Sometimes people claim that you won’t need to do your own reading, writing, or coding once AI gets good enough.

They are deeply, horrifically, wrong. This is only true if you’re planning to surrender all involvement in your own future.

But let’s take a step back and talk about what the point of these activities even is.

My argument is that these, like many other human activities, are part of how we think.

Even if the machine creates the final product, you trying to write out your own ideas clarifies them, develops them, much like how reading for yourself and struggling with it is for your own edification even if someone could have told you the point of the paper or essay.

Coding is very similar: a kind of writing, a way of clarifying thoughts into a concrete plan.

Many programmers can tell you that there’s value in messing around and trying out ideas, in doing exploratory coding as you figure out how a new domain works.

But even the act of implementing something on your own that’s well known still teaches you to internalize the concepts in new ways.

It’s like doing your math homework as a child: the point of practicing solving equations isn’t the end product. It is what you gain from the practice, from the learning.

Okay, so there are clearly two themes to our handful of examples: exploring new domains and internalizing already understood knowledge and techniques.

This gets into my rough taxonomy of knowledge work.

Category (1) work, rote work, is where there is nothing to be learned by doing it and there is no novelty required in the work itself. This is the work the machine excels at because there is no benefit to this labor beyond it being done and done competently.

Category (2) work is, well, what we generally call studying or practice. You’re not producing anything original, but you are exploring things you don’t already know how to do.

A lot of (2) work becomes (1) work once you master the concepts, wringing it dry of ideas to learn.

The third category is exploratory work. Maybe it’s true research where no one knows the answer; maybe it’s just something so new to you that you need to explore, try, play, brainstorm.

The relationship of the machine to category (2) and (3) work is more complicated than (1) work.

If you automate (2) work you learn and internalize very little, only the superficial regurgitatory knowledge of someone who has been told facts rather than worked through the insights themselves.

On the other hand the machine acts as a very competent “spotter” to help you get unstuck in your (2) work. It can help check the quality of your understanding.

As for (3) work, the machine is good for pursuing rabbit holes, looking for ideas and literature you haven’t seen, and building proofs of concept.

But how do you know what to suggest to it? How do you know what paths are worth diving down?

This is your element of play, exploration, divergent thinking.

If you pursue this work you can do incredible things in partnership with the machine that still have all the complicated, contingent creativity that you specifically have available to you.

So, yes, you’ll always have a need to study, to write, to code, to think.