft threshold for choosing instead of strictly following the 37% rule.
All things beings equal and choices disappearing after you pass over them, picking something from a wide variety of unknowns, set the number of considerations you can reasonably make, look at 37% of them without choosing, then pick anything better than those first 37%. I’ll give one more example to illustrate. If you’re at your work holiday party and they have 100 unwrapped holiday gifts being pulled from a bag and you get the first choice whether to take what’s pulled or let it go forever to your coworkers…your optimal choice is to pass over the first 37 items in that bag and pick the first gift better than those first 37.
Attempting to explore why this is the case would be futile in a blog post. Suffice to say, computers often have to pick a way to solve a computational problem with limited resources. This is the best way for a computer to do this.
Explore vs Exploit
Life is always a balance of the novel and the familiar. We want to find new and exciting things while savoring those we love. Computer science also has insights here. There is no perfect answer as to the balance of your Pandora station (new “recommended tracks vs your library) but there is an optimal strategy: minimize regret.
To develop your perfect library with minimal regrets, use the Gittins Index (see red line in image). Starting on Pandora, or any new area of life, everything is new. After about 50 songs, you should listen to about 50% new songs. After 100 you should listen to 25% new to old songs. As you approach 1000 songs, listen to 3% new to minimize your regret level.
Listening to only new music will expose you to many songs you don’t enjoy. While listening solely to your library will prevent you from discovering songs you’ll love even more. The optimal balance to reduce regret is listed in the chart above. As your library builds, you’ll experience more variety and novelty within your library and thus be less bored by the familiar. The optimal music library size may be around 500, where marginal gains in new music level off. At that point, you can be 90% confident your music is better than the next random song in your “recommended” track list. Listen to a new song every 10 songs to minimize new-song regret and you’ll be on your way to building the optimal music library. This doesn’t account for boredom with your current library. It assumes every song in your library is of a constant enjoyment value to you. You could, however, factor in a certain enjoyment decay factor to account for this.
Sorting your DVD’s A-Z isn’t necessary. But there is a best way to do it. Don’t take all your DVD’s off the shelf and put one back, then place one more before or after, then one more in it’s alphabetized place. This is inefficient because every new sort task needs to be compared against the whole library.
The optimal solution is called merge sort. You break the DVDs into several large groups, then merge those groups together.
But, when in doubt, err on the side of messiness in ordering/sorting problems. Take the DVD home library for example. Unless you have a massive collection, it will almost always be less effort to search through your collection than to intentionally sort it. There is a better third option to random organization and alphabetical: Caching.
Caching is choosing a small portion of your library to be quickly accessible. Computers use caches to make memoryfaster. For this a good example is your clothes closet. The tendency may be to sort articles of clothing by type. This is not the best way. In order to have the highest likelihood of accessing the right piece of clothing, you should place everything back in the order of most recent use in a prominent place. Because in almost all aspects of life, what is used most recently is most likely to be used again. This also applies to papers on your desk. You shouldn’t try to organize them into a “logical” order. Placing the most recently used papers on top is the best, assuming these are papers that belong on your desk and aren’t for archiving.
Another note: Keep things where you use them.
For scheduling tasks, make your goals explicit. Prefer tasks with soonest due date, weighted by importance, with roadblocks of any size as a priority. If a new task comes that is due sooner than your current task switch to it if the switch in context seems merited. When checking your email, do not sort it by importance level before opening. Going at random will prevent the waste of time in the “metatasking” of email inbox organization. It’s almost always better to batch a bunch of tasks together to get cognitive momentum and remove distractions from your working schedule in at least 90-minute chunks and avoiding “metatasks” such as spending lots of time choosing what to do first. Have a system for prioritizing and stick to it. However, some jobs and situations won’t allow this and you’ll need to be cognizant of that.
The optimal auction is called the Vickrey auction. Everyone bids once what they believe an item is worth. Then the winner pays the #2 bid. This prevents unneeded competition and potential misevaluation due to false trends in bids. An example of a pitfall in standard bidding wars: Consider five companies bidding on a bridge construction project. Consider: Company one bids $6 million instead of $8 million because of a surveyor’s error. Company 2 would have bid $6.25 million but is undercut by Company one so now bids $6 million as well even though they knew the project was worth $8 million. Now companies 3, 4, and 5 see the bids and are convinced the value of the project is somewhere between $6 and $6.25 million. The Vickrey auction would have avoided this by preventing comparative sequential bidding that leads to a battle over information.
It’s a common theme amongst my friends when asked “where do you want to eat for dinner” to say, “I don’t care, anywhere you pick is fine.” While feigning kindness this is a burden being placed on your friends. Without any information and parameters, choosing becomes more cognitively complex. In reality, it’s almost never true that anything is fine.
Computer Science and Mathematics can help us build a practical structure to life. Machines need to adapt to environmental changes and human demands like us. Observing how mathematicians use theory to design machines can be invaluable in implementing better systems in our own lives.