Mastery in the Age of ChatGPT

What is deliberate practice?

In a previous post, I explored what deliberate practice looks like and how it is the fastest route to mastery.

Developed by Anders Ericsson, deliberate practice is a combination of:

1. Focus: The learner must be completely focused on their objective.
2. Expert: The learner has access to a good teacher or mentor.
3. Difficult goals: The bar for achievement must be high.
4. Well-defined tasks: There is a good plan to achieve the goal.
5. Feedback: The learner must receive feedback as they progress.
6. Adjustments: The feedback must lead to an updated plan.
7. Mental models: The learner discovers their own way of doing things over time.

Most of us have experienced what difficult practice feels like:

• Sitting down at the piano each morning.
• Going to football training in the rain.
• Studying for a difficult exam.

We then continued deliberate practice in University where we would attack the optional maths problems and attend the lecturer's office hours for 1:1 support. Then, we encountered real-world, difficult problems in the workplace that required serious thinking and planning. All of these experiences led to significant growth and learning.

It is all being disrupted.

How AI disrupts the process of deliberate practice.

ChatGPT isn't affecting 'practical' trades as much as knowledge work. In knowledge work (think software development, copywriting, journalism) it is significantly changing the learning process. When it comes to those moments of extreme growth we used to experience, we need to know how ChatGPT influences the learning process so we can continue to experience that growth.

How we define focus

Deliberate practice requires extreme focus. The task at hand should be difficult enough that it is well beyond what you are currently capable of. Let's see what happens when you focus hard on something.

And let's zoom in on the key components of that focus.

Firstly, you look at the problem or task and attempt to break it down into manageable steps. If it's a difficult maths problem, you create a plan of the different sub-problems you need to prove. If it's a new technique for holding the tennis racket, you attempt to visualise how it should feel in your hand, your ideal body position and the perfect motion of the racket as it hits the ball. Whatever the objective is, you break it into its sub-components and attempt to get each sub-component correct, hoping it will culminate in the desired outcome.

Secondly, as you grapple with the problem or technique, you are constantly reflecting. For each step forward, you ask yourself "Did I do that well enough?". Maybe you didn't quite hit the ball correctly because your hand was in the wrong position. You figure these small adjustments by constant reflection.

Thirdly, you constantly reevaluate yourself against your objective. This is a larger piece of feedback to check where your progress is relative to your goal. Maybe you have found another interesting maths problem or a weakness in your swing, but that is not what we are concentrating on now.

These three things culminate in extreme focus. You constantly perform a task, reflect on it and then evaluate as it relates to your larger objective.

Disrupting Focus with ChatGPT

Let's look at what happens to focus when we use AI.

The first thing we should do is to break the problem into steps. But if I'm using ChatGPT to solve for X, then I will never engage in breaking down the problem. I can just ask ChatGPT "Solve for X".

The AI rarely returns the correct answer the first time, unless I am looking up an easy fact. So I prompt it again saying "That's not the right answer, try again". It returns a different response and I rinse and repeat.

None of the first 2 criteria for focus are met. I'm not (1) breaking the problem into steps and I'm not (2) reflecting and adjusting my tasks. You might argue that I'm re-evaluating the AI's answer in relation to the overall objective, but then I break that thought process by just re-prompting for a different answer.

Using AI skips the whole process necessary for deep focus.

The Practice Cycle

Next, let's look at the practice cycle in deliberate practice.

This is the intensive part of the learning process:

• We create a list of well-defined tasks.
• We choose a task to complete.
• We execute it and get feedback of some sort.
• We adjust our practice and rinse and repeat.

This is what happens when we're studying for a test, practising drills on the piano or practising a golf swing. But what happens when AI is used for knowledge work such as studying and learning?

By default, ChatGPT skips the whole practice cycle

Here's another way of looking at the flow of learning. When we focus on a task, our brain engages, takes in some new sensory information, encodes & consolidates that information and eventually we achieve our objective or complete the task. This encoding of new information is essential to learning. The more focused the person is on the task, and the more difficult the task, the more the brain works and takes in new information.

Contrast that to when we use ChatGPT to solve a problem: We type in our problem into ChatGPT, it spits out an answer, and we severely limit the encoding phase. This means that little to nothing is learned. Imagine the consequences for teenagers whose brains are still developing!

But I just want to get things done!

To be clear, I'm not suggesting we shouldn't use ChatGPT. But we need to ask ourselves a question every time we use it and skip the learning process: "Do I need to learn this?" If I need to learn about this task, then I should not use ChatGPT. If you're a student who has a paper to submit, you shouldn't use ChatGPT. If you're a software engineer who needs some skeleton code quick, maybe you should use it. But this leads us into another danger.

Using AI doesn't just stop your learning, it atrophies your current skills

Skills atrophy when we don't use them. Everyone knows the pain of learning a foreign language only to come back years later and forgot everything. The Ebbinghaus curve illustrates this.

Forgetting is not a weakness of the brain. It is a strength. People with hyperthymesia (the ability to remember a large number of life experiences) have tremendous difficulty in operating a normal life. So we need to recognise that while we want to remember the lessons in this chapter, our brain just wants to make room for more things.

Therefore, when we use AI to perform tasks we find tedious or easy, we should be careful. We may be tempted to throw the AI an "easy" task which we already know. The more you "automate" that skill with AI, the more it atrophies.

This is further emphasised by a recent early-stage study from MIT which demonstrates how brainwaves for people who use only ChatGPT hinders their brain processing, retention and engagement with written tasks.

This isn't surprising when we look at the "Learning Process with AI" diagram again. When we ask ChatGPT to solve our problem for us, we skip over the part where we engage our brains to complete the task. Our brain does not engage in any level of difficult thinking, and so the information and skill is neither developed nor used.

In the context of deliberate practice, this is clearly a destructive pattern if the person wants to learn. Discomfortis crucial for the brain to learn.

Why is discomfort and difficulty important?

Prior to studying this area, I always thought the optimal state for learning was the flow state. Flow state is that feeling of being in the zone and at your peak performance. You can usually achieve it by focusing on a task that is slightly beyond your current skill level.

Your body and brain wants to relax

We know that our bodies and brains are excellent at adapting to challenges. We build muscle by lifting heavy weights. We build expertise by engaging in challenging experiences.

But our bodies aren't good at growing by themselves. In fact, they constantly want to relax. This is called homeostasis. Our body puts on muscle or sheds weight because it sees that as the best way of relaxing in the future. When we go to the gym, our brain tells itself: If this weight is heavy today, I need to put on muscle so I can lift the same weight more easily next time. This anticipation to change is called allostasis.

We can use homeostasis and allostasis to our advantage. Think about it like this: The more pressure we put our bodies and brains under, the faster they will grow. And this is where deliberate practice comes in. Deliberate practice should not be slightly beyond your skill level (like doing tasks in the flow state). It requires discomfort to force the brain and body to grow fast and therefore requires tasks that are uncomfortably outside your skill level (but not so far that you fail to get any meaningful feedback).

Takeaways

If it feels easy, I didn't learn anything

This fact has now become my go-to litmus test on whether or not I'm studying properly. Whether I'm using ChatGPT, reading, watching videos or any sort of learning task, I ask myself how uncomfortable this feels for me. Learning isn't supposed to be easy - if it were easy then the brain wouldn't adapt and grow.

If I think I know a subject after watching a video. If I read a blog post and now feel like an expert. If I paste a difficult concept into ChatGPT, ask it to simplify it for me and suddenly think I understand it. These self-checks tell me I don't yet know the material.

If this is a struggle, I'm learning something

But if a task is difficult. If I'm struggling to understand something despite having mapped it out. If I need to revisit a study again and again (like the MIT study above) to clarify information. If I'm constantly finding new information that contradicts what I already believe. I know that these challenging moments are the moment where my brain is burning energy trying to make sense of things and build up a schema.

Importance of Expertise and Mastery

In a nutshell, difficult tasks are what everyone should constantly do. In the age of ChatGPT, the temptation is to get something done as quickly as possible and disregard whether we know it or not. But as ChatGPT and other AI tools become increasingly better at doing menial tasks, the bar for expertise gets higher, meaning we must not sacrifice our learning in favour of speed. As intelligence gets cheaper, our expertise must increase, not decrease. ChatGPT and other tools can help us work faster, but it's only worth it if we work to increase our abilities and deepend our mastery.