I think I've been outsourcing too much of my brain.

I've been using ChatGPT since it launched in November 2022.

Emails, research, difficult conversations, decisions, random questions at 2 AM, figuring out what to say to someone, and occasionally figuring out what I was feeling in the first place.

More than three years of having a remarkably patient, reasonably intelligent assistant who never gets tired of my questions.

A few months ago, I began noticing something uncomfortable. I was becoming increasingly dependent on AI for things I used to handle perfectly well myself.

It wasn't that I couldn't write an email anymore. I could. But why bother when ChatGPT could write a better one in ten seconds?

It wasn't that I couldn't think through a problem. But why spend twenty minutes wrestling with it when I could have five reasonably good perspectives before finishing my coffee?

And then there was the slightly embarrassing business of asking AI to help me understand my own mood.

I had somehow become a human API wrapper. Something happens in my life, I send it to an LLM, receive a response, and proceed accordingly.

An impressive technological achievement. A somewhat disappointing personal development.

I started wondering whether I was losing my edge.

The strange thing was that there was no obvious evidence of decline. My written communication was probably better than ever. I could research unfamiliar subjects faster, articulate complicated ideas more clearly, and get through work that previously took hours.

Everything I produced was improving.

I just wasn't entirely sure that I was.

The problem with removing all the friction

Before AI, finding an answer involved some effort.

You Googled something, opened several tabs, discovered that three people had three completely different opinions, and spent some time figuring out who was talking nonsense.

Sometimes you read a book. Sometimes you watched a YouTube video. Sometimes you asked someone who knew more than you.

It was inefficient. I don't particularly miss it.

But somewhere in that inefficiency, you were doing something important. You were comparing, questioning, remembering, connecting ideas, and occasionally arriving at a conclusion that was actually yours.

AI made much of that optional.

Now you can go directly from not knowing something to having a beautifully articulated answer about it.

We've become very good at getting answers without necessarily becoming better at answering questions.

And those aren't the same thing.

Removing unnecessary effort is one of the greatest things technology does. But not all effort is unnecessary.

Some of it is the actual process of learning.

The struggle to explain an idea forces you to find the holes in your understanding. The terrible first draft forces you to discover what you really want to say. Sitting with an uncomfortable decision forces you to confront what matters to you.

Remove all of that, and you might get a better result.

But what exactly have you become better at?

Invisible Knowledge Decay

I've started calling this Invisible Knowledge Decay.

It's not a scientific diagnosis. It's my name for something I began noticing in myself.

With AI, the warning signs are harder to recognise because the quality of your output can remain excellent.

From the outside, you appear more capable.

Meanwhile, something peculiar can happen internally.

You recognise a good argument but struggle to construct one. You understand an explanation while reading it but can't reproduce it later. You agree with a conclusion without being entirely sure how you would have arrived there yourself.

I noticed that distinction between recognising something and knowing it.

You know the feeling when you're reading a response and thinking, "Yes, exactly. That's what I was trying to say."

But were you actually trying to say it? Had you worked through the argument? Or did the answer simply sound so convincing that your brain decided to claim ownership?

I've caught myself doing that.

And the more I thought about it, the more I realised this wasn't limited to intellectual work.

Consider communication.

Suppose you're upset with someone. Previously, you might have written an unnecessarily dramatic message, deleted it, rewritten it, realised you were being unreasonable, and eventually figured out what actually bothered you.

Now you can ask AI to draft something mature, empathetic, and appropriately assertive.

The recipient gets a better message.

But perhaps you've skipped the part where you understand your own feelings.

And then there are professional decisions.

AI can give me a compelling argument for almost anything. It can also give me a compelling counterargument. Ask it to sound more confident, and suddenly the same mediocre idea starts looking like something worth presenting to a board.

But if somebody challenges that argument in a meeting, there's no prompt box between the question and my response.

I still have to understand what I'm talking about.

How I realised something needed to change

The biggest warning sign for me wasn't forgetting information.

It was the growing reluctance to attempt things without AI.

An empty document felt unnecessarily difficult. A question that required some thinking felt like an invitation to open ChatGPT. Even minor decisions started acquiring an extra step.

Not because I genuinely needed help.

Because asking had become easier than thinking.

There's another problem with this habit: AI is exceptionally good at making almost any thought sound reasonable.

Give it a weak argument, and it can make it sound impressive. Give it a half-formed opinion, and it can supply the missing structure. Ask it whether something makes sense, and depending on how you frame the question, you might get a very reassuring response.

If you aren't careful, you stop evaluating ideas and start evaluating how nicely they're presented.

And nice presentation is a terrible substitute for good judgment.

If I closed ChatGPT right now, how much of what I've just produced could I explain, defend, or recreate?

Sometimes the answer wasn't particularly flattering.

So I started changing a few things.

What actually helped me

I didn't quit AI.

That would be a rather theatrical response to a problem caused by using a useful technology too conveniently.

Instead, I began changing when I involved it.

The first change was simple: I started giving myself the first attempt.

If I wanted to write something, I wrote my version first. Usually an inferior version. Occasionally a spectacularly bad one.

But it forced me to figure out what I wanted to say before asking a machine to make it sound better.

I began doing something similar with decisions.

Before asking AI for an opinion, I tried to articulate my own. What do I think? Why? What am I assuming? What would change my mind?

Only then would I bring AI into the conversation.

The second change was more interesting.

I stopped asking AI to agree with me or improve everything I wrote.

Instead, I started asking it to challenge me.

What's weak about this argument?

What am I overlooking?

Ask me questions that would force me to explain this better.

Don't rewrite my answer. Tell me where my reasoning falls apart.

Getting a polished answer feels satisfying.

Having your argument dismantled, then figuring out how to rebuild it, is considerably less pleasant.

But I found the latter more useful.

The third change was making myself explain things without assistance.

Sometimes that meant writing for ten minutes without opening another tab. Sometimes it meant trying to explain something I'd recently learned, using only what I remembered.

No notes. No AI-generated summary to rescue me.

You quickly discover the difference between information you've consumed and knowledge you can actually use.

I also started paying more attention to what happened after interacting with AI.

Could I still explain the idea tomorrow? Had my opinion changed because I'd genuinely understood a better argument? Could I apply what I'd learned to a different situation?

Or had I simply enjoyed reading something intelligent?

I don't do all of this perfectly. I still catch myself reaching for ChatGPT when I could quite easily think something through.

But at least I notice now.

Not every problem needs to become a personal growth exercise

I should clarify something.

I don't think you need to write every email yourself, struggle through every research question, or refuse assistance whenever something gets difficult.

There is no intellectual virtue in spending forty minutes formatting a document that AI could fix in ten seconds.

Some tasks deserve to be automated.

Some decisions deserve outside perspectives.

Some questions genuinely benefit from an assistant that can explore more possibilities than you could reasonably consider on your own.

The distinction I'm trying to make is between outsourcing a task and outsourcing a capability you still want to possess.

I don't need to become better at formatting spreadsheets.

I do want to remain capable of constructing an argument, forming an opinion, understanding my emotions, making difficult decisions, and explaining what I believe.

Especially when someone disagrees with me.

These aren't inconveniences to eliminate.

They're rather important parts of being a person.

The question I now ask myself

I started this whole exercise because I was worried that AI was making me intellectually lazy.

I'm certainly not claiming that using AI inevitably makes people less intelligent. It can be a wonderful learning tool when used well.

But I do think there's a meaningful risk in allowing a system to perform every difficult mental step on our behalf, particularly when the results are so good that we stop noticing what we're no longer practising.

The solution, at least for me, wasn't to use less AI indiscriminately.

It was to become more deliberate about what I wanted it to do.

These days, when I finish a conversation with ChatGPT, I try to ask myself one question:

Did I just get a better answer, or did I become a little better at thinking?

Both are useful.

But I'd rather not spend the next four years collecting the first while quietly losing the second.