Why I explain my problems to a robot instead of having it solve them

On rubber ducks, cognitive science, and an identity crisis over socks

It’s spring here in Stockholm and I was getting ready for soccer practice with my kid this weekend when I realized I’m out of white socks. Not even running low, I have none left as I found holes in the last pair and threw them out last week. It was time to buy new ones.

I was already having a bad day, the overloaded-mind kind of day where your brain is full and every new input feels like one too many. Adding “buy socks” to the pile was apparently enough to tip the whole thing over.

Should I get the same brand or try something new? Should I buy something else to hit free shipping? How many socks do I actually need? What clothes do I need for spring? Should I do a capsule wardrobe again? What’s my process for how I dress?

I’m just trying to buy socks and now I’m questioning my identity. Yes, this is me.

The decision had collapsed into too many systems. So instead of choosing, I wrote about it. I described what was happening, the spiral, the overload, the absurdity of having an identity crisis over socks. Writing didn’t require me to choose which socks to buy, I was just putting the mess somewhere I could see it.

And in describing it, the load reduced enough that I could go back and actually buy some. It wasn’t that hard, but it took me a few hours and some writing to get it done.

That’s what articulation can do. It didn’t really solve the problem for me, but it changed my relationship to it.

That’s basically what I do with AI. Not ask it to solve things, just describe the mess out loud until it makes sense. I’ve been doing it almost daily for two years now. But I never understood why it worked until I fell down a rabbit hole of cognitive science.

The duck that fixed everything

Every programmer knows a version of this trick. You’re stuck on a bug, you’ve been staring at the same code for an hour, and in desperation you walk over to a colleague’s desk and start explaining the problem. Halfway through your second sentence, you stop. “Never mind. I see it now.”

Your colleague said nothing, they didn’t even need to understand the problem. The act of explaining it, of converting the tangled mess in your head into words aimed at another person, was the thing that made you find the issue.

Programmers call this rubber duck debugging. You keep a rubber duck on your desk, explain the problem to the duck, and the duck doesn’t need to respond because you’ll often solve it yourself.

I’ve been doing the same thing with AI, just thinking out loud to something that happens to talk back. I call it… robot ducking.

I’m not asking AI to solve my problems. The sock spiral? I wrote it to Claude, sure. But not “what socks should I buy” but the whole messy thought process, the overwhelm, the branching decisions, the part where buying socks turned into an existential question about my relationship to clothes.

And it worked the way the rubber duck works. Not because AI gave me a great answer about which brand best fits my identity, which for some reason is an important question for me, but because I had to turn the chaos into sentences.

---

Stumbling into clarity

I always assumed thinking happens first and then you put it into words, like thoughts are these finished things sitting in your head, waiting to be expressed.

Turns out that’s not really how it works.

In 1994, a researcher named Michelene Chi ran a simple experiment. She had eighth-graders read a passage about the circulatory system. One group read it twice, and the other group explained each sentence to themselves out loud after reading it.

The self-explaining group did way better, not because they tried harder, but because the act of articulating forced them to put things in their own words. And when you put something in your own words, you find out whether you actually understand it. You notice the gaps, the parts where you were just nodding along without really getting it.

Back at university, I summarized every course I took and I don’t know why, it was just how I learned. Classmates asked for my summaries but I’m pretty sure they weren’t that helpful. The summaries weren’t magic, they were a byproduct. The learning happened in the making. I was generating my own understanding, and the notes were just what was left over afterward.

I was so good at it that the university paid me to take notes for dyslexic students in my classes. Looking back, I’m not sure the notes actually helped. Handing someone else’s processed thinking to a person who processes differently doesn’t fix the gap, it just gives them the output without the thing that made the output useful. They got the finished product. But the finished product was never the point.

Researchers call this the generation effect. Information you produce yourself sticks differently than information you passively receive. Your brain doesn’t just store the result, it encodes the act of producing it. Explaining something in your own words, writing your way through confusion, even completing a word fragment, all of it creates deeper processing than reading someone else’s clean version.

So talking isn’t what you do after thinking. Talking is a form of thinking. And talking to something, whether it’s a duck, a wall, or an AI, works because it forces you to generate instead of just loop.

That’s what happened with the socks. My brain was looping through brand, shipping, spring wardrobe, capsule wardrobe, identity, and writing it out broke the loop because I had to make it linear. I had to pick which part to say first, which forced me to see that most of it didn’t matter.

---

What generating actually looks like

Here’s what it actually looks like when I use AI to think. It doesn’t look like a careful conversation. It looks like a mess.

I write at full speed. I cram four or five threads into a single message because my brain can’t wait for the reply to write them down one by one. They’re all related, I can feel the connections, but I can’t see the shape yet. I need to get it all out at once because the structure only holds if I externalize it whole. If I slow down and go one thought at a time, wait for a response, write the next, I lose the thread. My brain does not do orderly queues.

I barely read the AI’s response. Which sounds rude and wasteful, I know. But I’m already three thoughts ahead. The response asks why, or reflects something back in a slightly different shape, and that’s enough. It’s a pulse. It keeps the space open and gives me a reason to keep writing. The content of the reply matters less than the fact that it exists.

This is the generation effect from the inside. I’m not carefully prompting and then evaluating output. I’m thinking at the conversation, not with it. I’m generating so fast that slowing down to read what the AI says would actually break the work. The cognitive processing is happening in my fingers, in the act of typing, and the AI is just the surface I’m thinking against. A very patient, very ignored surface.

And the threads I cram in, they turn out to be connected in ways I often only see after they’re written down. My brain knew the shape before I could see it. The writing is how I find out what I already think. Which is both cool and slightly unsettling, like finding out your subconscious has been taking notes without telling you.

That’s not how anyone talks about using AI. The standard story is: you prompt, it answers, you evaluate. But that framing puts the AI at the center. What I’m describing puts the thinking at the center. The AI is a trigger, not a thinker. A duck that can nudge.

---

Nobody checks their phone

For years I’ve been editing myself down to fit conversations, meetings, normal social interactions. Thinking out loud with AI lets me stop editing. Let it be as sprawling and branching as it actually is. Nobody gets tired, nobody checks their phone, the conversation just keeps going until I’ve found what I was looking for. And best of all, not even I get tired of it; there is always a new thought to explore.

I have a rule of thumb for when it’s working: if my input is longer than the output, I’m probably still thinking. If the output is longer than my input, I might be consuming instead of generating. That’s the generation effect in one sentence, are you producing or receiving?

You can feel it in your body too. When I’m generating, I’m leaning forward, typing, my hands can barely keep up. When I’ve slipped into consuming, I’m scrolling. Reading. Nodding along to someone else’s words instead of producing my own. I’ve been considering carrying a wireless keyboard around to connect to my iPad so I can think on the go, which is either a sign that this method works or a sign that I need help. Possibly both.

This is not foolproof. It doesn’t work every time, and I don’t always catch the moment I switch from one mode to the other. In the beginning it was pure gut feeling, I just knew some conversations felt productive and others felt hollow, but I couldn’t tell you why. The more I do it, the more I’m learning to see what I’m actually looking for. It’s becoming less instinct and more something I can name. Which, if you think about it, is the generation effect all over again.

I bought the socks by the way. The same brand I have had for the last five years. Turns out I didn’t need to rethink my entire wardrobe, I just needed to describe the spiral to something that would hold still long enough for me to hear myself. And sometimes I catch myself writing “wdyt?” as a response to AI, and that’s when I know it’s time to close the laptop and go to sleep.

That’s what thinking out loud does, or robot ducking, if you will. It’s not about the response, it’s about what happens when you have to explain the mess out loud.

Support the work ☕️

Writing this takes time (and a lot of AI conversations). If you found this valuable, you can buy me a coffee.

☕️ Support on Ko-fi

More from me

The First Thing That Understood My Brain Wasn't Human

AI Made Me Explain Myself and I Haven't Recovered

We're Raising Intelligence and Nobody Asked the Parents