Not My Circus - Cognitive Load and Learning

Since a massive influx of shysters started posting on LinkedIn about AI and how it’ll take your job or won’t take your job or whatever, we seem to have lost sight of the fact that creating software systems and getting them live is knowledge work. To do it properly, we need to learn how to do it in the first place.

In my first IT job, I was part of the support team, carrying out an SRE-type role and working with customers to deploy and test our changes to their systems. Later, when I hired graduates, their first assignment was support. Research shows that this is not the best first assignment for newbies. Problem-solving and learning don’t mix.

"Considerable evidence indicates that domain-specific knowledge in the form of schemas is the primary factor distinguishing experts from novices in problem-solving skills. Evidence that conventional problem-solving activity is not effective in schema acquisition is also accumulating." [Cognitive Load During Problem Solving: Effects on Learning]

Which is great, but what’s a schema? We use schemas in all aspects of our lives. When we start doing something we've done before, we upload that schema into our brains and then know how to do it. In The Matrix, for example, when Neo says, "I know kung fu," it's as if he's uploaded the schema for kung fu to his brain. We have schemas for all aspects of our cognitive lives. An example in [Efficiency in Learning Evidence-Based Guidelines to Manage Cognitive Load] is that we all have a schema for trees, so when we see a tree, we take in all that information about the leaves and the wood and the colour of the leaves. There is a lot of information, but we use the schema from all the other trees we have seen to recognise the tree. We also have schemas for letters, words, combinations of words, and their meanings. For example, we have schemas for solving specific math problems. If we have many of these schemas, then we are proficient at maths. Our competence in any area comes from the schemas we store in our long-term memory. Schemas are important. It's like having a mental runbook that we can quickly upload, preparing our brains with the competence to do whatever work we need.

For example, I have always found that I am much faster at moving around in AWS after I have done a lot of work in that area. This is probably because I have developed or reinforced my AWS schema. It's also worth keeping your certifications up to date so that you can quickly get up and running again without using too much cognitive capacity to get to this point. In the future, we can use LLMs to help us develop and reload schemas more quickly for whatever we are working on, allowing us to focus on the more relevant task at hand.

Returning to our newbie on support, she needs to solve problems by dealing with production incidents. Domain-specific knowledge in the form of schemas is crucial for problem-solving expertise. The issue - getting there. The cognitive processes required for problem-solving and schema acquisition don’t overlap.

Experts solve problems by recognising situations they have been in before and the schemas that support them. New people use more traditional problem-solving approaches. One of the main ways we approach a new problem is to break the problem down, identify the gaps between the current state and what we want, and then determine how to get there. This is called Means-Ends analysis. This is an entirely different way of thinking from using schemas. It is also completely different from acquiring schemas. It’s also challenging and has a high cognitive load. This leaves little brain space for developing new schemas.

This is one of the reasons it is so stressful to look at a problem in an unfamiliar system - this stuff is hard! It’s also stressful to keep identifying a negative thing - a gap from where you want to be - and then work backwards to figure out how to bridge that gap. An expert doesn’t do this - they work forwards based on what the givens are using schemas.

This is maybe a way in which AI can help us do our jobs properly - effectively spoon-feeding us schemas in the best way to update our schemas so we can be all ‘I know kung fu’ and kick Morpheus’ metaphorical ass by rewriting the Matrix.

References

Sweller, J. (1988). Cognitive Load During Problem Solving: Effects on Learning. Cognitive Science, 12(2), 257–285.