BLOG

Simulated annealing: a life framework

Making good decisions despite being biologically bad at making them

I enjoy learning about different systems for living a good life. These can range from specific productivity practices to high-level formulations that attempt to reimagine the very concept of meaning in your life.

As an efficient software engineer, I really dislike repeat work when perfectly good tools already exist in the wild. (At the bottom of this article you can find a list of existing frameworks I find particularly interesting and thoughtful.)

But over the last few years I’ve been mulling over an idea. My background is in computer science and software design; the idea originally comes from those fields, and has in fact been around for a while. Recently, I’ve been thinking more and more on how the principles behind this “old” idea become new and interesting when you move beyond the bounds of technical applications, and start using it as systematic way to think about life.

In short, simulated annealing is a framework for life decision-making. It can help you with whatever you are meaningfully striving towards in your life. Specifically, it tackles the following fundamental assumptions that others don’t directly address:

  1. Humans are very bad at knowing what they want
  2. Humans are also bad at understanding probability
  3. Humans want to control how they grow and change


Simulated Annealing 101

Simulated annealing (SA) is a concept taken from computer science and applied mathematics. I first encountered it in college during a class about artificial intelligence, where we wrote toy programs to outcompete each other using various algorithms and strategies. SA is one of those strategies that turns out to have very practical implications for life.

SA is a particularly effective strategy under certain conditions; specifically, when you have:

  • A massive search space or unmanageably large number of combinations
  • incomplete information, and
  • limited time.

With these constraints, SA performs remarkably well at finding a very good solution. Note that it does not necessarily guarantee finding the overall optimal solution.

I can’t think of a better description for the general conditions of life. Humans have seemingly infinite options for what to do in life. However, we can’t spend forever searching for the best solution since we have a limited amount of time in our lifespans. So how do we look for the best possible outcome given this reality?

Simulated annealing is an improvement on naive search optimizations, like hill climbing, with two special “tricks”:

  1. Search the current set of available options for the best choice.

  2. Trick #1 (jump): At some probabilistic interval, jump to a new area in the search space and begin searching even if the new area has a worse state than the original area.

  3. Trick #2 (slowdown): Over time, decrease the amount of negative value you are willing to accept to jump to new states. This means that you only accept smaller and smaller differences in the current and new states when jumping, until you eventually stop jumping to new areas.

This will be summarized as the search-jump-slowdown cycle. It is a bit of an oversimplification but hopefully you get the idea. The name and inspiration actually came from the metallurgical process of annealing, a technique to reduce defects in metal by the heating and controlled cooling of the metal.

Simulated annealing is innovative is because it overcomes a major hurdle of more basic heuristics: getting stuck on local optima.

If you naively search by greedily choosing the best out of visible choices, how do you know the final “best” choice is truly the global optimum? Life is an incomplete information game and you are searching among endless possibilities, so optimizing from an arbitrary starting point is likely to just get you to a overall mediocre outcome.

process Searching for the highest peak using simulated annealing, with two jump phases.


A practical example

Let’s try a practical example: finding a new sport to do as a hobby. You have a set of criteria you want for your hobby: you want it to be fun, improve your health and fitness, and maybe even be a sport you can get skilled at given your natural talents.

sports-base The available search space for your new hobby.

An easy first simple initial search method is trying out sports related to your past experience. Suppose you were a runner in high school; you might begin by signing up for a local running group and maybe a team relay race with friends. Finding the best option within these choices is easy. You can even expand your consideration to sports with a big running component pretty quickly, like ultimate frisbee or soccer.

sports-running Searching among all running-related sports.

You’re currently searching in the space of “running-related” sports. But what if there exists sports outside of this range that would be better suited for you? More fun and faster improvement curve for your skill and physical fitness? What if it had nothing to do with running at all?

Many people at this point prefer not to deal with the uncertainty of whether running-related sports are best suited for them. They buckle down and commit to their running-type sport. It would be impossibly time-consuming to exhaustively try new sports one after the other.

And even if you’re wiling to try out other sports, society doesn’t teach us a good method to start or stop our searching. Maybe you try a few vastly different sports at random, until you get too busy or demotivated to keep looking. Perhaps you get lucky and have a friend or members of your local community introduce you to a great new sport. However both these approaches don’t constitute systematic ways of thinking; random searches based on willpower and relying on others are not reproducible decision-making patterns.

sports-jump

Simulated annealing offers a practical framework for finding that new sport. After an initial search phase of “running-related” sports, you should jump to a radically new area. Importantly, you should spend time to explore the new area even if it’s worse than the current state.

Let’s say we choose to try surfing. If you end up enjoying it, then you can straightforwardly switch to surfing as your main sports hobby. But if it is neutral to deeply horrible, you shouldn’t necessarily swear off all watersports. Following a simulated annealing strategy encourages you to do a little more searching in adjacent, related sports. It may turn out that the best watersport for you is actually superior to the best running-related sport.

water-sports You might stop searching before you realize you love kite surfing. Too bad it’s not as cool as your rich tech bro’s favorite sport: foiling.

The second trick of simulated annealing is also extremely suitable for life decision-making. Since we can’t search forever, we gradually reduce the number times we jump and how willing we are to suffer through new (bad) sports experiences. In our simple example for finding a new sports hobby, probably exploring 3-4 new categories of sports is sufficient to make up your mind. You slowly settle on your favorite sports hobby after searching through a much larger set of sports than you otherwise might have.

best-sport After repeating a few jump iterations, you conclude squash is the best sport for you.

So by not searching endlessly or until whatever point you are exhausted, you still have time to play and enjoy your chosen hobby. And because you have been deliberately searching through very different options, you have steadily increased the degree of certainty that the final sport is very suitable for you.

Which now brings us to how simulated annealing addresses those three fundamental assumptions…

1. Humans are very bad at knowing what they want

Many life philosophies are created in order to streamline your focus and output based on what you want to do.

But if humans actually are very bad at predicting their own desires, then those systems are doomed. Hacking your productivity or sticking to habits is useless if you made choices based on the wrong preferences. People will be left with a lot of existential angst and regret when they spend too much time pursuing ultimately unsatisfying results.

There are many factors that can contribute to someone incorrectly assessing their personal preferences. Preferences may change. Individuals can be influenced by what society says is the best or which is the most socially-approved choice. There is even evidence of a human cognitive bias called the introspection illusion, where we tend to believe we have direct insight into our own mental states while other people’s introspections are regarded as unreliable. This bias can lead to people falsely explaining causes for their behavior or inaccurately predicting future mental states.

Simulated annealing is unaffected, and maybe even improves in usefulness, when people are bad at knowing what they want. That is because SA is a metaheuristic, or a higher-level procedure to generate a search heuristic. So even if the underlying search criteria change, you can still apply the search-jump-slowdown cycle with the new optimization preferences.

experience Hopefully the improvement in your assessment of what you want, over time.

Another benefit of utilizing simulated annealing, is that the “jump” step forces you to quickly prove or disprove your hypothesized preferences with real experience. Nothing clears up any false preconceived notions of what your ideal date is by going on a dreadfully horrible date with someone who fits that type.

Clarifying your actual expectations through experience and testing out a vast array of possibilities has the added benefit of providing certainty to your decision-making. Many times I’ve heard friends say they aren’t sure about the long-term viability of their current partner. These are friends who have been dating their partners for a significant period of time – months and even years. Ostensibly they have had enough time to deeply understand the personality and quirks of the other person. So their lack of commitment stems from an internal uncertainty rather than not enough knowledge about the partner.

This hesitation either comes from being unsure of their own wants or lack of experience, oftentimes both. For those lacking experience, they often haven’t dated enough to decide on this relationship. In other words, people don’t know if their current perspective is accurate or not because they have no adequate reference points. It is not so much a “grass is always greener” problem as it is an inability to even know what the color green looks like.

So now we get to the next part, which is: how many different options are really even out there?



2. Humans are also bad at understanding probability

Science has proven that humans really stink at probability. Hindsight bias, base rate fallacy, gambler’s fallacy, neglect of probability… just to name a few failures in our mental capacity to understand probability. Buster Benson has a great article that breaks down cognitive biases into larger categories. Many of our biases stem from our inability to intuitively understand and apply probabilities.

sports-all

This failure to understand probability manifests in a lack of consideration for the space of possibilities.

If we look at the practical example again, the search space for a new sports hobby is not so large. It is possible to explore a significant area of the search space with only a handful of jumps. But this is not always the case for other life decisions. Questions in life, and especially the most existential meaningful ones, can have massively huge sets of potential answers.

For dating, there are probably hundreds of compatible romantic partners for each of us in the world. Even if you winnow down the available dating pool based on specific ages, location, education level, favorite movies, or whatever criteria you care about, there will still be more people than you can possibly finish searching through in one lifetime.

dating Dating is a life search space that has gotten larger and more complicated over time.

It used to be that we could quickly choose between the options for romantic partners. There were only so many eligible people similar in age living in the local village or town back in the day. But with the globalization of communication, increased accessibility to travel, more frequent moves to new cities and countries, and the meteoric rise of online dating, the search space for romantic partners has exponentially increased.

Simulated annealing pushes for a more deliberate and diverse-seeking mindset towards the problem of dating, not only to clarify your own desires and expectations, but also to decide how much of the available search space to explore. Before, you might only need to date one or two people to feel like you’ve understood most of the available dating pool. Nowadays it will take many more jumps to achieve the same level of knowledge and certainty.

I want to emphasize that I’m not saying this is a sure-fire system for achieving success in dating. And I am also not advocating for dumping long-term partners and dating 100 people in an attempt to exhaust the search space.

Rather, it’s a helpful mindset when you are considering a new direction to take. If there was a person you would normally reject, but you don’t actually have any experience motivating that reason, it might be worthwhile to accept the date invitation instead. Not only because you might prove or disprove something to yourself, but also because you will be intentionally expanding your coverage of the dating search space.

The other systems that I’ve encountered tend to ignore the difference in size and complexity of search spaces. Productivity-focused frameworks push you to think about how to quickly and effectively improve the prioritized list of (five? three? one?) outcomes you are focused on. Never have I seen a productivity system ask you to reflect on the overall set of choices, and whether the ones you’ve picked are really the best fit for you. Coupled with the assumption that humans are really bad at knowing what they want, it’s very unlikely you can reach the optimal solution based on mental rationalization alone.

I am also very suspicious of such productivity systems since they tend to come with branded swag they are trying to sell you. This starts edging out of life philosophy territory and into the category of sellable, marketable self-help products. While the underlying ideas might be useful (ex. I used the basic concept behind Bullet Journal for many years), beware that they have an incentive to push you to get on with your tasks. The more unquestioning attitude you have then the quicker you’ll be reading their book or downloading their app. It is against their business interest to tell you to stop and evaluate why you’re doing things in the first place.

So by its very nature, the system of simulated annealing forces the user to pause and consider the size of the search space. You can use your estimation of the size of the search space to decide how many times you want to repeat the search-jump-slowdown cycles before you are likely satisfied. A large search space may require many more iterations, while a smaller search space can be narrowed down more quickly. Deciding on your life’s work or romantic partner probably needs more thoughtful searching than picking your new sports hobby.



3. Humans want to control how they grow and change

In addition to not thinking about probability and the range of possibilities, I think many life philosophies mistakenly focus on personal growth. Like I was saying above, these systems end up just being productivity “hacks” that push you to unquestioningly make plans and habit trackers rather than doing the hard work of thinking about what you should be focusing on.

what-you-can-imagine

I believe the popularity of personal growth frameworks comes from an underlying desire to control how we ourselves change. You might be okay with thinking about children as being heavily influenced by upbringing and environment but, as an adult, it’s very alarming to think about how your future self will be a mix of only a small amount of your own input and mostly random chance and external forces. It’s even worse to consider a future self who you can’t imagine at all based on your present values.

Funnily enough, I think kids actually think of their adult selves that way. I remember being young and failing to imagine what I would be like when I grew up. Not because I lacked any imagination, but because there wasn’t one thing that I thought I would definitely be. There seemed like an almost infinite set of options that my adult self might become. After all, what kid ever really thinks they are going to grow up?

new-state

Simulated annealing can restore some of the child-like carefree attitude towards radical, unpredictable change. By accepting that the search space of life is huge and impossible to fully comprehend, you are acknowledging that the “best” choice for you might be something that you can’t conceive of at this moment. And even if external forces push or pull you to arrive at certain choices, SA will naturally resist that by encouraging you to go way out of your comfort zone.

One other thing: “personal growth” not only evangelizes that people should try to control how they change, but also that people need to be constantly growing. This fixation on growth can cause you to discard worse-off choices that could over time lead to a better answer. Accepting certain levels of discomfort and negative experience, tempered by a continual decrease in the tolerance for those negative situations, could help you discover answers you would otherwise have thrown out.


Simulated annealing is not a “formula” for success

What I like so much about simulated annealing is it gives very concrete steps for how to expand the range of possibilities for important life decisions.

lazy-possibility

Simulated annealing provides just enough high-level structure to encourage me to explore more diverse versions of myself that could exist. Without an explicit forcing function like the jump phase, I can too easily fall into the trap of rationalizing away why the new state would be a waste of time. Lazy people like me will just revert to the default in the absence of systematic accountability.

Yet for all it’s tactical usefulness, simulated annealing should not be viewed as only a formula for extracting the best solutions out of your life.

Personal growth mindsets tend to obsess about reaching the optimal outcome as fast as possible, providing techniques concerned with boosting your execution and cutting any mental dead weight. Even the more “enlightened” versions of this, which try to balance mental wellbeing and prevent burnout, still preach running as fast as possible towards the goal. Events like the Ultraworking Pentathlon sell attendees on the chance to blitz towards success, promising:

Peak Performance, Now.
Permanent Gains, Forever.

Using that personal growth outlook, you can view SA’s search-jump-slowdown cycles as another strategy to speedily locate a solution. But this ignores a significant part of simulated annealing’s value as a life philosophy.

Simulated annealing makes no judgement on whether each iteration gets you closer to that global optimum. You don’t necessarily know if the new area you’ve jumped to is where you want to be heading. If you only care about rapid completion, you can end up regretting the time spent exploring a suboptimal area. Here the built-in inefficiencies of simulated annealing combined with the personal growth mentality results in an extremely negative frame of mind.

So it is important to view simulated annealing itself as a process to enjoy. Within each iteration, you are intentionally giving yourself space to respond both rationally and emotionally, consciously and unconsciously, to the experience. By having a positive framing that is unconcerned with the fastest route towards the final outcome, you can view each twist and downturn in the path as leading to valuable insights, building a greater understanding of your own self.

By contrast, when you are only focused on personal growth, reaching the end of your goal is the ultimate payoff for all the hard work. But what about the experience at the top? Unless you drop dead the moment you reach the top, you have to keep living and doing things.

For personal growth lovers, this state is an ultimately boring one. You either spin endlessly at your goal, moving nowhere because there is no room to go up and no point in moving downwards, or switch to pursue some other goal.

Meanwhile, the simulated annealing framework does provide motivation for lateral or downward steps. The satisfaction of discovering new insights and self-knowledge produces value beyond the achievement of success. It treats life as a game to be played, not won.

self-actualization

In this way, simulated annealing fosters a self-actualization mindset instead of one of personal growth. Rather than striving for some optimal version of the self, “actualization” just means making something into reality. The term cares nothing about whether or not you are moving towards, or have already found, the best solution. It also doesn’t matter the speed at which you’re moving – if you’ve slowed down, sped up, or are staying still. We can extend the challenge and engagement in life however we see fit.


Jumping: a year abroad in Japan

I left my job and the Bay Area in 2018 and moved to Tokyo with my partner and dog. I am essentially taking a year-long sabbatical to learn Japanese and explore the country, living off my savings. I took the leap (ha!) to move internationally in large part because I viewed this as an important “jump” - one phase in my great exploration of what it means to live a rich and fulfilled life.

Although it turned out to be an incredibly positive experience, I had plenty of concerns and reservations before coming here. I don’t think I would have been willing to move if not for the simulated annealing mindset: there were too many potential risks and negatives for taking a year off in my career, international move logistics were terrible, it could cause challenges in my romantic and platonic relationships, I was sacrificing a year of money-making potential, etc.

Simulated annealing allowed me to reframe my anxieties about moving to Japan to as a natural cost of jumping to a brand new part of the search space of life. It was normal to experience a potential negative hit to my life during a jump phase, and because I am early enough in my search for a diversified life, I am okay experiencing a larger negative cost. And by not doing it, I could be throwing away the chance to discover a better answer to my question of: “What is my ideal life?”.

Without the forcing function of a jump phase, I might have never moved to Japan. It also means I might have never started a podcast or built a game to help language lovers learn animal sounds from around the world. I would have never traveled to dozens of areas in Japan or had new revelations about how I wanted to spend my retirement. There is so much more that I could write about the impact of this year in Japan that it requires it’s own dedicated post.

So is Tokyo where I can ultimately live the most rich and fulfilled life? It’s hard to know for sure. Remember that life is a decision-making game with rigged odds: it’s messy, complicated, and impossible to be fully understood.

And like I mentioned in the last section, the answer doesn’t matter that much. The more important thing is that I’ve learned a lot about myself, made some wonderful memories, and have started forming an idea of where my next big jump will land.


A list of other life frameworks

Have other frameworks you like? Simulated annealing can certainly be used alongside other life philosophies.

Take the deathbed game by Buster Benson, a lovely framework that helps you reconsider what is valuable by continually reflecting on what you will regret on your deathbed. One of the top regrets people have on their deathbed is “wishing I had let myself be happier.”

It’s easy to think of straightforward changes in your life that are preventing your happiness. Eliminating those is a clear first step. But after that, what proactive changes should you take to add more happiness to your life? The deathbed game doesn’t define an explicit process to make new changes, so pairing simulated annealing and the deathbed game together can be effective at adding practical steps for your decision-making.

Here is a list of systems that are useful to check out:

  1. Getting Things Done
  2. Bullet Journal
  3. Decision Journal
  4. Your Life in Weeks by Wait But Why
  5. Life Intensification by Venkatash Rao
  6. Deathbed Game by Buster Benson
  7. Codex Vitae by Buster Benson
  8. The Elephants by Nick Crocker
  9. Cognitive Journaling by Richard Ragnarson
  10. Cadence by William Van Hecke

I’m always looking for new frameworks to investigate. I’d also love to hear any feedback on simulated annealing, you can send me a note here.


This article was last updated on 07/12/2019. v1 is 4,102 words and took 9 hours to write, 8.25 hours to edit, 2.5 hours to draw the illustrations.
«
»