Skip to content

Bayesian Goal Updating

Goals are hypotheses that you can test and update.

Pamela J. Hobart
Pamela J. Hobart
4 min read
Bayesian Goal Updating

One day, you wake up and set a goal. Goal A.

In the first day/week/month, you make significant progress towards Goal A. Hooray!

But your progress stalls, and you fall off the wagon.

You read some productivity tricks: break Goal A into pieces, set a reasonable timeline, track your progress, etc. You take this advice.

Still, you do not achieve Goal A. It's not even clear you're making good progress.

There are actually many possible explanations for your apparent failure:

  • Goal A is too difficult for you - outside your "zone of proximate development." Too-hard goals invite procrastination. You should set an easier goal and/or develop your relevant skills.
  • Goal A is too easy for you - you don't experience "flow" working on it, i.e. it's boring. Too-easy goals therefore invite procrastination. You're going to have to avoid this task or just power through it.
  • You want to have completed Goal A, but you don't like the intermediate steps B-Z. You are failing to identify closely enough with your future self. Find a way to do that.
  • You never really A set as a Goal at all. If it were a proper goal, it would motivate you, but it doesn't. You need to take mental action by recommitting to the goal.

What now?!

Productivity people can tell you what to do in light of your existing goals. There are cute little guidelines about setting a SMART goal (specific, measurable, achievable, relevant, time-bound). And there's some personal development-type talk of how to make use of your failures.

But all of the intermediate steps between goal setting, revision, and abandonment remain basically murky. This is where the magic (or the mess happens). We can theorize this better.

What Goals Really Are

As far as I can tell, the commonsense view of goals (and related phenomenon like intentions, choices, etc) holds that goals are illocutionary acts: goals get made in the stating. In other words, goals are commissives like promises and oaths, though they are often made privately to oneself instead of to an individual speaker or peer group.

On this commonsense view of goals, it is very easy to set a goal - and commensurately easy to break one. This incurs the heavy psychic cost of constant failure. Yuck.

Recently, People Online have tried to snatch failure victory from the jaws of defeat with a constant insistence that learning via failure is often (always?) worth it. This new failure mantra is illustrated with a wide variety of failure porn accompanied by commendations of the failee's bravery/openness, etc. This is a suboptimal solution to an avoidable problem.

I do not think goals are illocutionary acts at all. Instead, they're more like hypotheses. Goals are hypotheses about what one ought to pursue.

These goal-hypotheses live at the confluence of what you want now, what you want later, what you're currently capable of, what you might later be capable of, what seems good all-things-considered within the totality of your life, and what else is going on in the world.

Bayesian Goal Updating

This is the absolutely crucial point: hypotheses require updating in the face of evidence. They are cyclical, not binary. And you can be wrong about your goals.

Basically, you need to think like a Bayesian in order to make proper sense of goal setting, revision, and abandonment.

(I am not the right person to explain this concept if you're unfamiliar - please check out a Bayes theorem explainer such as this one).

You start with a prior hypothesis that you ought to do something. You can hold this hypothesis strongly or weakly at the outset. But this is where the goal process starts, not where it ends.

Next, you go try whatever it is and get some feedback from the world. Or, you fail to attempt reaching your goal, which also constitutes a piece of evidence.

You must update your prior hypothesis in light of the new evidence. What the heck happened?

Probably, you encountered some evidence to suggest that your goal-hypothesis is not correct. Failure even to attempt a step towards the goal, for no good reason at all, is goal-defeasing.

Other types of problems carry weight but not conclusive weight - like partial progress up to a surmountable but difficult roadblock. Or maybe you're not really good at the thing you're attempting, but you really want to do it.

You must figure out the relative strengths of these considerations to know how to update your goal-hypothesis in light of your continued struggles to do the thing.

On the other hand, you may have often experienced flow while working towards your goal, achieved some markers of success, and felt basically good throughout. All these signs point to becoming more certain that your goal-hypothesis is a good one.

The goal updating never ends, but it can only help

This process can and will happen over and over and you gather new considerations about what it will take to learn something, what the payoff is, what your opportunity costs are, and how you feel.

Now, I'm not suggesting that you literally attempt to do math on your goals. But the schema is a sound one. Nothing is lost by looking at goals as updatable hypotheses, and there is much clarity to be gained.

Instead of gluing yourself more firmly to the goal merry-go-round, or going full aimlessness, why not try the Bayesian approach?


I like to think that I've made the goal hypothesis updating process sound fairly intuitive.

Unfortunately, it's harder than it seems to do in practice, on your own messy stuff with your limited and biased powers of introspection.

If you want to try this new way of approaching goals but don't know how to start (or if you start but get stuck), drop me a line or check out how we can work together.

Pamela J. Hobart Twitter

Philosophical Life Coaching in Austin, TX. Also mother of 3, Miata driver, and DIY manicure aficionado.