Learning From Surprise

Each surprise is an opportunity to update our model of the world to be more effective in the future at meeting our needs.

But not all surprises or errors are accurate or useful. If the brain learned from all errors indiscriminately, it would be as ineffective as if the brain never learned anything at all. 

The brain instead goes through a process called “Bayesian Updating” – it weighs the new information against its prior beliefs about the world, and only updates its model if it has enough confidence in this new information. Otherwise, it rejects new information and sticks with its prior mode. 

To illustrate how this learning process works, let’s look at a quick example.

Example: Your Favorite Restaurant.

Every time you go to your favorite restaurant, you order the same dish — mac and cheese. Every time, it tastes the same — exactly what you expect and want.

This past Friday, though, you showed up, ordered your mac & cheese, and woah! It tastes different. It tastes a bit bitter, not as creamy. Surprise!

If you have no more information, you can make one of two choices:

1️⃣ Learn from the surprise: this restaurant is no longer good, and I should never come back, or

2️⃣ Ignore the surprise: it’s a fluke — maybe the chef is out sick, or they couldn’t get their typical supplies — I’ll come back next Friday and hope my mac & cheese is back to normal.

Option 1 immediately updates your model of the restaurant: it’s no longer good; Option 2 may change how confident you are in the restaurant being good, but doesn’t update your model of the restaurant entirely.

The next Friday, you come back again and order your mac & cheese. Once again, it’s not the same. Now with your confidence diminished from the prior week, you have to make the same choice — learn from or ignore the surprise. This second surprise is a bit harder to ignore. A pattern is starting to form.

You come back for the third week, and while you’re hoping for your favorite dish, you’re almost expecting to be disappointed. But, surprise! 😯 The pasta is back to its normal, delicious self! Thank goodness you stuck with it and didn’t update your model of the restaurant too quickly.

Now let’s add one more twist: when you ask the host at the restaurant whether the chef was gone for the past two weeks, they reply with one of the two following responses:

a) Yes, she was out sick, but he’s feeling much better now. Hopefully it didn’t disrupt your experience too much.

b) Yes, she’s opened a second restaurant and hired a second chef — she’s planning to split her time between the restaurants from now on.

In scenario a, you’re pretty confident things are back to normal, and your mac & cheese is safe. Scenario b, however, changes everything. Now each time you come to the restaurant, it will be a dice roll of whether you get good or bad mac & cheese.