Noticing and Overcoming Bias

post by Neel Nanda (neel-nanda-1) · 2021-03-06T21:06:36.495Z · LW · GW · 1 comments

Contents

  Introduction
  Identifying Bias
  Noticing Bias
  Overcoming Bias
  Conclusion
None
1 comment

Introduction

There’s a classic optical illusion about two squares on a chessboard. Square B looks significantly lighter, but they’re really the same shade - my mind is automatically correcting for the perceived shadow.

I don’t think this particular quirk of the eye is that interesting. But I think there is a valuable lesson to take from this illusion. Look at the picture, bearing in mind that the two squares are the same colour, and try to will yourself to see them as the same thing.

If you’re anything like me, this is utterly impossible. My eyes are a lens through which I perceive this picture, and they have a flaw. And to perceive the true world, I can’t fix this lens - I need to learn to work around this flawed lens, and find a different way to understand the ground truth.

This is a post about biases, how I think about them, and how I try to overcome them. And this lesson is key to understanding biases, because my mind is the lens through which I view the world. And biases are the flaws within that lens. And I cannot think my way out of a bias, or overcome this flaw with sheer force of will. It is insufficient to just be aware of my biases and to continue life as normal. To overcome biases, I need to learn when not to trust my own mind, and when to find work arounds to find a different way to access the ground truth.

Having true beliefs is a key sub-goal for essentially everything I care about, and biases are one of the main things holding me back. Overcoming bias is important. But it’s hard. If I want to find truth, there are times when I’ll need to believe things that feel intuitively wrong. Because my intuitions are the lens through which I find truth, and this lens has flaws. This is a difficult skill, and a constant balancing act, and I’m going to try to outline my thoughts on how to achieve this.

Identifying Bias

The first step is finding your biases!

A classic approach here is what I call the Laundry List of Biases approach. There are long lists of biases, with various snappy names and motivating examples. There is value in this, I think these biases are very real, and important to be aware of! And naming things makes them easier to understand and recognise.

But I’m personally not a big fan of this approach. I find this easily frames bias as an academic thing, something to be understood in the abstract, or as something that happens to other people. And, worse, it can make overcoming bias feel like an exercise in ticking boxes - learning a long list of increasingly arcane and obscure biases, and ensuring I’m not falling prey to any of them.

Personally, I find that this is distracting. Biases don’t go away just because I’m aware of them. And so most of my problems come from a few really common biases, that come up everywhere. And thus, to 80/20 this, the vast majority of my effort should go into overcoming those common biases. Often these don’t have snappy names, because they’re obvious and mundane, and don’t feel as exciting to think about. But I think very few people are actually, consistently good at overcoming common, mundane biases. And you should focus on the most important bits and the basics, before trying anything fancy.

For me, the obvious, daily biases are:

• Insecurity: I systematically underestimate my own abilities, how well other people perceive me, how well things are going, etc. My mind jumps to worst case scenarios
• Eg, I am terrible at asking people for favours, or generally making requests, even when it’s an obviously good idea (though I’ve gotten a lot of mileage out of working through this!)
• Defensiveness: I feel a strong drive to be contrarian, and to contradict anything I don’t immediately agree with. Especially if it feels like it comes from an out-group, or attacks me or my in-group.
• Motivating reasoning: Favouring arguments for conclusions that I want to believe - things that are convenient, things that save me from costs or risks, things that fit my aesthetics and existing worldview
• Planning fallacy: I assume everything will take far less time and less effort than it will in truth - most of the time, this is the root cause behind procrastination for me
• Bias towards the default: I want to find reasons to stick to the default actions, or default beliefs, and resist any idea that this should change
• This partially comes from a desire to avoid effort, to justify procrastination and why I can avoid something aversive
• This partially comes from loss and risk aversion - I want to find reasons to avoid needing to go outside my comfort zone

Exercise: Before reading on, set a 5 minute timer and list your obvious, common biases. Bonus points for examples!

I expect the most important biases to vary a lot between people, but I think we’re all aware of our’s on some level, even if they aren’t at the top of our mind. It may be hard to notice a specific instance of a bias, but we have a lot of data about our most common biases

Even if these don’t feel immediately obvious, there are other ways to generate them!

• Any time I screw something up: I make a mistake, am proven wrong, or lose an argument; there’s often some bias at the root of that. Analysing this can point me in the right direction!
• Seek feedback, and listen to criticism. Other people are biased too, but they have different biases. By combining their warped view of the world with mine, I can often identify new flaws
• Note: I highly recommend working on defensiveness before or in conjunction with this!
• Finally, noticing biases in other people! It’s much easier to spot a bias from the outside than from within.
• But, rather than going for the smug sense of superiority from pointing out somebody else’s bias, use this as just another data point! The people you surround yourself with are generally pretty similar to you, so their biases will correlate a lot with your’s.
• What are the common biases you see around you, and how do those manifest in your life?

Noticing Bias

But it’s not enough to just understand a bias, or to have identified which ones matter most. Biases matter most in the moment that you make an error. But in the moment, a bias feels exactly the same as a true belief. And so, you need to learn how to recognise and notice a bias on the five second level [LW · GW]. Most biases matter most when I am operating on autopilot, and I can only make progress when I turn off autopilot, zoom out and consider things. And in order for this to be possible, I need to find a way to automatically turn off autopilot

The main skill I find useful here is Noticing [LW · GW]. I recommend reading the linked post for my full thoughts on how to implement this, but roughly, noticing breaks down into finding a clear, salient trigger that occurs with that bias. And then creating a reflexive alarm in your mind to take that trigger seriously - something is wrong, and your default reactions are inadequate.

Another route, is to focus on situations where overcoming bias is most important. When I’m planning, or when I’m doing anything high-stakes or with significant consequences. Or the situations where I’m most likely to be making a mistake. When I’m disagreeing with somebody I respect, or in a heated argument. In these situations, it can be good to zoom out, go through my list of common biases, and try to figure out which ones are triggering.

A good trigger is something simple, atomic and concrete - it needs to be simple so that you can automatically recognise it in the moment, rather than something nuanced that would require processing, like “when I’m committing neglecting data that’s systematically not in my sample”. For me, some good triggers are:

• Justifying why I don’t need to do anything - a sense of lethargy and resistance, which feels obviously correct but I struggle to put into words
• Planning - any time I think about my future actions (seriously, any time. I suck at planning)
• Defensiveness - anxiety and discomfort, a sense that somebody must be wrong, and feeling internally like I’m searching for reasons why

Exercise: Pick one of your common biases, set a 5 minute timer, and list examples of that bias triggering. Extract out common cues and contexts from this, and design a trigger

Overcoming Bias

So, you’ve now learned how to identify a bias, and how to notice it in the moment. I still have a false belief! Sometimes I get lucky, and the bias simply dissolves under scrutiny, but sometimes not. What can I do about this?

And recall, this is hard because biases feel true from the inside. In order to overcome a bias, I will need to run with a belief that doesn’t feel true.

My main tool for this is to form policies. These are rules that I try to follow when I notice a bias, as a default action. This means I follow the policy for the sake of following the policy, rather than because I evaluate the situation and decide it’s necessary. The entire point of a bias, is that it matters most when it doesn’t feel like a bias. And so in order to overcome it, I need to find ways to robustly change my actions, rather than being vulnerable to lying to myself.

I have a few useful categories for thinking about policies:

• Counter-biases: Systematic biases in the opposite direction
• Eg, when I notice myself feeling tempted to Actually Do Something, can’t see any costs, but feel reluctant, just do it!
• Eg, when I’m generating reasons why I’m less capable than a trusted friend thinks I am and they’re pushing back, just accept that I’m wrong.
• I think it can be dangerous to try to overcome biases with counter-biases in general, but I find these extremely useful for resolving borderline cases of major biases - if the bias is big enough, then anything that feels borderline would be obviously wrong without the bias.
• And ultimately, if you have a systematic error, shifting the mean should resolve that
• Reframings: Different ways of thinking about a situation
• Eg, when I have feelings about something politically charged, come up with an analogous situation that inverts right and left and see if that changes anything
• Eg, when struggling with a decision, imagine I was advising a friend about the decision
• Eg, when intending for my future self to do something, ask whether I’d be surprised if it fails to happen
• Eg, when I feel resistant to a suggestion that will involve effort and has some risk, assume the action went incredibly well, and try to flesh out what that world might look like and why it’d be awesome
• Alternate ways of finding an answer
• Eg, when evaluating something about myself, just ask a friend for their opinion and trust that
• Eg, when estimating how long something will take, ignore my intuitions, look at past data and take the average time

Exercise: Set a 5 minute timer, and generate policies that could help with each of your biases

Conclusion

In conclusion, I think overcoming bias is an incredibly important skill. But it’s not something that will just happen by default. And it’s not sufficient to just identify and understand your biases.

There will be a few really common and important biases that come up all the time, and it’s worth focusing your effort on really getting these down. A bias is something you need to notice in the moment, and this requires finding appropriate triggers and cues. And it’s hard to overcome them by applying discretion and judgement - you need to find policies, rules that can help you make better decisions and form better beliefs.

And remember, these policies won’t always feel sensible, or justified! And there’s a skill of noticing when a policy is actually a bad idea. But to overcome bias, sometimes you’ll need to stick with rules your past self set, even if they don’t feel necessary or important in the moment.

And if you’ve read through this post, skipped all the exercises, but the ideas in here resonate, I highly recommend going back and trying them out :) We all have our own personal biases, and you won’t be able to make progress on them without trying!

1 comments

Comments sorted by top scores.