HomeScience EducationScience TeacherMeeting 18 of the Inclusive Science Education group

Meeting 18 of the Inclusive Science Education group

Notes from the Inclusive Science education group on 15th November 2023. This was the 18th meeting of our group.

A summary of the presentation.

Adam Wray gave a presentation to the group on predictive brain theory – application to teaching autistic students.

Predictive brain theory is at the cutting edge of cognitive science, it’s a potential link that explains much about how the brain works and is potentially very useful for autistic learners. Adam started by explaining his background and some of the books he has based his understanding on – and a recommendation for ‘Autism and the predictive brain’ by Peter Vermeulen 

We are all familiar with Willingham’s model of memory with working memory and long-term memory.

Adam started by explaining how a neurotypical brain processes seeing a coffee cup. This is interpreted as a series of pixels at the 1st level of processing. We recognise lines & shapes and using schema (information) in our long-term memory we start to recognise these shapes together as a cup of coffee. There are several levels of processing in seeing this image.

Using a predictive model, the brain uses probability to predict what you are likely to be seeing – this isn’t using the senses, it is just a prediction at this stage. We then deconstruct this model all the way down to a pixel map of what we should see if the brain prediction is correct. The brain then compares this prediction to what we see and checks if this is what we expect. 

If we don’t see what we expect, the brain adjusts the prediction of what we are seeing. This is quite a different model to the one that many people are familiar with. This model is being processed at multiple levels which constantly test this prediction.

Checking this model with computing, this allows us to have a much more rapid appreciation of what is happening around us, and this model explains our rapid reaction rates and explains the bandwidth that the brain is working with. This prediction mechanism is what we would refer to as our working memory. 

‘Priors’ feed into the prediction – priors can be what’s just happened, or they can come from our long-term memory.  If there is a prediction error, the differences are communicated back up to the prediction engine to correct the errors.

Sensory input from the real world is noisy – Adam gave the example of a pure tone, and when he played the tone with noise our brains are still able to discern the tone. Comparing the predicted state with the actual state is a ‘fuzzy compare’ – we can change the precision of this comparison. High precision is when we believe this prediction and, in this state, we ignore errors from our senses. If we are in a low precision state, we believe our senses more and pay attention to what is going on around us.

An example of fuzzy logic at work – most people are able to read this text.

At each stage of processing by the prediction engine, we are able to switch between these precision states. When get a prediction error (what our senses tell us doesn’t match what the prediction engine thinks we should be seeing) then there are a number of possibilities.

  • We can increase the probability of alternate prediction and adjust the next predictions.
  • It can adjust or create new schema in the long-term memory (learning)
  • Take action 

As teachers the priors are very important (these are prior experiences/knowledge of similar situations) or context. This will significantly affect the prediction and precision in the processing engine.

Adam played this file as an example

This shows the schema that we activate before learning can be important for activating prior knowledge.  Adam played another example

Adam showed an optical illusion in which the brain prediction engine expects shadow, the tones you see look different at different places on a grid when they are in fact the same shade of grey. Adam showed an image with a series of white and black dots moving in opposite directions, and our prediction engine interprets these as a cylinder rotating (not everyone sees this)

Precision in this model is important, it is about you paying attention to what is going on.  In a low precision state, you believe your senses more. At high prediction you are believing the prediction more. An example of the states in the brain is when you drive to work without remembering the drive – this is high precision mode and you have been believing the prediction and your model performs as expected.

When our prediction notices errors, we have to decrease the precision so we pay more attention and believe our senses more (this is when we are learning, updating our predictions and schema). When we are working in a fluent state, we aren’t learning because our predictions are expected. When we overlay emotions on this process, we experience anxiety and fear when reality doesn’t match the predictions. When we pay attention, if we keep experiencing errors, we feel frustration and anger. When we reduce our error rates and enter a learning state, we are concentrating and focused.

In an autistic brain we see a higher level of connectivity locally but fewer longer connections between the layers/areas of the brain. When we apply this to the prediction engine, because the signals are not communicated as well, they end up in a low precision state and they believe their senses a lot more. They get a flood of prediction errors. Autistic people think they are in a high precision state, but their precision is actually low, which leads to a high rate of prediction errors leading them to focus on details, more uncertainty and high levels of anxiety as a result. 

This sustained rate of prediction errors explains many autistic traits like a focus on details (spotting minute changes in their environment like a small poster changing) Living in this permanent state of predictions leads to living in permanent uncertainly and they use lots of energy being in this low precision state.

Adam suggests that as teachers we become managers of prediction errors for autistic learners. We need to reduce uncertainty and increase predictability – following the same lesson structure, same types of diagrams etc. We want to reduce extraneous load (perhaps by creating a low arousal environment) which will cut down on the extra noise. We can increase dopamine by sensory breaks, movement breaks will help refresh dopamine levels.

Lessons can be planned with lots of familiar content and then introduce small amounts of new content to help these being incorporated into the schema.

Adam’s talk to the group included lots of ideas and animated sliest that are difficult to translate into written text. Adam has agreed to share his video and presentation so you can watch it for yourself. 



Rizwan Ahmed
Rizwan Ahmed, founded by Rizwan Ahmed, is an educational platform dedicated to empowering students and professionals in the all fields of life. Discover comprehensive resources and expert guidance to excel in the dynamic education industry.


Please enter your comment!
Please enter your name here

Most Popular

Recent Comments