Decisions
I, as an agent in the world, decide between actions that lead to different macrostates in the world. (See Carlo Rovelli, “Agency in Physics”.) The scary part here is not the action, but the potential irreversibility of it. First insight:
If possible, make decisions reversible.
It helps enormously if we can “try” something, observe what happens, and decide again — or factor out a reversible part and test it first.
Often that is not the case, and this part is more what this essay is about. This essay is also not about the “relatively simple” case where, like in poker, I choose my action with both a clear goal in mind and the probabilities of all the possible outcomes in principle knowable. This example clearly shows: We don't choose the outcome when we play a card - the actions of the other players and “luck” in combintion with my actions decide the outome, and we can only control the latter.
This essay is more about if either goals are unclear, probabilities are unclear, or even both. Let's take some examples. We might be choosing:
- bike tires
- a chess move
- where to live
- a job
- a business strategy
The goals are: ride as fast as possible, win, be happy, be successful, make money. Or? No. Life is mostly multi-objective.
For bike tires: We want speed, but also puncture resistance and grip.
For chess: We want to win ... but how?
For a place to live: close to nature, close to work, close to friends and family, quiet, enough space.
For a job: interesting work, enough money, autonomy, good people.
For a business: grow, make money, preserve freedom to act, or survive.
They are sometimes in direct contradiction to each other - like puncture resistance and rolling resistance.
Constraints and Pareto fronts
A first tool is constraints. We can use them to cut away bad options. (See Constraints.)
We might restrict the “where to live” search to Central Europe.
We discard all chess moves that immediately lose a piece.
This is also the first way to make a mistake: maybe we just threw away the best move. Maybe we would actually be happiest in an igloo.
Pareto optimal does not mean perfect. If the criteria are chosen well, Pareto analysis has now done all it can. Unless we can state a genuine preference among the remaining trade-offs, every point on the front is equally good for the decision—not identical, but no longer rankable by the criteria we chose.
Of these examples, bike tires come closest to being quantifiable, at least as a vector of properties.
For where to live, I can quantify distance to work, distance to family, rent, noise, and square meters. But does this cover it all? And how do I map it to “being happy”? I am convinced that personal “utility” is not a “number” (see Not everything is a number) - which makes such a decision so hard.
When the map is unclear
Sometimes the goal is clear, but the mapping from options to the goal is not.
We can calculate a few chess moves ahead, producing a tree of possibilities. But at the edge of our calculation, we still have only positions—not outcomes. We still need to evaluate them.
Fortunately, chess has many regularities: doubled pawns, good and bad bishops, piece values, activity. A master learns to weigh these against each other.
As Daniel Kahneman argues in Thinking, Fast and Slow, expert intuition becomes reliable in a sufficiently regular environment, with repeated practice and clear feedback, as in chess.
Under deep uncertainty, they are absent. For a business strategy, there are too few true comparison points. The market, competitors, technology, and economy are not the same game. Here even expert intuition fails.
Instead, we have to look somewhere else, like robustness:
which option remains viable across many plausible futures?
We can also look for failure modes: Under which plausible futures does this fail, and can I survive those failures? This is not the same as optimizing for the literal worst case. Almost no strategy survives the worst case we can imagine.
But perhaps we are running a startup. Then a strategy optimized only for survival may be the safe way to lose.
The question becomes:
Which adverse cases must I survive, and which upside can I not afford to miss?
So before choosing a method, we should know what kind of decision space we are in.
Finding options
Before choosing between options, we need options in the first place.
A useful practice in design is to separate generating options from evaluating them.
For a place to live, we can talk to friends and colleagues who made similar choices, look at places we had not considered, or spend a few days in candidate cities. We can iterate ideas, generate new ones, and combine them.
Of course, there is a trap here too: endlessly generating options and gathering information. At some point, one more option—or one more piece of information—is no longer useful. It is delaying the decision.
Ask: What could I learn that would change my choice? If the answer is nothing, stop researching. Keeping options open costs time. Some options disappear while we test others.
More than one way to reach the goal
More than one path can lead to the same goal.
For where to live, I may be happy with many friends in a loud city. Or alone, close to beautiful nature. The mind easily makes one option great and the other terrible. Usually, both have advantages and disadvantages. If neither is clearly better once the trade-offs are explicit, we are probably close to a Pareto front again. Then the question is personal:
Which trade-off can I live with better—now and in the future?
A useful technique is to think not only in OR, but also in AND.
First, name what each option makes us give up. Then see whether we can actively compensate for it.
Live in the city and make regular weekend trips to the mountains.
Live close to nature and deliberately plan holidays and time with friends.
The same is true even in the simple choice of bike tires for a gravel race. To win, I can attack uphill, downhill, or pull away on the flat.
Breaking down distant goals
Sometimes the goal is too far away to map directly to options. Then it helps to break it down into nearer goals and proxys.
In a project, nearer goals can become milestones: build a prototype, find a first customer, break even.
For where to live, “Will I be happy there?” might become the proxies:
- Can I build a social life there?
- Can I get outside often enough?
- Can I find work I would actually want?
But neither a milestone nor a proxy is the goal, and both become dangerous when one mistakes these!
When we still do not know
For personal decisions, well-being is often the goal and the body often knows what it wants before the articulate mind can say why. “I have a weird feeling in my stomach” is precisely that. But this is not a verdict, that instinct needs to enter into the discussion with the articulate mind. We have to talk to that unarticulated feeling like we would to an injured kitten: carefully and slowly, trying to find out where it hurts - the paw? the tooth? or somewhere else? Ask, in a quiet moment:
What is this feeling really about?
and wait. Try an answer and notice whether the feeling settles or resists it.
For a business strategy, this feeling may tell me something about my fears or wants, but it cannot tell me what the market will do.
Here, the process has to carry more of the load: find options properly. Keep quantitative and qualitative judgments distinct. Test what can be tested. Look at adverse cases and upside. Iterate the options.
Eventually, we move from options to a decision.
If we have done all this and several options still remain, there may be no hidden calculation left to discover.
A sound decision can end badly; a careless one can get lucky, like in poker. We should judge a decision by what was knowable when we made it and by how well we used that knowledge.
We cannot choose the optimal outcome. We can only make the decision optimally.