What I Learned When My AI Kermit Slop Went Viral
First, I want to apologize. My Kermit the Frog post was not entirely sincere.
This particular post of mine has been viewed more than 10 million times, which is far more than I expected. But I did expect something. Social networks have never been the realm of good faith or authenticity; trolls and other engagement baiters have been able to engineer their own virality for years and years, simply by correctly predicting what large numbers of people will respond to. Donald Trump’s TikToks don’t happen by accident; nor did Kamala Harris’s embrace of “brain rot” videos. Each campaign is constructing media that it believes can travel in algorithmic feeds. That’s also what I did when I put together my post, which featured a couple dozen AI-generated images of Kermit the Frog.
Allow me to explain. Last weekend—delirious from a lack of sleep and hoping that my screaming toddler would soon settle down in his crib—I was tapping around on my phone in a kind of fried stupor. My mind struggled to latch on to anything. Each of the apps on my home screen seemed to promise only more boredom. I was the sort of trapped that many parents of young children might recognize: A demand for attention could come at any moment, so I couldn’t lose myself in a book or a bike ride. But I was looking for a diversion.
[Read: What did people do before smartphones?]
Then I had an idea. I decided that it would be fun to use Bing Image Creator, based on OpenAI’s DALL-E technology, to help me replace each app icon on my iPhone’s home screen with a thematically appropriate image of the world’s greatest muppet. (Why? You’d have to ask my psychiatrist.) Instead of the basic Gmail icon, I contrived an image of Kermit buried under a massive pile of envelopes. Instead of the basic green phone icon, Kerm chatting on a yellow landline.
The final product was an absurd, borderline-deranged home-screen grid of 24 bespoke frogs. The creation of each one required a series of specific prompts from me. There was Calculator Kermit and Photos Kermit. Authenticator Kermit was dressed like a police officer and wielded a massive baton. My job complete, I took a screenshot and sent it to a friend, who replied, “Damon I truly truly fear for you.” About halfway through the project, I had developed an inkling that her message seemed to confirm: People on the internet would probably respond to this. I could use my Kermits to go viral.
Everyone loves Kermit, of course, and that could only help me. But just as important was the fact that I had made the images using generative AI, a hyper-polarizing technology with passionate boosters and passionate critics. My content would have to appeal to both groups in order to go as far as possible. So I tried to walk a middle path. I typed an ambiguously worded post that nonetheless contained a sharp opinion that people could react to: “People will be like, ‘generative AI has no practical use case,’ but I did just use it to replace every app icon on my home screen with images of Kermit, soooo.” Then I embedded the before and after images of my home screen, and published simultaneously on X and Threads.
The reactions were swift, and they haven’t stopped. A lot of people just love the images. Others have accused me of destroying the environment, thanks to generative AI’s water and energy use. (I suppose I’m guilty on that count; alas, every online action takes its toll.) Quite a few people have criticized me for leeching off Disney’s intellectual property. (Another fair knock, given that generative AI is trained on tons of copyrighted material.) Some seem to view me as a tech bro or 4chan creep, perhaps because for the YouTube app, I had generated an image of Kermit watching Pepe the Frog—I meant it as a reference to the purportedly radicalizing content that the site has hosted, not as an endorsement of the symbol.
And many people have posted that I played myself, allowing the AI to do the “fun,” imaginative stuff while I took on the rote task of changing the app icons. Those people are wrong: Writing the prompts, looking at the outputs, and adjusting my asks in response was like playing with a toy. By contrast, one person attempted to write a program that would automate every step of the process I had undertaken. Although arguably impressive on its own merits, it appeared to produce bland, interchangeable, witless icons. No fun.
The truth is that the AI didn’t just do everything for me. I came up with little details that some people delighted in (a blond-wigged Kermit snapping a selfie for the Instagram icon, Kermit climbing out of a filthy sewer for X), I tweaked and iterated on the prompts until the outputs were right, and I selected the options I thought looked the best. Even the images that some took as evidence of the uselessness of generative AI (an icon for The Washington Post app bearing the nonsensical headline “NEW HASPELES”; a calendar icon showing the month “EOMER”) were chosen on purpose. It seemed funny and appropriate to include art with some glitches, given AI’s well-documented problems, though avoiding them would have been easy. (For the Atlantic app, of course, I made sure to choose an output with the correct spelling.)
[Read: Generative art is stupid]
That’s not to say that I believe what I did was creative, exactly. The feeling reminded me a bit of editing a talented writer (albeit a nonhuman plagiarist in this case): I gave direction and received something in response, but the fundamental essence of the work did not emerge from my mind. As in working with a person, there was room for surprise—when the image generator took it upon itself, for example, to add a pair of breasts to Kermit for the Instagram icon. (I promise I did not ask for them.) You can nudge the program in one direction or another, but every press of the “Create” button is a bit like pulling a slot machine.
This is one reason generative AI is such an ideal match for the social-media era. These programs are now nested within X, Facebook, Instagram, and Snapchat—apps that are defined not just by endless scrolling but by the downward tug from the top of your screen to refresh and get something new. AI images are a confection just like the other algorithmically served junk people now spend so much time consuming. Having a home screen filled with Kermits isn’t actually practical. The effort was entirely about entertaining myself and getting engagement, not remaking how I actually navigate my phone. (I reverted to the default app icons almost immediately, because the Kermits all blurred together and made the device harder to use.) It’s no wonder that social-media companies are pushing generative AI; the technology feels like it offers both a way to melt time and a shortcut to the kind of numbers-go-up posting that makes these networks so compulsively usable. As my colleague Charlie Warzel wrote last month, that plug-and-play quality has given generative-AI images a certain utility for the MAGA set, who routinely embrace outrageous falsehoods for political gain. They can now illustrate and post in seconds whatever meme they’re using to rally the base on a given day. Likewise, spammers have found that it pays to flood Facebook with attention-grabbing AI slop.
So here is a use for generative AI: It is lubricant for broken algorithmic machinery. Pour it into a social network, and if you’ve done the alchemy right, the gears will turn and turn. This is the internet’s synthetic maximalist moment, where fake content leads easily to superficial interaction. I soon started to notice that many of the typed responses to my post seemed to be following a script, that they were sent from anonymous accounts that barely followed (or were followed by) anyone at all. I’m certain that many were bots, interacting with a JPEG file that had also been made by one—albeit with my mischievous prompting.
The informational environment has become hopelessly junked up, and the way it works can be dispiriting to even the most cynical of the extremely online. But I have to admit that watching my Kermit post go viral was, dare I say, fun. I’m sure many of the actual people who responded to me felt it too. I was amused. Perhaps when we look back on the generative-AI revolution, we’ll realize that chasing this feeling is the ultimate reason for many of these programs—especially as they enter social apps that are designed to prioritize engagement.
We’re a long way from Amusing Ourselves to Death, Neil Postman’s famous 1985 book, which argued that television would lead the public to privilege spectacle over substance. But it’s clear that Postman saw around the right corner. Many prognosticators have said quite a lot about AI’s existential risks, that the technology could be used to construct bioweapons and God knows what else. In the meantime, aided by other sophisticated machines—and, sometimes, an exhausted parent on an iPhone—it’s a grade-A brain softener. Use with caution.