You’re sure it’s a dog. And you’re sure it is a big dog.

The street is residential, it’s the middle of the afternoon, and no one is about. The bark from behind you, just beyond your left shoulder, is loud - thunderous, even. It can only be a monstrous, drooling guard dog.

You turn around slowly, heart racing, sweat beading, and…

A tiny black pug, no bigger than your hand, is backing slowly away across a lawn three yards away, barking up a storm but retreating and shaking as it goes.

You laugh, and carry on, not realising that you just learned something. What just happened, you see, was a prediction error.

“The immediate effect of this prediction error is to update our belief from thinking there is a big dog to there being a tiny dog,” explains Peter Kok, a senior research fellow at the Wellcome Centre for Human Neuroimaging at University College London (UCL). “The more long-term effect is that we update our model of the world: loud barks don’t always come from big dogs.”

In neuroscience, a so-called “prediction error” occurs when an expected event fails to materialise. Such incidents cause a range of metacognitive functions to leap into action that force us to re-examine our initial flawed expectations and ask ourselves what we can learn from our errors.

Floris de Lange, a professor at the Radboud University in the Netherlands, reveals that prediction errors can arise from a number of different brain stimuli. They can be sensory - for example, a visual surprise such as a sudden flash; or they can even be semantic. He gives an example of the sentence “the soup was too hot to cry” - the last word elicits surprise because our brains had predicted a different end.

Kok says it’s the system by which we educate ourselves about life. “Making predictions and updating our beliefs based on prediction errors is a computationally efficient way of making the best guess of what’s out there in the world,” he argues.

Efficient learning? Impact on memory? A research-backed reason to create a bit of drama in your classroom? What’s not to like?

Many teachers already build surprise into lessons - for example, teasing out predictions for what happens in the class novel and then dramatically revealing the truth; offering multiple possible options for the outcome of an experiment so that the actual result confounds what many expected; directly anticipating misconceptions.

What this research has the potential to do is make the creation of surprise a more purposeful and planned part of the classroom - influencing practice in the same way as retrieval practice has over the past few years: codifying and fine-tuning something many teachers are already doing.

So surely, prediction error is ripe for a conversion into pedagogy? Well … maybe.

Prediction errors are closely tied to dopamine - the chemical in our brain that many associate with the feeling of pleasure and reward, but which also plays a key role in movement, motivation, memory and focus.

If a pupil enjoys a lesson, then dopamine comes into play. It reinforces memories by strengthening neural circuits that are associated with the “enjoyed” information.

As Joydeep Bhattacharya, director of research in the department of psychology at Goldsmiths, University of London, told Tes in March, there is a “huge body of literature” showing how dopamine is crucial for learning, and it also suggests “prediction errors” are key.

“When we predict something completely correctly, there is no change in the activity of the dopaminergic neurons in the brain - but if something unexpected happens, then they respond,” he explained. “For example, if an expected prediction fails to materialise, they quieten down, and when something happens that goes beyond expectation, they fire up.”

This leads to the hypothesis that dopamine is strongly related to coding a reward for prediction error. And that has potential benefits to learning.

“Dopamine neurons code for these prediction errors, and thereby pave the way towards updating our knowledge through a process known as ‘reinforcement learning’, which is fundamental for adaptive survival in this ever-changing world,” Bhattacharya says.

De Lange, who is also principal investigator in the Predictive Brain Lab at the Donders Institute for Brain Cognition and Behaviour, adds that “prediction errors are very useful, because they indicate that the model isn’t fully accurate and needs updating”.

He continues: “Surprise often activates both the reward system and the hippocampus - a key structure involved in learning.”

Hanneke den Ouden, associate professor at the Donders Institute, pushes that further, arguing that, in theory, educators could tap into prediction errors to improve students’ learning.

“If [teachers] make sure there is an expectation [among their students] first, then there can be a prediction error, and learning will occur,” he says.

Such is the preoccupation with research-informed practice in education that you would not be shocked if there was a memo out to all staff by the end of that sentence calling for “at least three surprising moments in every lesson”.

It’s likely it would be eagerly enacted, too: teachers who advocate for frequent low-stakes testing could easily apply the theory to their way of working (wrong answers being a pure form of prediction errors), as could those more keen on experiential, project-based approaches (the element of surprise a key feature of this approach to pedagogy).

However, just as with a lot of the research finding its way into schools, it’s not as simple as it first seems. As Jared Cooney Horvath argued in Tes in June, taking evidence from academic research and ensuring that it has an impact on teaching practice in schools is complex. Many a theory has suffered on its route into the classroom - the road is littered with the burned-out shells of teacher attempts at growth mindset, dual coding and cognitive load theory, to name but a few - and prediction error is likely to be just as fiddly.

Michael Hobbiss, a former secondary school teacher who is now completing a PhD in educational neuroscience at UCL, says that while ideas about the use of prediction errors in pedagogy are “very interesting”, “they are a fair way off being ‘classroom ready’.

“In other words, they don’t yet point to clearly definable strategies that we can implement in the classroom - and for which we can be confident that they are more worthwhile than any of the other possible strategies that a teacher might use,” he says.

Instead, what we have is a series of working hypotheses about how this research might impact teaching.

For example, one area of interest is the effect of working a prediction error into your lesson in a purposeful, planned way.

Marlieke van Kesteren, an educational neuroscientist at Vrije Universiteit Amsterdam, suggests “starting each day or lesson with something that yields a prediction error … will probably trigger dopamine release that can aid learning”.

But she adds: “We are still questioning, though, whether it helps all kinds of learning and to what degree - type of memory, how long the effect lasts … and when it becomes too much.

“You could use it in your teaching, but … the brain quickly habituates to a novel situation or a novel stimulus so it will only work [in the relative short term] I think.”

Too much novelty and surprise, van Kesteren says, might even “overflow” the brain, and have a negative impact on learning. “In my experiments, participants learn stimuli that are congruent or incongruent with prior knowledge,” she explains. “You would expect the incongruency to generate a prediction error and hence enhance learning, but I have never found this. In fact, these memories are generally very poor, and I have related this to the fact that there are just too many prediction errors in combination with learning events happening in a short time-period.”

What her evidence suggests so far is that when prediction error is used as a technique sparingly, you do get positive results. So, while starting a day or a class with something that contradicts expectations might sharpen the mind, too many prediction errors in a short time-period can “easily lead to the reverse effect, so you have to be picky”.

“In educational settings there is often a lot to be learned in a short time frame, so finding the right balance is key,” she says. “Experiments that only show an occasional incongruent item have more success generally, so you could use this for one specific learning event that is generally hard to learn, for example.”

Helen Nasser, a research fellow at the Florey Institute of Neuroscience and Mental Health, affiliated with the University of Melbourne in Australia, offers a more pastoral potential application: reducing pupil anxiety.

“If [for example] a test is performed in a particular room … and a student happens to do poorly in that test, [there is an] aversive outcome. Then the student might show anxiety-like behaviour whenever they enter that room,” she says.

Prediction error is at work here, she explains, because before getting test results back, the room in which the exam was taken would likely not have been part of a pupil’s rationale for a strong or weak test performance.

“When they see the bad grade, they feel dread, so what actually happens is an aversive event,” says Nasser. “The discrepancy between what was expected and what happened is therefore large, and they learn a lot about that experience.

“So next time they go to this room, they will experience a significant amount of dread. This can be problematic for the student as this expectation might decrease their ability to concentrate which, in turn, might again cause them to perform poorly on another test.”

Being aware that this is happening might enable teachers to mitigate the issue through exposing the student to that room in different contexts. Repeated testing in that room may also help. If the next time they sit an exam in that room they receive a good grade, there is another large discrepancy in what they expected, says Nasser: “Then they learn that the test room is not as anxiety inducing.”

This same process may be useful in managing behaviour, or in looking at engagement - how discrepancies in previous experiences influence a pupil is a factor that may need more attention.

Finally, Geoffrey Schoenbaum, a senior investigator on the Intramural Research Program at the US National Institutes of Health, argues that the theory may become important in the ideology of teaching - more specifically, by influencing how teachers view their students.

Understanding that your pupils are “active agents, with beliefs, and that how you teach against those beliefs can either facilitate or hinder your success - due to the concept of a prediction error - is probably not a bad thing to consider if you are a teacher,” he says. “Although I doubt it would modify anything a good teacher does in a complex classroom setting - they are already probably doing things that take advantage of this - perhaps when things are not going well, thinking in these terms would help.

“If teachers present their material in a way that does not violate those beliefs, then their teaching will not be very successful.”

Understanding your pupils well enough to not only know their academic weaknesses and where misconceptions lie, but also their bias, ideologies and quirks is all useful in general for a teacher, but the argument here is that when you consider prediction error, it is even more important.

So where does that leave prediction error’s place in the classroom?

It’s likely that our knowledge of prediction errors will continue to expand in an academic sense and that, as a result, the potential educational applications will also expand. How the teaching profession chooses to engage with that process will be a good test of how far it has come in becoming better able to forge a mature relationship with research findings and their translation to the classroom.

For the moment, Schoenbaum believes that “things like sleep, food, exercise and making sure lessons are entertaining … are all more important in the context of classroom teaching than prediction errors”.

But it is certainly something to keep an eye on - not just because it might surprise us and turn out to be more important than we think, but in case it is shown to be less important yet still gets a foothold in how children are taught. Unfortunately, judging on past experience, the latter would not be as much a surprise as it should be.

Chris Parr is a freelance writer

This article originally appeared in the 27 September 2019 issue under the headline “The (learning) element of surpr!se”