Seems like the issue is less about optimization per se but the reckless application of optimization to areas where people don't really understand what they are optimizing and the reliability of the control/feedback mechanisms at play. Scanlon makes the point that she considered herself an optimizer but finally understood she didn't understand what she was solving for. I like satisficing as a juxtaposition for optimization because under the assumption that one 1) doesn't fully understand what we are optimizing for; and 2) we don't really understand the mechanism for feedback, satisficing is a robust way to make small improvements (by gradually raising the threshold for outcomes) without expending massive effort and intervention to pursue possibly spurious gains relative to the poorly understood true objective function. Sometimes we do have a well specified and understood objective in which case we should optimize away.
I totally agree with your comment, but am not sure satisificing is the right term. I flagged this upthread, but while I suppose you could claim that "finding the first thing that moves the needle" is satisficing, it seems to be a more general feasibility principle that doesn't quite capture what I'm looking for.
Good point, worked on it a bit with Claude and we landed on ‘cultivation’ and ‘practice’. In the context of ‘maxxing’, the object of cultivation is improving the self, practice is specifically focused on a craft or activity that can be part of cultivation of the self. Cultivation evokes progress and increasing depth of understanding, it encompasses the idea of sharpening and extending what we are solving for over time. Both cultivation and practice encompass quantitative metrics and qualitative nuance, feel, taste. What I like about both of these words is that they also evoke the requirement of time, effort, and a generalized pursuit of excellence rather than a narrow ‘number go up’ focus.
In safety engineering, designers optimize but only to a point. They back off from the "optimum" and leave a "margin of safety". In many cases, this is equivalent to a form of robust optimization (i.e., minimaxing against an uncertainty set of objectives). Regularization in machine learning is another instance of this. When we are uncertain about what we are optimizing, robust optimization can protect us to some extent. It can even protect us from *some* unknown unknowns. Perhaps one reason we find maxxing repellent is that we know it is risky (and not necessarily only to the maxxer but potentially to the rest of us as well).
IMO, *maxxing is a sign that you don't have really hard problems in your life. If you do, you focus on removing the hard problems (like your friends who tried the elimination diet). Only after you've removed all the hard problems making you miserable will you have the luxury to start obsessing about maximising quantified measurements of various aspects of your life.
Seems like Clavicular clearly has hard problems in his (at the very least interior) life that he's very obviously using looksmaxxing to avoid dealing with in a meaningful way. Frequently people chase quantifiable goals because their real problems have much more nebulous and uncertain solutions.
On mechanisms, Marcia Angell's book, The Truth About Drug Companies, had this fantastic analogy. Aspirin is salicylic acid; it reduces fever; ergo, fever is caused by lack of salicylic acid. and then (she was talking about Prozac), SSRI's provide dopamine; they reduce symptoms of depression, therefore depression is caused by lack of dopamine.
On improvement and optimization. I finished reading The Score, and I like the point that games are OK because there are millions of them. So, even if you have a common objective *within* a game, you are free to move on to another game if you don't agree with that objective. Or, you can get together with your friends and change the objective. Therefore, its not necessarily the optimization that's the problem (although it might be), but what Nguyen calls "value-convergence" where everyone is optimizing the same thing.
My defense of economics is that traditionally done, that was the fundamental driving force behind what people were trying to do. Everyone is different, everyone wants to optimize their own thing, we can't see what they are optimizing. Then, the problematic becomes: Under what circumstances can there still be win-win improvements that allow people to do better, even if everyone is optimizing something different. You could argue `value-convergence' is totally ok because you have made `everyone's thing' into `your own thing.' But I think there are many ways in which you can get artificial value convergence if you tweak things around a bit, while still committing to the rationality (transitive preferences) that economics demand in its neoclassical form. For instance, I want to give up instagram, but I really can't because everyone else is using it. So, if we all committed jointly to giving it up, we would all be better off etc. I *think* if you are willing to define value-convergence as the real issue, rather than optimization, we have the language and the formal methods of showing where a problem may arise.
Then, if you want to move away from the individualistic approach to thinking about society and what society should optimize given that there are many, many circumstances where you may not get a win-win improvement, but you still want some kind of action (any redistributive act, for instance), I still think that Sen's idea of Capabilities remains the most powerful.
On the games point, playing strictly for score optimization has been somewhat controversial in some games communities too, especially those like MMORPGs with more of a social component. The term "theorycrafting" (originally derogatory) was coined for people who based their playing primarily on trying to reverse-engineer game mechanics to build a mechanical "theory" of the game in order to play optimally. There's an interesting discussion of that in World of Warcraft in this academic paper: https://gamestudies.org/1801/articles/paul
A word that juxtaposes optimizing life? I dunno. Perhaps common sense? Enjoying life? Finding beauty in imperfection? Imagine if we optimized Berkeley purely for academics. The campus would be awfully ugly and boring.
You've been on a tear recently. The essays on postmodernism have been fantastic, and I get excited when a new arg min post shows up.
This one in particular resonates for me personally, as I've been working hard on improving myself with very little quantification for a few years now. At some point fairly early in the process, I stumbled on the idea of thinking of self-improvement as a posture or orientation (plus action). That is, I work to point myself in the direction of self-improvement and then take steps in that direction, however big or small.
I can't think of a pithy phrase or single word to capture this approach in contrast to optimization, though.
Utilitarian thought and practice requires both quantification and optimization, but virtue is a character trait, an orientation towards the world and / or yourself.
In reinforcement learning, folks try to formalize this as "empowerment". Taking actions that do not close off future options. This is also a standard strategy for acting within a hierarchy: you try to avoid closing off options for your supervisor(s).
The GLP-1 example might turn out to be more broadly indicative of the hazards of maxxing as an objective, because that nigh-universal weight reduction is on the far side of a truly dizzying array of side effects. Of course you lose weight when your gastric emptying has slowed to the point of making your burps smell like farts. I suspect that after maybe making some progress towards acknowledging rounded-but-basic facts like 'bodies that are different are still functional but you probably would be better off not being obese' (which seems to be in line with the 'make things better but don't be a psycho' ethos you're suggesting here) we're going to be entering a hammer-looking-for-nails situations where every medical condition that is a correlate of weight gets a MuD-PhuD study treating it with Ozempic and a shiny p-value from all the pukers dropping out.
As for the word(s)? Sanity? Eudaimonia? Maybe 'ordinary improvement'? 'Relaxed optimization'? 'Diverse qualitative improvement (DiQI)'?
I tend to agree with you about that last point. There's definitely a lot of irrational exuberance around GLP-1s as wonder drugs. In an optimization society, everything--even perfectly good, amazing things--must be overhyped.
"Growth"? (It interests me that in the culture of maxxing and optimisation, the idea that the economy as a whole could reach a stable point from which no improvement is possible is anathema!)
Yes, there is something special about maximization with an unbounded objective function. Most of economics is like this, as is reinforcement learning in spirit.
It's funny because people seldom model minimization problems with the hope they are unbounded below.
Formalizing 'trying to be better' is a different optimization problem. There, the goal is to get better, with a constraint on energy invested in getting better. KKT conditions mean we can re-write that differently. Technically, I think that is still 'maxxing', just a different thing to maxx?
A decade ago, “maxxing” meant maximizing your RPG character on one dimension or another. This is fine. An RPG character is in fact nothing more than a spreadsheet that is maxxable.
The reason maxxing is grotesque is that the subcultures that promote it today are impossibly grotesque.
Is a runner who's trying to improve their personal best 10 km time reducing themselves to a RPG character? The number they're optimising is just one facet of them. The reduction would occur if they treated everything in their life as numbers to be optimised. But few people actually do it.
I think that’s exactly right. If you’re optimizing a number, and you understand that number is just one facet of yourself, you’re not reducing yourself to an RPG character.
Another distinction is that DPS has objective value within the limited scope of the game world. Whereas for looksmaxxing, you need to do the extra work of inventing your own game world and then layering it on everything else.
Exactly so! I'm trying to improve my running and weightlifting myself at just this moment, but I'm not reducing myself to it because there's more to my life than numbers.
Parenting small children also has a strong optimisation component:
- you're weightmaxxing them until they reach their birth weight
- then you monitor the number to "follow the curve"
- many parents soft-compete on how early their children achieved a particular milestone ("baby olympics")
There are sound medical reasons for monitoring baby weight and height, obviously, but the mindset it induces is kinda similar to building your RPG character 😁
From "by that time in the game, your Fighter should have DPS at least 10 or you got the build wrong" to "by six months, your baby boy should weight at least 6.4 kg".
I think lookmaxxing is connected to dating apps, which gamified courtship and sexual life. That said, people have been reducing sexual to a numbers game for centuries: the "Catalogue Aria" from Mozart's Don Giovanni is about that exactly.
Seems like the issue is less about optimization per se but the reckless application of optimization to areas where people don't really understand what they are optimizing and the reliability of the control/feedback mechanisms at play. Scanlon makes the point that she considered herself an optimizer but finally understood she didn't understand what she was solving for. I like satisficing as a juxtaposition for optimization because under the assumption that one 1) doesn't fully understand what we are optimizing for; and 2) we don't really understand the mechanism for feedback, satisficing is a robust way to make small improvements (by gradually raising the threshold for outcomes) without expending massive effort and intervention to pursue possibly spurious gains relative to the poorly understood true objective function. Sometimes we do have a well specified and understood objective in which case we should optimize away.
I totally agree with your comment, but am not sure satisificing is the right term. I flagged this upthread, but while I suppose you could claim that "finding the first thing that moves the needle" is satisficing, it seems to be a more general feasibility principle that doesn't quite capture what I'm looking for.
Good point, worked on it a bit with Claude and we landed on ‘cultivation’ and ‘practice’. In the context of ‘maxxing’, the object of cultivation is improving the self, practice is specifically focused on a craft or activity that can be part of cultivation of the self. Cultivation evokes progress and increasing depth of understanding, it encompasses the idea of sharpening and extending what we are solving for over time. Both cultivation and practice encompass quantitative metrics and qualitative nuance, feel, taste. What I like about both of these words is that they also evoke the requirement of time, effort, and a generalized pursuit of excellence rather than a narrow ‘number go up’ focus.
In safety engineering, designers optimize but only to a point. They back off from the "optimum" and leave a "margin of safety". In many cases, this is equivalent to a form of robust optimization (i.e., minimaxing against an uncertainty set of objectives). Regularization in machine learning is another instance of this. When we are uncertain about what we are optimizing, robust optimization can protect us to some extent. It can even protect us from *some* unknown unknowns. Perhaps one reason we find maxxing repellent is that we know it is risky (and not necessarily only to the maxxer but potentially to the rest of us as well).
IMO, *maxxing is a sign that you don't have really hard problems in your life. If you do, you focus on removing the hard problems (like your friends who tried the elimination diet). Only after you've removed all the hard problems making you miserable will you have the luxury to start obsessing about maximising quantified measurements of various aspects of your life.
Seems like Clavicular clearly has hard problems in his (at the very least interior) life that he's very obviously using looksmaxxing to avoid dealing with in a meaningful way. Frequently people chase quantifiable goals because their real problems have much more nebulous and uncertain solutions.
Two thoughts.
On mechanisms, Marcia Angell's book, The Truth About Drug Companies, had this fantastic analogy. Aspirin is salicylic acid; it reduces fever; ergo, fever is caused by lack of salicylic acid. and then (she was talking about Prozac), SSRI's provide dopamine; they reduce symptoms of depression, therefore depression is caused by lack of dopamine.
On improvement and optimization. I finished reading The Score, and I like the point that games are OK because there are millions of them. So, even if you have a common objective *within* a game, you are free to move on to another game if you don't agree with that objective. Or, you can get together with your friends and change the objective. Therefore, its not necessarily the optimization that's the problem (although it might be), but what Nguyen calls "value-convergence" where everyone is optimizing the same thing.
My defense of economics is that traditionally done, that was the fundamental driving force behind what people were trying to do. Everyone is different, everyone wants to optimize their own thing, we can't see what they are optimizing. Then, the problematic becomes: Under what circumstances can there still be win-win improvements that allow people to do better, even if everyone is optimizing something different. You could argue `value-convergence' is totally ok because you have made `everyone's thing' into `your own thing.' But I think there are many ways in which you can get artificial value convergence if you tweak things around a bit, while still committing to the rationality (transitive preferences) that economics demand in its neoclassical form. For instance, I want to give up instagram, but I really can't because everyone else is using it. So, if we all committed jointly to giving it up, we would all be better off etc. I *think* if you are willing to define value-convergence as the real issue, rather than optimization, we have the language and the formal methods of showing where a problem may arise.
Then, if you want to move away from the individualistic approach to thinking about society and what society should optimize given that there are many, many circumstances where you may not get a win-win improvement, but you still want some kind of action (any redistributive act, for instance), I still think that Sen's idea of Capabilities remains the most powerful.
So, my vote would be for "Capabilities"
On the games point, playing strictly for score optimization has been somewhat controversial in some games communities too, especially those like MMORPGs with more of a social component. The term "theorycrafting" (originally derogatory) was coined for people who based their playing primarily on trying to reverse-engineer game mechanics to build a mechanical "theory" of the game in order to play optimally. There's an interesting discussion of that in World of Warcraft in this academic paper: https://gamestudies.org/1801/articles/paul
Arete.
This is a good one.
Flourishing.
"Making things better" is about partial orders. "Maxxing" is about total orders.
There are very few total orders in real life.
Yes. Making things better also applies to partial orders that can be entirely ephemeral.
A word that juxtaposes optimizing life? I dunno. Perhaps common sense? Enjoying life? Finding beauty in imperfection? Imagine if we optimized Berkeley purely for academics. The campus would be awfully ugly and boring.
Satisficing 🙂
I love Herb Simon, but don't find that word satisfactory! I'll explain why in a future post.
You've been on a tear recently. The essays on postmodernism have been fantastic, and I get excited when a new arg min post shows up.
This one in particular resonates for me personally, as I've been working hard on improving myself with very little quantification for a few years now. At some point fairly early in the process, I stumbled on the idea of thinking of self-improvement as a posture or orientation (plus action). That is, I work to point myself in the direction of self-improvement and then take steps in that direction, however big or small.
I can't think of a pithy phrase or single word to capture this approach in contrast to optimization, though.
It's basically virtue ethics vs utilitarianism.
Utilitarian thought and practice requires both quantification and optimization, but virtue is a character trait, an orientation towards the world and / or yourself.
Constraint relaxation, the pursuit of making life less constrained is a form of betterment that is not maxxing---I think.
In reinforcement learning, folks try to formalize this as "empowerment". Taking actions that do not close off future options. This is also a standard strategy for acting within a hierarchy: you try to avoid closing off options for your supervisor(s).
The GLP-1 example might turn out to be more broadly indicative of the hazards of maxxing as an objective, because that nigh-universal weight reduction is on the far side of a truly dizzying array of side effects. Of course you lose weight when your gastric emptying has slowed to the point of making your burps smell like farts. I suspect that after maybe making some progress towards acknowledging rounded-but-basic facts like 'bodies that are different are still functional but you probably would be better off not being obese' (which seems to be in line with the 'make things better but don't be a psycho' ethos you're suggesting here) we're going to be entering a hammer-looking-for-nails situations where every medical condition that is a correlate of weight gets a MuD-PhuD study treating it with Ozempic and a shiny p-value from all the pukers dropping out.
As for the word(s)? Sanity? Eudaimonia? Maybe 'ordinary improvement'? 'Relaxed optimization'? 'Diverse qualitative improvement (DiQI)'?
I tend to agree with you about that last point. There's definitely a lot of irrational exuberance around GLP-1s as wonder drugs. In an optimization society, everything--even perfectly good, amazing things--must be overhyped.
I like sanity. or chill maxxing.
"Growth"? (It interests me that in the culture of maxxing and optimisation, the idea that the economy as a whole could reach a stable point from which no improvement is possible is anathema!)
Yes, there is something special about maximization with an unbounded objective function. Most of economics is like this, as is reinforcement learning in spirit.
It's funny because people seldom model minimization problems with the hope they are unbounded below.
craft
William James used words like “expedience” and “workableness”
Formalizing 'trying to be better' is a different optimization problem. There, the goal is to get better, with a constraint on energy invested in getting better. KKT conditions mean we can re-write that differently. Technically, I think that is still 'maxxing', just a different thing to maxx?
A decade ago, “maxxing” meant maximizing your RPG character on one dimension or another. This is fine. An RPG character is in fact nothing more than a spreadsheet that is maxxable.
The reason maxxing is grotesque is that the subcultures that promote it today are impossibly grotesque.
I would simply not reduce people to RPG characters.
Is a runner who's trying to improve their personal best 10 km time reducing themselves to a RPG character? The number they're optimising is just one facet of them. The reduction would occur if they treated everything in their life as numbers to be optimised. But few people actually do it.
I think that’s exactly right. If you’re optimizing a number, and you understand that number is just one facet of yourself, you’re not reducing yourself to an RPG character.
Another distinction is that DPS has objective value within the limited scope of the game world. Whereas for looksmaxxing, you need to do the extra work of inventing your own game world and then layering it on everything else.
Exactly so! I'm trying to improve my running and weightlifting myself at just this moment, but I'm not reducing myself to it because there's more to my life than numbers.
Parenting small children also has a strong optimisation component:
- you're weightmaxxing them until they reach their birth weight
- then you monitor the number to "follow the curve"
- many parents soft-compete on how early their children achieved a particular milestone ("baby olympics")
There are sound medical reasons for monitoring baby weight and height, obviously, but the mindset it induces is kinda similar to building your RPG character 😁
From "by that time in the game, your Fighter should have DPS at least 10 or you got the build wrong" to "by six months, your baby boy should weight at least 6.4 kg".
As of last summer, I have had to start reaching up in order to hug the baby boy. So I must be doing something right!
The 10K run example hits for me, because I impulsively signed up for a 5K fun run and now I am literally trying to improve this stat right now!
Hmm. Should I start fitness blogging?
I think lookmaxxing is connected to dating apps, which gamified courtship and sexual life. That said, people have been reducing sexual to a numbers game for centuries: the "Catalogue Aria" from Mozart's Don Giovanni is about that exactly.