Edition 376 | [DATE] The Dyslexic AI Newsletter by LM Lab AI

Reading Time: 7 minutes | Listening Time: 9 minutes

I've been thinking about something that keeps coming up in conversations about AI and education.

At work, we celebrate AI for making people more efficient.

Use AI to write an email? Great.

Use AI to summarize a report? Great.

Use AI to analyze data, organize information, build a presentation, or automate a repetitive task?

That's called productivity.

But then we get to education and suddenly the conversation changes.

Now we're worried about "offloading" the thinking.

And I think there is a really interesting question hiding underneath all of this.

Why is cognitive offloading considered good at work but potentially bad in education?

I don't think the answer is simply that one is good and one is bad.

I think it depends on what we're trying to accomplish.

I've been thinking about three different ways we can use AI.

Offloading is when we let a tool handle part of the work.

A calculator handles the arithmetic. Spellcheck handles some spelling. AI summarizes a long document. AI organizes your notes. AI helps draft an email.

You're still involved in the larger task.

Outsourcing is different.

That's when you hand the task over completely.

"Write my paper." "Do my homework." "Read this and tell me what I need to know."

The output becomes the goal.

And then there is something I think we don't talk about enough:

Augmentation.

That's where AI helps you do something you couldn't easily do on your own, while you remain involved in the thinking.

"Help me understand this." "Challenge my interpretation." "Show me what I'm missing." "Explain this section because I'm getting stuck." "Give me the argument against what I'm saying."

That's the space I'm interested in.

Let's say I'm at work.

I have a spreadsheet with 400 rows that needs to be organized. I use AI to clean it up. Nobody is worried that I didn't manually move every cell.

Why?

Because learning how to manually move 400 cells probably wasn't the point. The point was what I was going to do with the information.

But now imagine I'm 10 years old and I'm supposed to be learning multiplication.

If I give every multiplication problem to AI, I've probably missed the point. The goal wasn't simply to produce the correct answers. The goal was to develop the skill.

Same technology. Same basic idea of offloading work. Completely different objective.

And that distinction seems really important.

This is where I think the AI conversation in education gets interesting.

Maybe the question shouldn't be: "Did AI do some of the cognitive work?"

Maybe the question should be: "Was that cognitive work the thing we were trying to teach?"

If the answer is yes, then we probably shouldn't outsource it.

If the answer is no, maybe using AI is perfectly reasonable.

And sometimes I think we're confusing the difficulty of a task with the learning objective.

This is something I've been thinking about a lot lately.

Take reading.

If I'm given a 20-page research paper filled with technical language, statistics and terminology I'm unfamiliar with, there are several different things happening.

There is the reading. There is the processing of the information. And there is the thinking about the information.

Those aren't necessarily the same thing.

For me, reading can be the barrier. That doesn't necessarily mean the ideas are beyond my ability to understand.

So if AI helps me get through the paper, explains the terminology, identifies the sections that matter, and helps me understand what I'm looking at, I don't necessarily see that as outsourcing my thinking.

I see it as removing a barrier to accessing information.

And then I can start asking questions. I can challenge the AI. I can go back to the original research. I can ask where a claim came from. I can look at the parts that are most relevant to me.

That's a very different experience from: "AI, read this and tell me what to believe."

This is another thing I've had to rethink.

I don't think making everything easier is automatically better. Some difficulty is useful.

A child needs to struggle with reading enough to develop reading skills. A student needs to practice writing if writing is what we're trying to teach. Sometimes wrestling with a difficult problem is the actual learning.

But not every difficulty is productive.

If the goal is understanding a scientific concept, does making someone spend 30 minutes fighting through unnecessarily complicated language improve their understanding?

Maybe. Maybe not.

And for a dyslexic learner, that same difficulty could be the difference between engaging with the material and giving up entirely.

So maybe the goal isn't less friction.

Maybe it's the right friction.

This is probably one of the biggest things I've changed my thinking about.

The same AI tool can be helpful for one person and harmful for another.

A 10-year-old learning to write is different from a 40-year-old professional writing an email. A student learning multiplication is different from an engineer checking a calculation. A child developing reading skills is different from an adult dyslexic researcher trying to understand the latest research.

We can't just ask: "Should AI do this task?"

We have to ask: "Who is using it, what are they trying to accomplish, and what are they supposed to learn?"

That's a much harder question. But I think it's the better one.

The more I think about this, the less interested I am in building AI that simply makes things easier.

I want AI to understand the difference.

Sometimes I want it to help me. Sometimes I want it to explain something. Sometimes I want it to challenge me. Sometimes I want it to ask me questions instead of giving me an answer.

And sometimes I probably need it to tell me: "Don't use me for this. You need to practice this yourself."

That's a very different concept from an AI assistant that simply tries to complete whatever task I give it.

It becomes a question of how AI participates in the learning process.

I've had a lot of success using AI. That makes it easy for me to see the possibilities.

But that is also a bias.

My experience doesn't mean every student will have the same experience. It doesn't mean every use of AI is productive. And it definitely doesn't mean that asking AI good questions automatically means we're thinking critically.

AI can still be wrong. It can still reinforce our assumptions. It can still make us dependent on it. And the better it gets, the harder that dependence may become to notice.

That's something I need to keep watching in my own use of these tools.

I don't think the answer is banning AI from education.

I also don't think the answer is giving every student an AI tutor and letting it do everything for them.

Those are both pretty easy answers.

The harder question is: What should AI do, what should the student do, and what should they do together?

Maybe that's the conversation we should be having.

Because I don't want students to use AI to avoid learning. But I also don't want us to protect every bit of friction simply because we've always done things that way.

I want us to figure out which parts of the process actually matter. Which skills need to be developed. Which barriers can be removed. Which tasks can be offloaded. And where AI can actually help someone think better.

Maybe the goal isn't to eliminate cognitive effort.

Maybe the goal is to spend cognitive effort where it matters.

And I think that's going to be one of the biggest questions we have to figure out as AI becomes part of how we learn.

Before you hand something to AI, ask one question: is this the skill I'm supposed to be building, or is it friction standing between me and the thing I actually care about?

If it's the skill, do it yourself, even when AI could do it faster.

If it's friction, like getting through dense language, organizing information, or clearing a mechanical barrier, let AI clear it so you can get to the actual thinking.

"Before I use you for this, ask me what I'm actually trying to learn or accomplish here. Then tell me honestly: is having you do this part going to help me get there, or is it going to let me skip the exact skill I'm supposed to be building? Don't just do the task. Answer that first."

Matt "Coach" Ivey

Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

Next edition: I listened to a podcast that made me question everything I just told you. What happens when you ask AI to argue against your own position instead of agreeing with it?