What it means to optimise

by Neel Nanda8 min read25th Jul 2020No comments

7

Pitfalls of RationalityRationalityAI
Frontpage

(This is a post from a daily blogging experiment I did at neelnanda.io, which I thought might also fit the tastes of LessWrong)

If you’ve gotten to my ninth post, it’s probably no surprise to you that I’m a big fan of optimising my life! Optimisation, as I define it, is the process of taking actions that best achieve your goals.

As stated, I consider it pretty tautologically obvious that this is a good idea. You can take actions that bring you closer to your goals, or you can take other actions. But I get into pretty frequent arguments about whether this is a good idea. And it’s easy to get defensive, and fall back to this definition, and argue that optimisation is by definition a good idea. But I think it’s valuable to engage with criticism and take it seriously, and to see what you can learn from it.

My best explanation is that this criticism is really directed at what I’ll call naive optimisation. Like a robot, who’s told to clean the dust in your room, but never told “priceless Ming vases are important”, and smashes the vase to clear the dust beneath it. An approach that’s focused on being cold, calculating, and driving as hard as you can towards a specific utility function, with no regard for restraint or common sense. And I completely agree that it’s dumb to be a naive optimiser - but I think it’s dumb because naive optimisation does not best achieve its goals. And if someone can explain how an approach to optimisation doesn’t best achieve your goals, that’s not a problem with the concept of optimising, it’s a problem with how you are trying to optimise. You shouldn’t aim to “try as hard as you can” or “do the cold, calculated action”, you should aim to do whatever action best achieves your goals. And this includes asking whether what you think your goals are, are truly your goals.

But this is all easier said than done. And especially for the kind of person who enjoys thinking about optimisation, I think there are a lot of ways to accidentally fall into the trap of being a naive optimiser. So the rest of this post will be a list of the most common criticisms, and how best to account for them:

  • Objection 1: Optimisation is bad for you. Eg, someone who optimises for their studies, works 12 hour days with no breaks, and burns out within a few months.
    • I think the error here is being shortsighted. Life is a marathon, not a sprint. If you learn 10% more for 3 months, and then burn out, you learn far less than if you work sustainably for 3 years.
      • It can feel unnecessary to do things like taking breaks, or taking care of yourself, or maintaining your mental health. But this is a problem with your intuitions, not with self-maintenance.
      • A true optimiser focuses on the goal, above all else. If your intuitions tell you you shouldn’t take breaks, and this is holding you back from achieving your goals, then you need to find ways to take regular breaks, and ignore your intuitions.
    • The high-level point here is a distinction I call inner vs outer optimisation.
      • Inner optimisation is “always, in the moment, do the thing that feels optimal”. This is sometimes useful, but often not, and can often close off things like spontaneity
        • This is an especially bad idea when your instincts for “the best action” aren’t great.
        • Remember: Intuitions are not automatically correct - you shouldn’t ignore them, but you also shouldn’t follow them blindly!
      • Outer optimisation, is about planning your life on a much higher level, and choosing the situations you put yourself into. And then in these situations, you could completely ignore inner optimisation, but might still get closer to your goals
    • In practice, I find it’s valuable to be an outer optimiser almost always, and an inner optimiser only sometimes.
      • One useful approach: Explicitly schedule time to review your life at a high-level, like a weekly review. Optimise as hard as you can within the review, and outside of that just live your life in the moment
      • Eg, set yourself a regular break schedule, where your laptop automatically locks during break times. Then, in the moment, you have no choice but to do the right thing, no matter what your intuitions say.
    • Another error: It can be easy to optimise by focusing only on things that are obviously necessary. And cutting everything that doesn’t have an obvious causal pathway to working towards your goals.
      • I think this is often a bad idea, because the world is complicated. Things can be important, without feeling obvious to you.
      • If someone with more experience gives you advice, or there’s a cultural convention, pay attention to it. Even if you can’t see an obvious reason for it, you should have a good reason to ignore it.
  • Objection 2: If you are always optimising, you’ll never encounter new things, and spontaneity. Some of the best things in life, some of the greatest friendships, started from an unexpected chance encounter
    • I think this is a super good point! This is a well known problem in computer science, called the explore/exploit trade-off. Roughly - when you try something new, the expected value in the moment will be lower. But there’s a chance it’s much better than you thought, and if you discover something new and awesome - doing new things gets you information, and information is valuable for achieving your goals
      • Eg, if you want to have amazing restaurant meals, you want to balance between trying new places (exploring) and going to your favourites (exploiting). If you always go to your favourite place, you might be missing something amazing. But if you always go to new places, then you’ll miss out on the awesome meals you could have at your favourite place.
      • A great podcast on this idea
    • It may not feel like information is valuable, but this is a problem with your mind, not with the world. Information is a valuable resource, and a true optimiser will realise and account for this.
    • And if spontaneity is truly your goal, this is definitely something that it’s important to realise, because it can be optimised for!
      • Eg, if you want to encounter new ideas and ways of thinking, ask your friends to introduce you to their weirdest and most interesting friends
      • Eg, if you want to expand your comfort zone, go to events you’d normally never dream of. Like a pole-dancing class, or a meeting of your local right wing party.
        • One idea: Scroll through meetup.com, and pick a random event every Sunday for the next month
      • In contrast: If you value spontaneity and novelty, but don’t really think about this and never go outside your comfort zone, you are not taking actions that achieve your goals.
    • This is another example of inner vs outer optimisation - I find my inner optimiser is often bad at acknowledging the value of information, while my outer optimiser is fine with this.
  • Objection 3: You can’t optimise, because you don’t understand your true goals
    • Eg, Someone who optimises for money, so they work 18 hour days as an investment banker
      • This is a terrible idea. Seriously, don’t become an investment banker, it sounds awful.
    • Eg, Someone who thinks they care about status, so gets the most prestigious PhD advisor they can. But then realises that mentorship is actually more important, and that their advisor has no time for them.
    • I think this is a really important point. My goals change over time, and my goals are complicated and not clear to me. I care about happiness and fulfillment, but won’t take a drug that just gives me the feeling of happiness and fulfillment. I care about learning and knowledge, but only about certain things, in a fuzzy and ill-defined way.
    • The first lesson: Have uncertainty about your goals. You will change your mind over time, and you will make mistakes, and you will learn from them. Be open to this, and be looking out for evidence that you’re wrong.
      • This means you should gather evidence - information is valuable! Do experiments, try new things, ask others for advice, go outside your comfort zone. Notice when you aren’t enjoying things, and ask yourself why.
      • I find introspection is a really useful skill to cultivate here. Eg, If you think money is all that matters to you, try to imagine a concrete future where you have a lot of money, and ask yourself why you care about that. And ask yourself whether anything is missing.
    • The second lesson: Preserve option value. Some decisions we take are irreversible, some are not. And when you don’t understand your goals, the ability to change your mind in future is key!
      • Eg, I currently don’t want children. So it would be useful in the short term to become sterilized. But this would be a terrible idea, because I’d lose the ability to change my mind in future! I’d be trading a great deal of option value for a short term gain, which does not achieve my goals.
  • Objection 4: Optimisation is stressful/If you overthink everything, you’ll never get anything done
    • I think this is an extremely valid point! Optimisation and planning take time and energy. These are a limited resource, and they have a genuine cost. If you spend them to make a small decision 1% better, that’s often a bad trade
    • I think the high-level error here is that a naive optimiser thinks of themselves as a purely rational agent. Your goal is to take the perfect action, and it is a failure to do otherwise. No matter how long it takes to
    • However, humans are bounded rational agents. We have limited resources to spend optimising things,
      • Again, this is something my inner optimiser has issue with! I think this partially stems from my maths background, where a proof of existence, or an intractable but valid algorithm feel like solutions. I’ve found it valuable to learn more about economics and computer science, where these questions of resource constraints are taken seriously
    • This means that you need to optimise how you optimise.
      • Eg, I tend towards overthinking trivial decisions. So I now carry dice everywhere, and use them to randomise things that just don’t matter, like choosing a meal in a restaurant
    • But for big decisions, it is valuable to optimise. And this is something humans are intuitively bad at.
      • Eg, If I’m choosing a career, that’s something that will consume most of my time for the next few years. This is a really important point, and gathering as much information as I can, asking for advice, testing things, and thinking deeply, are all great uses of time.
      • It’s often valuable to do a Fermi Estimate of how big a decision this is. The sheer magnitude of time doesn’t feel visceral to me, because humans are bad are understanding large numbers. But this is a problem with my mind, and my goals live in the world.
    • Further - it’s not enough to just try to optimise big decisions. Your goal is not “spend time agonising over a career”, it’s “choose a good career”. If thinking isn’t going anywhere, and you keep going in circles, this is not bringing you closer to your goal. And you are spending resources for nothing.
      • Uncertainty is unpleasant, and it’s easy to agonise over this. But often it is impossible to reach certainty, and no amount of thinking will resolve this. A good litmus test: “If I spend another hour thinking about this, do I think I will make a systematically better choice?”
    • With regards to “it’s stressful to optimise” - you can also optimise optimisation to be lower stress
      • Inner optimisation is generally more stressful, and consumes more attention. So carving out time to be an outer optimiser, and ignoring optimisation the rest of the time, can reduce the costs a lot. Your goal is not "spend time and energy optimising”, your goal is “be optimised”, and you want to achieve this as efficiently as possible
      • I personally find optimisation super fun, and pretty low stress. If you differ, this doesn’t mean you should never optimise, just that you should focus it on the important things!

Overall, optimisation is hard. It’s easy to fall into traps, it’s easy to confuse proxies for your true goals, it’s easy to neglect important but abstract ideas. And it takes time and energy to do. But I only have one life, and there are a lot of things in it that matter to me. And if I can spend some of this life making the rest of it better, then that is important to me. And when something is important, it matters all the more to be deliberate, to always drive towards the goal. To strategise and ensure I do it well. And every time this goes wrong, that is a chance to learn, and ensure I achieve my goals even better next time.

It’s easy to consider this too much effort. An obligation to feel guilty about it, or a source of stress to ignore. But every single thing you do is an action that brings you further or closer from your goals. Not optimising is simply trading the resources you could have spent optimising, for the world where you follow the default path. I am not happy with the world where I follow the default path, and I choose to strive to do better.

Perhaps that feels like a good trade for you. And for some people, it probably is the right call! But be aware that it is a trade-off you are making.

7

New Comment