There's something that's been on my mind for a few days, and I finally decided to start this blog just to write about it. Maybe it'll go nowhere and nobody will ever read this. But even so, at least I'll have gotten it off my chest.
I'm sure that by now you've seen the 80s photos. (Right?!) At least for me, my entire feed became 80s photos. Coworkers, college friends, my family, even brands got into the "80s trend". From one day to another, everyone suddenly started looking like they'd just walked off the set of a Timbiriche music video. Neon jacket with lively colors, feathered hair, soft-focus glamour lighting. Maybe a sunset in the background that screams "synthwave playlist cover."
It was super fun at the beginning, but after the next five and then fifteen all of that started land (at least for me) in a weird way. Why does everyone look veeeery suspiciously similar, like almost the same?
Like someone almost did a Ctrl-C Ctrl-P on an entire generation of people who are very much not from the same generation or even same country. My friend from São Paulo and my colleague in Mexico City somehow ended up as twins separated at birth in a fictional 1985 high school. Even when clearly São Paulo and Mexico City looked very different from each other in the 80s, they ended up wearing a similar jacket, similar haircut, pretty much similar everything.
In any case that was when I came to the realization of "I think I've seen this before."
The déjà vu
A while back, I was chatting with one of my best friends about a kinda comparable situation, except with text. Not sure if you remember when everyone started using AI to write their LinkedIn posts and suddenly the entire platform sounded like the person that says in the middle of their posts "But here's what nobody tells you..."
People who are genuinely smart and interesting and unique suddenly sounded the same.
Now, with text, it's sneaky. You read one post, move on, read another. They feel similar, but you rarely see them side by side, so the pattern is harder to catch.
With images though? You see ten 80s portraits on your feed at the same time and the pattern jumps out at you. So naturally, I went down a rabbit hole as I had to understand "why" this was happening.
Nerd moment, but stay with me
Here's the thing about how these image models work. When you type "make me look like I'm from the 80s" into an AI image tool, the model doesn't think about YOU. It thinks about probability.
These models were trained on millions and millions of images, and during that training they picked up on statistical patterns, basically what "the 80s" looks like according to the aggregate of every 80s-related image it's ever seen. And when you ask it to generate something, it just goes straight for the most statistically likely version, not the most creative one.
There's a concept in statistics called regression to the mean. If you ask a huge population to do the same thing, the results cluster around an average. The AI model does exactly this but faster and with more confidence. It finds the statistical center of "80s portrait" and parks itself there every single time.
Think of it like this, if you surveyed a thousand people and asked them to draw a house, most of them would draw the same thing. A square with a triangle roof, a door in the middle, maybe a chimney, the same basic mental prototype everyone seems to carry around for "house," regardless of what their actual house looks like. The AI is doing that, except with your face and a neon jacket.
In machine learning there's also this thing called mode collapse. The "mode" is the peak of a probability distribution, basically the most common outcome. When a model collapses toward that mode, it gets stuck generating the same type of output over and over. Basically, your 80s photo is just the model's best guess at the statistical average of "80s portrait," not a personalized read on you. And everyone else got the exact same best guess.
There's also a parameter called 'temperature' that controls how much randomness the model uses. Low temperature means the model plays it safe: predictable, conservative. And guess what, default settings tend to keep temperature low, because safe outputs make fewer people complain. But safe outputs also mean everybody gets handed the same playlist on shuffle.
So you, your friend, your coworker, and that random person from high school all looking alike in the 80s? That's just math doing its job. The model optimized for the most probable 80s portrait and delivered the same answer a million times because that's exactly what it was built to do.
Ok so what does this actually mean for us?
Here's where I want to be careful, because I really don't want to be the person who tells you to stop having fun or stop following trends. I'm the first silly goose to do things that are just fun, and I'll always encourage you to have fun and post whatever you want.
What I do want to say though, is a bit deeper about ourselves, our unique-ness, and how to show how we really are in this AI era.
If a million people typed roughly the same prompt and got roughly the same output, what does that say? In a way it tells us a million people didn't push back. Nobody said, actually, in the 80s my mom had this very specific ridiculous perm and paired it with these absurd oversized glasses and I want that. Nobody added the weird, personal, specific detail that would've made the image actually theirs.
This is what has really been bothering me because I see it happening in different formats at different moments. Someone opens an AI tool and hits generate without thinking, in emails, presentations, blog posts, documents. The 80s trend just made it visible in a way that was impossible (at least for me) to ignore.
We have these incredibly powerful tools that can do almost anything. And what do we do with them? Basically the same thing, over and over. It's like being handed a lightsaber and using it to open your mail.
It was never about the tool
I think part of what bothers me about this has nothing to do with AI being bad or people being lazy. I think it comes down to something simpler: what makes people interesting is that they're different. You have a perspective, a history, a very specific way of seeing the world that no statistical model can replicate.
And when you use AI without injecting any of that into the process, you're basically volunteering to be average. You're letting the model decide who you are instead of telling it.
The 80s photo trend is honestly the perfect metaphor. The model gave everyone the same jacket because nobody told it about their jacket. Nobody said, I want the one my dad wore in that photo where he's trying to look cool at a party in 1987 and failing SPECTACULARLY. That would've been a portrait. What we got instead is a stock photo with your face on it.
So next time
If I could sum this up, what I'd say is next time you open an AI tool, whether it's for an image or text or a presentation or whatever, before you hit enter, just ask yourself:
Does this prompt have any of you in it?
If the answer is not or not really, throw some of you into it. The weird stuff, the specific stuff, whatever the model has never seen before because it only exists in your brain.
Give AI the same input as everyone else and you'll get the same output as everyone else. That's just probability doing its job.
So yeah. Try to be the weird(er) next time.