Anyone who studies psychology and looks to improve their thinking has come across cognitive biases: our tendencies to think in ways that deviate from rationality or good judgment. It’s worthwhile to peruse them and understand them, but it would be a mistake to claim that we have to avoid them completely in order to be a rational person.
As an example of what I mean, I want to explore the difference between learning and confirmation bias, and whether there is in fact any difference. When you look at the processes involved in each, as I’m about to outline, it’s clear that there is some strong similarity between the two. Looked at this way, I think there are some interesting takeaways for anyone interested in constant learning and self-improvement.
What is Learning?
There are a lot of different models of the learning process, but let’s take a simple one. In this version, teaching and learning involve a process of advancing one’s state of knowledge through the mastery of small topics (or lessons) that get successively harder or more complex.
Take math skills as an example. Students generally progress from counting and simple arithmetic to functions and algebra, then basic geometry, before finally moving to more complex formulas, trigonometry and calculus.
If you were a tutor responsible for ushering a student along that path, your teaching process would look something like this:
- Assess the student’s current state of knowledge
- Give lessons and assignments that are just within grasp of their current state of knowledge
- Repeat
A good tutor will devise every lesson as in a sweet spot of not too close yet not too far from the student’s state of knowledge (i.e. proceed from 8th grade algebra to 9th grade geometry, not straight to college level calculus). Too close and the student is not learning fast enough, too far and they will struggle to keep up.
In this way, each lesson builds on prior knowledge in as efficient way as possible. Each lesson should confirm and incrementally advance the student’s state of knowledge. I argue that this doesn’t read much different from the standard description of confirmation bias.
Confirmation Bias
Let’s examine a few of the common definitions of confirmation bias:
- “A confirmation bias is a type of cognitive bias that involves favoring information that confirms previously existing beliefs or biases.” [link]
- “[Confirmation bias] is the tendency to search for, interpret, favor, and recall information in a way that confirms one’s preexisting beliefs or hypotheses, while giving disproportionately less consideration to alternative possibilities.” [link]
A classic example would be someone who believes in astrology – that the positions of stars and planets have predictive bearing on earthly events. If this person is exhibiting confirmation bias, they will interpret any correlation between an astrological event and a borne-out prediction as evidence that astrology works, while avoiding possible conclusions to the contrary.
Putting Them Together
Compare those definitions of confirmation bias to our prior statement about efficient tutoring / learning:
- Tutoring and learning involve the presentation of information in such a manner that confirms and extends the students’ existing knowledge in as efficient way as possible.
See the difference? No?
With such a close definition, it’s actually hard to distinguish when you are really doing one or the other:
In some sense, when you’re looking to make good headway in a new subject area, you want there to be some confirmation bias. Otherwise you’re not learning in an integrative way, you have random facts disconnected by miles of ignorance, and your payoffs to learning might be too far out to be immediately useful.
The Sweet Spot
We can map any point in that Venn diagram to one’s “slope of learning” as they try to progress in a subject area or grasp a new piece of information. We can base this slope on the subjective rating by the individual for how easy or hard it is to understand the new piece of knowledge.
We’ll use a ski slope metaphor for how cognitively burdensome it is to integrate a new piece of knowledge or fact:
If you stick to the green Hard Going slope you will make slow progress, but the difficulty may help reassure you that you’re not just combing over territory you’ve already covered. If you stick to the Easy Going path, you may move a lot more quickly towards tangible intellectual progress, but it is likely to be incremental at best.
Confirmation bias is still learning, it’s just learning in a way that most economizes cognitive effort.
So as you learn new information you should ask yourself: how cognitively challenging was it to integrate that new piece of knowledge? What slope do I perceive myself as on? You can use this as a litmus test for how likely it is that you’re doing real hard learning, or simple confirmation.
Knowledge Economies of Scale
There’s another paradox that comes after this process picks up steam. Take someone like Tyler Cowen. Wikipedia identifies him as an economist, but as his interviews and blog make clear, his knowledge is much broader than that, covering a wide field in areas like sociology, psychology, and global culture.
We might imagine that his personal “Knowledge Cloud,” inside the space of all possible knowledge, looks something like this:
Let’s pretend there are economies of scale at work in knowledge acquisition. This means that the more knowledge you have in your initial repertoire, the easier it is to bring in knowledge that directly neighbors it.
In the case of some new idea at the crossroads of sociology and economics, Tyler Cowen’s effort to grasp that idea might look like this:
Now take someone else with a very different background, like Elon Musk. Elon’s knowledge cloud, and his effort required to grasp the same concept, might look like this:
Now we have a scenario where the “ski slopes” for each individual differs greatly, for that particular new idea:
Tyler’s slope is much closer to what we’d be tempted to call confirmation bias. But is it still confirmation bias? Is it a bad thing? Can we accuse domain experts that publish and adopt research at the cutting edge of their field as simply engaging in confirmation bias?
Bias Bad, no Bias Good
I don’t pretend to have the answers here, though I think this is a much richer description of what we mean when we talk about confirmation bias. Confirmation bias is always at or near the top of any list of cognitive biases, and it makes sense why this is so. We are learning all the time and we’ll mostly gravitate towards the easy slope, towards learning with the least effort or in the most economical way.
If people continually economize their learning, they will build models solely based on what’s the least cognitively burdensome knowledge for them to acquire. That’s a big way to miss out on other important models, and why we should recognize confirmation bias as a potential pitfall.
But if we take the simple approach that says “I shall not commit any of the biases on this list,” we are misunderstanding what that list is telling us. Behind every “bias” is some truth to the way we operate, some core insight about how we evolved as intelligent beings. It would be a shame to turn our backs on thousands of years of evolution just because we have a fetish for what’s supposedly rational. Instead of fighting ourselves and seeking to rid oneself of biases, you might try to take the idea a little deeper and use it to work with, instead of against, your brain.
Joseph Kelly's Blog 