Surprise, Learning, and the Predictive Brain
By: Curtis Kelly
Nicky and Heather just gave us a delightful look at surprise as an unexpected spice for your classes. I will continue their discussion, repeating part of it (by accident. Really!) and giving more of a perspective on what is happening inside the brain. No rabbits, but there is a spider.
Surprise is not just some random phenomenon of the brain. It is there for a reason, a good one, and it has a lot to do with learning. The key to understanding surprise is knowing it is a prediction error. If you have been following our Think Tanks for the last four years, you probably know that our Think Tank team is big on predictive processing (archived TT search results). The theory of predictive processing holds that, rather than taking in extensive sensory data to figure out our current environment, we instead predict what is most likely to be there according to previous experiences of similar situations. After taking in just a smidgen of sensory data, we fill in the rest automatically.
For example, imagine you are in class. You say, “Open your books to page 18,” and the class starts moving. You see them open their textbooks, but then one student who isn’t moving at all catches your attention. You look her way and see there is no book on her desk. The student says, “Sensei…,” and you already know she is going to say, “I didn’t bring my book” (60% probability), “I left my book at home” (30%), or “I lost my book” or some other still expected but less likely utterance (10%).
So how is prediction happening in this situation?
Of course, your prediction of what the student will say even before the words come out of her mouth is part of it. This is what’s known as predictive language processing, and we do it all the time. It also explains why our language has grammar. Grammar aids prediction, limiting the possibilities of what words will come next. Once the student says, “Sensei,” and “I left…,” the probabilities of the following words, because of the situation and grammar, become “my book” (90%) or “my house” (10%). Grammar greatly narrows what can come after “Sensei, I left my book” as well, with basically just three potential words: “at” (80%) “in” (10%), or “on” (5%).” and so it goes for what comes after those words as well. Our brains use probability statistics to figure out likelihoods: “…the human brain can be seen as a Bayesian inference machine” (2017, para 1).
Our predictive language processing gives us huge advantages. Being able to predict what words finish an utterance frees your mind from the need to do a ton of language processing and lets you use the mental resources freed up to deal with the situation.
I mean, think about it. The traditional model of how we listen–that we process each phonetic unit one at a time before ascribing meaning–is almost laughable. If, to get the words, you had to listen to each phoneme of an utterance one by one, some of them blended together or barely comprehensible, and then figure out how each word fits into a sentence, and only after you get the full sentence, process it for meaning, you’d spend your social life frozen in linguistic computation for minutes or hours after each utterance. Instead, our brains take a proactive approach. They look at the priors (a distillation of previous experiences) and grammar and predict what will be said, usually with just a tiny bit of input to get the predicting started: “Sensei” in our example above.
So, that explains how we processed the student’s utterance, but now think about how much processing was going on in the rest of this situation. When the rest of the class started moving after you said, “open your books to page 18,” did you have to look at each student one by one to see if they were opening a book? Did you have to direct your attention to the books, one after another, to verify each one was the textbook? (I’ve taught in places where this was necessary!) Did you have to walk around the room to see if each student opened to page 18? No. Just perceiving motion, the smidgen of sensory input, let your brain fill in the rest. In fact, your higher cortical areas are sending more sensory signals downwards through the cortical hierarchy than they are receiving upwards from your senses. The downwards signals, representing the predictions, cancel out upwards bound sensory signals that match them, letting only the non-matching perceptions (the prediction errors) through (source). Your brain tells you what you are seeing, not your eyes.
Sometimes the emotional components of models are so strong, especially those related to danger, that we override our Bayesian statistics and see or hear something that is not even there! That is why police might shoot down a boy in a hoodie pulling out a phone. Their strong instinct for survival makes them actually see a gun1. That is why your child’s chocolate ball rolling across the kitchen floor becomes a spider. That is why your spouse talking in a quiet voice on the phone makes your antenna go up.
1 This does not justify such shootings, but shows that it is not just a matter of moral ineptitude. Sensitivity or discrimination training will not change actually seeing a gun in a situation with less than 0.4 seconds to respond without being shot first.
How does this relate to surprise and learning? Well, imagine rather than the expected, your student had followed “Sensei…” with: “I have cancer so I won’t be taking your class anymore.” You’d be stunned, wouldn’t you? And so would everyone else in the room since this violated their own predictions. In this case, your priors failed you. They were completely wrong in making predictions and left you hanging. That’s prediction error and what surprise is all about. When that happens, your brain releases neurotransmitters like noradrenaline and cortisol to make you focus your attention on the situation (source) and these, in turn, lead to feelings of excitement, anxiety, or alertness. It might also activate the reward system if the surprise is pleasant. Along with your bilateral inferior frontal gyrus (attention), your brain activates your hippocampus associated with memory (source), and releases dopamine, a neurotransmitter that causes curiosity and reward (source). Dopamine also causes new synapse formation. After all, your models were wrong. They need to be revised. Stephen M. Ryan explains this error-driven revision of models in a much simpler way: “It is learning!” He continues:
This is learning: using the unexpected elements of our environment to modify our expectations so that the next time we encounter this kind of environment, we can make better predictions, using our bandwidth more effectively, more efficiently. This is learning. This is learning about life. This is learning about our immediate environment. This is learning about our social environment, and the same principles apply to learning language. (iTDi Course video: 16:04… written out here just in case you didn’t click on the same link to this fine bit in Nicky and Heather’s article.)
That intense emotional reaction to prediction error only happens when the error is about something meaningful to us, not for the small stuff, and for good reason. Across our evolution, surprise often meant life or death: a springing predator, a crumbling slope, or a violent kinsman. Learning, deep learning, was critical to survival and social success, and in many ways, it still is today. If you meet that student again, it is good to remember that she has cancer. But even with little things, our ability to make accurate predictions aids us at every level.
In short, surprise increases the potential to learn. Those neurotransmitters mentioned above arouse curiosity, increase engagement, and prime the brain to remember whatever information was present at the time. The reward system is activated, creating curiosity, and this makes the learners want to know more. Incorporating surprise facts, discovery-based methods, unexpected activities, and other novel experiences makes learners focus and engage, creating the “ever-sought teachable moment.” Retention is enhanced as well.
So, to repeat what I wrote at the outset, “Surprise is not just some random phenomenon of the brain.” It is the basis of all learning and a wonderful resource for making your classes more satisfying and effective.
Keep that in mind as you read through the surprises our contributors experienced, and how they helped these teachers grow.
Curtis Kelly (EdD) was surprised in class every day. What wonders our learners are, even the difficult ones.