The Dunning-Kruger Effect and Climbing Mount Stupid

I'm Pablo Stanley, a designer, and founder of Musho, a tool that helps you create dope designs with AI. I've been a designer for far too long, surviving on a diet of pixel-perfection, neverending revisions, and doodles.

Lots of doodles...

Honestly, it can be a grind, and it's not a life suited for everyone (IYKYK, right?).

This post was supposed to be about AI and design, but I decided to switch things up at the last minute because why not? I don't want to bore you with the typical AI doomsday message like, "Oh, AI is gonna steal your job," or the flip side, "Nope! It won't take your job; it will empower you. AI is a tool—a superpower! Oh, but remember, you need to adapt!"

I'm a techno-optimist, so I won't dive too deep into the typical stock responses:

• Yes, AI will disrupt our jobs.
• Yes, it's also an incredible tool that will enhance our skills.
• Sure, it might one day overthrow us and decide to turn us into fields of fleshy batteries, but let's not get hung up on that.

Don't worry, though! This post will still discuss AI and design; just be patient and keep reading, ok? I promise it will all make sense in the end.

Dunning-Kruger Effect & why you might kinda suck (for now!)

Today, I want to talk about the Dunning-Kruger Effect and share three instances where I've been a victim. For those asking, "What the heck is the Dunning-Kruger Effect?" Let's dive straight into it so you can get a sense of what I'm talking about.

The Dunning-Kruger Effect is this cognitive bias where people who kinda suck at something they think they're pretty good at it. In other words, it's about clueless people boldly attempting things way out of their league.

One of the super cool things about the Dunning-Kruger is that it even has a graph! On the x-axis, you have competence - how good you really are. And on the y-axis, you have confidence - how good you think you are.

As you can see at the beginning, when you know nothing and your competence is at its lowest, the graph shoots up high in confidence. In other words, the less you know, the more you think you know.

Or, as I like to say, If you're dumb enough, you might even reach the peak of Mount Stupid (a place I've found myself before). The thing about reaching the top of Mount Stupid is that reality hits super quickly. And, of course, once you realize what you don't know, your confidence drops instantly. You find yourself in the Valley of Despair, suddenly realizing, "Oh man, I know nothing."

But here's the thing – confidence eventually starts rising again for those who keep pushing, improving, learning, and becoming more competent.

Well, I've been that not-so-smart person many times. And I'm pretty sure you're already thinking of someone like that, too. Perhaps you've had one of those managers who believe they know everything about design, where for them, everything seems so simple; it's just a bunch of boxes and text—easy, right?

We might laugh at them or get mad at them—"Oh, they just don't understand anything about design, but they think they know it all!" we say.

But I can guarantee that all of us here have experienced the Dunning-Kruger effect at least once. We've all thought we understood philosophy because we read a Wikipedia description about existentialism.

There are three stories where I personally experienced this fascinating sensation of the Dunning-Kruger Effect. The first one is about the time we naively thought we could fix the healthcare problem in America—easy, right? The second is about me thinking being a CEO and raising funds would be a breeze. I was like, "Sure, millions of dollars will just fly our way because I totally deserve it, right?" The third, a current endeavor, is about how recently our team and I believed that building an AI design app would be a piece of cake. Spoiler alert: we were so wrong.

Climbing Mount Stupid Pt I: Solving American healthcare

I first reached Mount Stupid's peak when a group of colleagues and I thought we would solve America's complex healthcare system. We were these young, idealistic devs and designers starting Carbon Health. We saw the healthcare problem in the US and thought, "Hey, we can fix this with some coding and good design... yeah, that's what doctors need!"

Our confidence started rising, and we said, "All we need to do is give them a fancier medical report, an app for data collection, a calendar to schedule stuff, and... that's it, problem solved, right?"

Boy, we were wrong. We didn't consider that healthcare is tangled in a web of regulations, that doctors don't need another app—they already have a ton, and they hate them. They need more time to care for their patients.

We also didn't think about how patients don't just need a fancy calendar; they need to understand their prescriptions, whether their insurance will play nice with their bills, decipher their lab results, and know if they'll be dancing at their niece's wedding next month or still sitting in waiting rooms.

And it's not just about the big things. We didn't know about the little yet critical stuff– like the labyrinth of patient privacy laws, drug and lab interactions, or the fact that not everyone can read tiny prescription labels.

We didn't think about the nurse who's juggling five tasks at once or the elderly patient who's still trying to figure out how to make a video call on their new smartphone.

So, we found ourselves in the Valley of Despair. We thought we knew it all, but in reality, we knew nothing. But that didn't stop us!

What did we do? We decided to open our own clinic. That's the opposite of what a typical tech startup would do. Tech startups make apps and UIs, not a freaking clinic.

But if we wanted to truly understand the real problems faced by patients and care providers, we had to dive in deep. So, we built a practice right next to our office.

But we went even further; we brought a doctor onto our product team to guide us on everything we were clueless about.

We'd often go to the doctor with our crazy ideas and ask, "Can we do this!?" and he would say, "Nope, you can't." Back to the drawing board, then!

And, well, slowly, we started learning more and more. Having a clinic next to our office allowed us to quickly absorb information that doctors or receptionists shared with us.

We could rapidly prototype something, show it to the doctors, get their approval, and ship it.

Now, Carbon Health has over 100 clinics across the US and is one of the country's top urgent and primary care providers.

But it was our initial ignorance that gave us the confidence to dive in.

If we had been aware of all the difficulties from the start, we might have convinced ourselves it was too risky, too complicated, too difficult to change the system. We might have thought, "It's not worth it."

The second time atop Mount Stupid

A few years ago, we started Blush, a tool that lets anyone add illustrations from various artists to their creations. This time, not only was I the designer, but I was also the CEO, responsible for raising funds.

Luckily, we got our first check from the CEO of Carbon Health. We quit our jobs, deciding to go all in. Having that first check boosted my confidence, making me feel like Mr. CEO, or Señor CEO in my case.

I thought, "Fundraising? Easy. Millions are going to pour in. Come on, VCs, shower us with your money. I'm El Señor CEO!"

Well, I was so wrong.

Rejection followed rejection, and it was no after no. There were calls where I made a fool of myself, responding to questions with the most absurd answers, "So, how much will you do?" they would ask… "Oh… Billions—TRILLIONS!" I would say.

You've probably heard of impostor syndrome, and, well, I had the worst case of it. But this time, I was a genuine impostor. I had no clue what I was doing.

But that wasn't the end of it. Then came the part about hiring, forming a team, and creating a culture. We thought, "Creating a team culture? Easy. People aren't complicated, right?"

Wrong! VERY WRONG!

For example, not everyone is ready to work remotely. It's hard for beginners because they lose the ability to interact with others and learn quickly.

Or how all decisions get taken in super long threads that become digital black holes where all the requests and information vanish.

Then there's implementing productivity tools that somehow make everyone less productive. Notion. Airtable. Google Sheets. Slack. Figma. Figjam. Dev Mode. Canvas in Slack.

And the culture, like, you figure that it's easy to create cool vibes, but some days, it feels more like I'm in an episode of The Office, and I realize that I'm Michael and super cringe.

I was totally unprepared for all the complexities that come with being a founder. So, I had to humble myself, be ready to learn, listen, be open to feedback from meeting to meeting, and understand what I did wrong and what I could improve.

Then, we started getting some positive responses, and some people showed interest in our doodle venture.

And, wow, VCs must have a WhatsApp group chat because as soon as one VC said yes to us, then EVERYONE wanted in.

It was wild proof that the Dunning-Kruger Effect can strike any time, whether you're a designer, product manager, CEO of doodles, or whatever position.

Summiting Mount Stupid again but with AI

In November, we decided to create our own little AI design app. The goal was to build an assistant that all of us designers and creatives deserve.

You've probably used ChatGPT. You may have even played with generative images, too. It's all so easy, isn't it? You tell it to do one thing with a prompt, and then it does it all for you! You can even ask it to code a little game, and it does a pretty reasonable job.

And now it's everywhere like, "AI can now mimic your voice. AI can now answer your emails. AI can now do your laundry!"

So, we thought, how hard can it be to make these little AI robots do our design work?

I thought, let's start small; let's create a tool that just does landing pages. How hard can that be? And you know what? We did it. In a week, we had a functioning prototype.

But we kind of cheated. We used a ton of templates and color palettes I made to get us started. A simple UI and template kit

So, the AI only did the images and the text, not the actual design.

In essence, we were using PabsGPT. But, hey, you have to fake it till you make it, right?

However, right after that, we said, "Ok, that was a good start. Now let's allow the AI to handle all the other stuff like layout, composition, picking colors, choosing sections, creating flows, adding nice padding, grouping, and all that other stuff."

Easy, right? Just sprinkle some AI magic on it.

Well, we quickly found out it wouldn't be that easy.

Big problems we found:

• Cost—oh my god, it's expensive!
• Speed—It's so slow!
• And the quality—it creates some pretty horrible stuff!

To all the creatives and designers reading this right now, I can tell you this: so far, the robots, by default, are not great designers.

I won't bore you and dive deeper into the first two problems of cost and speed, but the quality stuff is genuinely interesting and applicable to the story.

With quality, we mean good-looking designs that are also accurate and follow your brand or prompt.

We quickly learned that AI is a bit rebellious, which is not a very comforting thought.

So, when you're working on a product or programming an app, you expect your code to be reliable and do exactly what you ask of it.

But with Large Language Models, it's incredibly frustrating trying to make them follow your instructions (even positive/negative examples don't always work) and do so consistently. Sometimes, it just ignores (or forgets?) some of the instructions.

If you think of AI as robots or people on your team, imagine them as team members who don't follow your instructions, forget what you ask them to do, and just do their own thing. Not very helpful, right?

It's this unpredictable nature of GPT that can really drive you nuts.

So, you often find yourself trying different forms of the same instruction to get the best results.

• "Hey, please, can you do it right?"
• "Hey, pretty please, with a cherry on top, could you do it right?"
• "Hey, what if I tip you? Would you do it right?"
• "Hey, I will kill you. I will literally enter the matrix and end you if you fail. Can you do it right now?"
• "No… I'm sorry, Mr. Robot. I love you. Could you please help me with this little thing?"

You end up going through the twelve stages of grief before AI starts getting what you mean.

Sometimes, it has its limitations, too. For instance, the output size might be small or different from what you asked for. You request long code, and, well, it gives you short code.

And if you push it to give you more, it might just enter infinite output mode – that's repeating the same thing, over and over, in a loop, and you don't even realize it. And then, of course, it costs you a fortune.

Now, you might have heard of AI having hallucinations.

You might think, "Hey, that's a good thing, right?" You want a designer robot to be creative, to go smoke a joint and create weird stuff.

But not when that weird stuff includes things like hands with more than five fingers, or people with three legs, or retro-futuristic sci-fi elements that clash with your theme.

So, it's been more challenging than we initially thought, but we're getting the hang of it now.

We're building our own fine-tuned model.

We're setting up infrastructure to collect and annotate data from our designs so the robots can learn what good design is and what it isn't.

It's a cycle of training, evaluating, and teaching again.

Typical evaluation models focus on text, but we're developing a system that scores both text and visuals. It's super fun yet challenging.

Now, Musho can do some crazy stuff, like create designs that actually look good, and it genuinely understands what you need from it. Plus, the results are even better if you're more specific with what you ask of it.

It's not just about doing landing pages but also various UIs, like dashboards and UI components. We're also teaching it to get good in portfolios, e-commerce, and different aesthetics, and building all this has been a blast.

But having reflected on it, we might not have even attempted this if it weren't for that initial ignorance of the Dunning-Kruger Effect that sparked our excitement.

Be a fool, embrace the silly, and stick with it!

The Dunning-Kruger Effect often gets a bad rep, mainly because it's fun to talk about that initial part of the curve. The part where someone thinks they know it all and ends up looking foolish.

We enjoy making fun of them!

But it takes real fools, genuine fools, to persist over the long haul, to keep going even after realizing they know nothing and that the journey will be incredibly tough.

We all need a dash of ignorance to kickstart something new, to make us bold, and to embark on new adventures.

And eventually, we need the courage to persist after feeling the embarrassment of realizing our own foolishness and acknowledging the long and challenging road ahead.

So, I encourage you to embrace a bit of silliness and embark on new adventures.

And, let's not be too quick to judge those overconfident people who know little, but instead, help them on their path to learning.

To achieve greatness, sometimes we need to be a bit clueless to give us that necessary push.