On Confirmation Bias

Last week I ran a simple experiment on Twitter, inviting people to find a pattern in a three-number sequence. The numbers were 6–12–18. In order to solve it, people had to do two steps. First, propose their own sequence of three numbers that would conform to the pattern they thought I was thinking of. They could propose as many three number sequences as possible, and for each one I would tell them if it fit the pattern I had in mind or if it didn’t. When they were satisfied that they understood the pattern, they could guess what it was, but only got one guess.

As I predicted, people had an extraordinarily difficult time guessing the pattern that I was thinking of. Over the course of about four hours, I’d say about 40 people sent me DMs with guesses. Nearly all of them submitted three-number sequences that worked. All but two of them guessed patterns that were wrong.

The pattern I had in mind was simple: any three numbers in ascending order. Overwhelmingly, people who submitted answers suggested it was something more complex: intervals of 6, multiples of 6, the third number being the sum of the first two, etc. Almost everyone only submitted a single number of sequence of their own before attempting to guess the pattern. Since their submissions all fit the simpler pattern I had in mind, they all assumed their more complex pattern was right, and then nearly all of them processed to…