The Way You Think About Value is Making You Miserable

The cheapest way to compute a value gradient is to value the present at zero

Note: This is advice for younger me. It might work as advice for you, if you think about value the way I did.

How You Currently Think About Value

You see yourself as having a utility function, which maps possible states of the world onto numerical values. You think the purpose of this utility function is to help you decide between possible courses of action.

When you deliberate between courses of action, you attempt to compute the expected utility of future world states resulting from each course of action. Once you feel confident that one course of action produces higher expected utility than the other, you choose that utility-maximizing course of action.

Of course, this is how you’d like to be. It’s not really how you are, but how you aspire to be.

A lot of the time, you are tired or stressed or scared or sad.

The frequency of these negative emotional states means you often follow courses of action that you understand are not making you happier in the long term, but you feel so crappy right now that you just want something that feels good.

You often cannot act the way you want to act, because you just want the pain to stop.

“blueprint for a happy life”, midjourney. Note the many layers of framing; the AI implicitly understands that it takes sufficient framing to have gratitude for the present.

You are making a conceptual mistake. This conceptual mistake is making you miserable.

The reason you are miserable is that you have misunderstood the job of the utility function, so you rarely use your utility function to evaluate the present moment.

After all, you think the purpose¹ of the utility function is to help you make choices which have the effect of altering the present. Instead of computing your utility function on the entirety of the present moment, you most often compute the delta of your utility function between the present and some hypothetical future.

You are miserable because you have substituted this counterfactual-derivative of the utility function, for the utility function itself.

You never bother to compute the actual utility function of the present in full.

When was the last time you computed the utility of the present moment?

What do you think is the utility of just breathing, smiling, and doing nothing to alter the trajectory of the world from its present course?

My experience is that, in most situations, that’s literally the way to maximize future utility. Do nothing. Don’t bother to change how events around me are unfolding, other than to add a smile, a warmer, softer, wider gaze, a more upright posture, and slower, more deliberate breath.

You never actually enumerate your blessings. Instead, you frequently compute counterfactual blessing gradients at your present location. You give the blessing gradient a value of zero unless you are considering going about mucking with your environment.

I constantly enumerate the goodness of the world around me. I often feel grateful that I have air to breathe, a floor to walk on, consciousness, and life itself. I note the blessings that come with each packet of energy from the sun, adding new structure, new life, new energy to the Earth. My default utility gradient is +173,000 Terrawatts because I recognize that the sun is always adding more energy for us to play with here on earth.

You don’t do any of those things. That is why I am happy most of the time, and you are not.

You take all of those good things for granted, or even worse, you focus on some of the bad aspects of net-positivity utility phenomena, because the bad aspects have higher short term variance, and the good aspects - like, the sun keeps us alive - are so reliable that they show up as zero in both terms of your “alternate future minus the present” gradient. Reliably good things are not included in your utility function.

Sometimes you obtain good things, but once you obtain them, you expect them to persist, and thus stop computing them in your utility function. Most of the surprises you experience are thus negative, and you work incredibly hard at the impossible task of making the gradient of your utility function consistently non-zero at the present moment, while ignoring the reliably value-adding phenomenon that is carried across the sky on the back of a giant invisible tortoise². This is why you are miserable.

MidJourney, “reliably value-adding phenomenon that is carried across the sky on the back of a giant invisible tortoise, disney”

In order to be computationally efficient in computing expected utility of worldstates other than the present, you consistently value the present at zero.

If you don’t repeatedly evaluate your utility function, in its entirety, on the present moment, you don’t actually compute your utility function. Your brain is great at optimizing, but this is both a blessing and a curse. If you stop using your ability to appraise the goodness of your current environment, as is, you run the risk of losing it.

Your brain optimizes for the only way you use your utility function, which is to compute its gradient under various counterfactual scenarios. Instead of seeing the present as ‘one million utility units’ and a future state where you have washed a dish as ‘one million and one utility units’, it is computationally cheaper to simply compute the delta between the dish being clean, and it not, as being one.

Of course, doing this makes you miserable because it makes your choices and actions in each moment seem more consequential and meaningful than they really are. It makes you miserable because you don’t give yourself permission to do nothing, to not react, to just smile and blink when things go poorly. It makes you miserable because it convinces you that perfectly fine situations have negative utility simply because the utility gradient that you have computed exhibits a downward slope at your location.

You are mistaking the derivative of your utility function for the reality that your utility function evolved to represent. This confusion makes you miserable.

Why You Should Value The Present at More than Zero

I think the mistake you are making comes from not realizing how deep in the stack your representation of value is.

I think pleasant, joyful feelings come largely from evaluating the present as being good. If you consistently ignore the value of the present (because it doesn’t help you in computing how different courses of action will change the present), I think you’re consistently depriving yourself of the ability to feel joy and gratitude for the things which are constantly present in your life.

MidJourney, “a robotic monk washing dishes, Hokusai Katsushika”

What’s worse, I think this expectation of net-zero utility eventually becomes ingrained.

If you know, at some level, that you never value the present by itself, absent a comparison to some possibly different future, you have no reason to expect future happiness. The only thing you can expect to make you feel good in the future is the temporary tonic release of dopamine, which has to revert to a baseline because the whole dopaminergic system exhibits zero gradient over long periods of time; you brain produces dopamine at some fixed rate and there’s not much you can do to change that consistently, over long periods of time.

I’m not a neuroscientist or brain expert, but I am a parent of multiple children. This means I’ve watched a human cognitive stack come online multiple times. I have a good memory and I’ve kept notes on this.

Each of my children seemed to ‘come online’ as an agent navigating the world, with a will, desires, and sense of ‘good’ and ‘bad’, long before they learned to speak.

Seeing this happen multiple times, for multiple embodied brains, makes me think that our utility functions are represented deep in our brains, and that a lot of what they do, before we start tinkering with them by means of language, is evaluate just how good the present is.

Each of my children, once they gained the ability to speak, soon gained the ability to evaluate counterfactual situations as good or bad. All of my children have seemed to become less consistently happy after this point in time.

They start getting ideas about what good and bad are, they start evaluating the world based upon what they think is likely to happen and how good they think that is. This practice often means they get upset about things they they believe ought to or ought not to happen.

My older children don’t do as much of what my 10 month old does now, which is wander around joyfully, finding wonder in all kinds things in her environment, and curiously playing with the things she finds. When my older children do this, they are often extremely joyful, although sometimes this means making giant messes and loud noises, which often means I get upset and tell them they are being bad.

MidJourney, “a joyful rabbit, m.c. escher”

The words “good” and “bad”, are, I think extremely powerful words. If our brains are learning from each moment, the words “good” and “bad” are the labels on that training data. With good labels, a model can do very well even on relatively simple data. Without good labels, though - good luck.

Adam and Eve, that old training data says, were kicked out of paradise when they gained knowledge of good and evil.

What would it look like if they had never actually left Eden, but only thought they had, because they were confused about what, exactly, was good to eat? What would it look like if an evolved model for seeking and appreciating the good started over-writing its own labels, and stopped adding ‘this is good’ labels to anything other than hypothetical situations?

What would it look like if you were currently living in an environment that provides for all of your needs as best it can, subject to its constraints, but that it sees your need for freedom and autonomy as being fundamentally your most important need?

What kind of life do you want?

Do you want a life where you are consistently mucking around with the state of the world but almost never happy, because your conception of good is entirely limited to what kind of changes you can make to move away from a zero that follows you everywhere?

Or do you want a life where you are consistently feeling good, and therefore spontaneously adding joy and love to the world, contributing to the human project, expanding the tree of life, and helping the universe increase its entropy all at the same time - with occasional pauses to compute how good it is now, and sometimes, even more rarely, an evaluation of different hypotheticals, each of which is scaled heavily by the knowledge that the world is full of goodness and most of your choices simply involve tiny marginal differences in that net total good, like asking into which ocean it is better to spit?

All you have to do, to live in the latter world, is get in the habit of computing your utility function on the world as a whole, in the present moment.

MidJourney, “photo of a child smiling.” Those teeth are terrifying.

1. To the extent that you think anything has a purpose, because you think purpose is an illusion and that everything is meaningless bits of stuff bumping into each other for no good reason.

2. Our ancestors were idiots for thinking that the universe consisted of some phenomena which loved them. We modern men are smarter, which is why we are consistently weaker and more miserable despite living in safer, healthier environments.