Say Wrong Things

Gordon Seidoh Worley24 May 2019 22:11 UTC
115 points
13 comments4 min readLW link

There are many ways you might approach being less wrong.

A popular one is to make fewer wrong statements; to say fewer wrong things.

Naively it would seem this is a recipe for success, since you just say more things that are true and right and fewer things that are false and wrong. But if Goodhart has anything to say about it, and he does, you’ll find ways to maximize the measure at the expense of the original objective.

Assuming the real objective is something like “have a more complete, precise, and accurate model of the world that better predicts the outcome of subjectively unknown events”, then we can quickly see the many ways Goodharting can lead us astray if we focus too much on appearing less wrong. We might:

  • make fewer claims than we could, pulling us away from completeness even as we appear less wrong;

  • make weaker claims than we could, pulling us away from precision;

  • and, a perennial favorite, filter the claims we publicly make so we appear less wrong than we really are by hiding our least confident claims.

The first two can be corrected with better calibration, that is by making statements with confidence intervals or likelihoods that proportionally match the observed frequency of correctness of similarly confident claims. But simply suggesting someone “be better calibrated” is not a motion they can make; it’s an outcome of taking actions towards increasing calibration. As good a place to start as any for improving calibration is the forecasting literature, if that’s what you’d like to do.

The third is more tricky, though, because it’s less directly about claims being made and their probability of correctness and more about social dynamics and how you appear to other people. And for that reason it’s what I want to focus on here.

Appearing Less Wrong

I’ve met a lot of people in my life who are experts at not looking as stupid as they are.

That’s kind of harsh. Maybe a nicer way to say it is that they are experts at appearing to be better at making correct predictions about the world than they actually are.

Some of their techniques are just normal social tricks: projecting confidence, using social status, the ever-abused term “gaslighting”, and other methods of getting people to believe they are right even when a more careful examination would reveal them to be mistaken. These are people we all love to hate and love when we can call them on their bullshit: overconfident academics, inflated politicians, self-important internet intellectuals, and those people whose idea of social interaction is to say “well, actually...”.

But there’s a way to avoid looking stupid that is more pernicious, less amenable to calling out, and that subtly drags you towards local maxima that trap you mountains and valleys away from more complete understanding. And it’s to shut up and not tell anyone about your low confidence beliefs.

It is extremely tempting to do this. Among the many benefits of keeping low probability claims to yourself:

  • you have a high accuracy ratio of publicly made claims, making you look right more often when observed;

  • you say only things that, even when wrong, turn out to be wrong in conservative ways that still make you look smart;

  • and you accrue a reputation of being right, usually conferring social status, which can feel really good.

The only trouble is that this approach is too conservative, too modest. It’s easy to justify this kind of outward modesty as keeping up appearances in a way that is instrumental to some bigger goal, and you say to yourself “I’ll still make low probability claims; I’ll just keep them to myself”, but down that path lies shadow rationality via compartmentalization. You can try it, but good luck, because it’s a dark art that hopes to do what human brains cannot, or at least cannot without some sufficiently powerful magic, and that magic traditionally comes with vows not to do it.

Meanwhile, out in the light, finding models that are better predictive of reality sometimes requires holding beliefs that appear unlikely to be true but then turn out to be right, sometimes spectacularly so, although semper caveat, all models are wrong, but some are useful. And then you have to go all in sometimes, exploring the possibility that your 10% guess turns out to be 100% correct, minus epsilon, because if you don’t do this you’ll do no better than the medieval Scholastic holding to Aristotelian physics or the early 20th century geologist ignoring the evidence for continental drift, forever locked away from taking the big risks necessary to find better, more accurate, precise, and complete models.

Okay, so let’s say you are convinced not to try so hard to appear more right than you are. How do you do that?

Say It Wrong

So I suppose it’s nothing so much harder than just telling people your claims, even if you have low confidence in them, and seeing how they react, although depending on the circumstances you’ll probably want to adequately explain your confidence level so they can update on it appropriately. The trouble is getting yourself to do that.

I can’t change your mind for you, although thankfully some folks have developed some techniques that might help if you’re not interested in over-solving that problem. What I can do is point out a few things that might help you see where you are being too modest, nudge you towards less modesty, and create an environment where it’s safe to be less modest.

  • Look for the feeling of “pulling your punches” when you are telling people your ideas.

  • Ramble more and filter less.

  • Alternatively, more babble less prune.

  • Worry less about how you look to others.

  • Relatedly, increase your ability to generate your own esteem so you less need to receive it from others, giving you more freedom to make mistakes.

  • Encourage others to tell you their half-baked ideas.

  • When they do, be supportive.

  • Take a collaborative, nurturing approach to truth seeking.

  • Play party games like “what do you think is true that you think no one else here also thinks is true?” and “what’s your least popular opinion?”.

  • And my personal favorite, write up and share your immodest, lower-confidence ideas that you think deserve exploration because they have high expected value if they turn out to be right.

I suspect that the real reason most people try too hard to appear more right than they are is fear—fear of being wrong, fear of looking stupid, fear of losing status, fear of losing prestige, fear of losing face, fear of being ostracized, fear of being ignored, fear of feeling bad, fear of feeling lesser. I see this fear, and I honor it, but it must be overcome if one wishes to become stronger. And when you fear being more wrong, you will be too careful to ever become as less wrong as you could.

To sum it all up pithily:

To become less wrong, you must give up being most right.

Gordon Seidoh Worley24 May 2019 22:11 UTC
115 points
13 comments4 min readLW link