First, I want to apologize: my Kermit the Frog post wasn’t entirely honest.
My post has been viewed over 10 million times, far more than I expected. But I also expected it. Social networks have never been the domain of sincerity or authenticity. Trolls and engagement hunters have been able to create their own virality for years simply by predicting exactly what most people will respond to. Donald Trump’s TikTok didn’t happen by accident, nor did Kamala Harris’ endorsement of “brain rot” videos. Each side builds a medium that they believe can go viral with an algorithmic feed. I did the same when I put together a post featuring dozens of AI-generated images of Kermit the Frog.
Let me explain. Last weekend, I was sleep-deprived and disoriented, working on my phone in a kind of numbness, hoping that my wailing toddler would be settled in her crib soon. My mind was struggling to focus on anything; every app on my home screen only seemed to invite further boredom. I was in a kind of caged-in state that many parents of young children will recognize. I couldn’t immerse myself in a book or go for a bike ride, because a demand for my attention might come at any moment. But I was looking for a distraction.
So an idea came to me. I thought it would be fun to use Bing Image Creator, based on OpenAI’s DALL-E technology, to replace each app icon on my iPhone home screen with a themed image of the world’s best Muppet (why? Ask your psychiatrist). Instead of the basic Gmail icon, I came up with an image of Kermit buried under a pile of envelopes. Instead of the basic green phone icon, I came up with an image of Kermit chatting on a yellow landline.
The final product was a silly, almost insane, grid of 24 custom frogs on my home screen. Each one required a specific set of instructions from me to create: Kermit the calculator and Kermit the photo. Kermit the verification was dressed like a police officer and wielded a giant baton. When I was done, I took a screenshot and sent it to a friend, who replied, “Damon, I’m really worried.” About halfway through the project, her message seemed to confirm that people online would probably respond to this: Kermit could help spread the word.
Of course, everyone loves Kermit, and that can only be a plus for me. But just as important is the fact that I used generative AI to create the image. Generative AI is an extremely polarizing technology with ardent supporters and ardent detractors. For my content to be most widely accepted, I need to appeal to both groups. So I tried to walk a middle ground. I typed out a vaguely worded post with some sharp points that people could respond to. “Some people will say, ‘generative AI has no practical use,’ but I actually used it to replace all my app icons on my home screen with images of Kermit.” I then embedded the before and after images of my home screen and published them simultaneously on X and Threads.
The reactions were swift and endless. Many people just loved the image. Others accused me of damaging the environment with the generative AI’s water and energy use. (I think I’m guilty of that too. Unfortunately, every online action has a cost.) Quite a few criticized me for exploiting Disney’s intellectual property. (A valid criticism, given that the generative AI is trained on a ton of copyrighted material.) Some people seem to have dismissed me as a tech nerd or a 4chan weirdo, probably because I generated an image of Kermit looking at Pepe the Frog for the YouTube app. I didn’t mean it to endorse the symbol, but as a reference to the supposedly extremist content the site hosts.
And many have posted that I got an AI to do something “fun” and imaginative, taking on the tedious task of changing app icons. These people are wrong. Writing the prompts, watching the output, and adjusting the requests accordingly was like playing with a toy. In contrast, one person tried to write a program that automated every step of the process I did. While certainly impressive on its own merits, it seemed to generate bland, interchangeable, meaningless icons. Not fun.
In truth, the AI didn’t do it all for me. I came up with little details that some people would be happy with (Kermit in a blonde wig taking a selfie for the Instagram icon, Kermit crawling out of a filthy sewer for X). I tweaked and iterated the prompts until I got the output right, and then chose the option I thought looked best. I also purposely chose images that some people took as evidence of generative AI’s uselessness (The Washington Post app icon with the cryptic headline “NEW HASPELES,” a calendar icon showing the month “EOMER”). AI issues are well-known, so I thought it was odd and appropriate to include some glitchy art, but it would have been easy to work around them. (Of course, in the Atlantic app, I made sure to choose the correctly spelled output.)
That’s not to say I believe what I did was creative. The feeling reminded me a little of what it’s like editing a talented writer (albeit, in this case, a non-human plagiarist). I gave instructions and received responses, but the underlying essence of the work didn’t emerge from me. As with working with humans, there was room for surprises, like when the image generator took it upon itself to add two breasts to Kermit for his Instagram icon. (I didn’t ask for the breasts.) I could move the program in either direction, but each time I pressed the “Create” button, it felt like pulling a slot machine.
That’s one reason generative AI is a perfect match for the social media era. These programs are now built into X, Facebook, Instagram, and Snapchat. These apps are defined not just by endless scrolling, but by pulling down from the top of the screen to refresh and get something new. AI images are a treat, just like the other algorithmically-fed junk people spend so much of their time on now. It’s not really realistic to want to fill your home screen with Kermit. This effort was entirely about entertaining and engaging ourselves, not remaking how we actually interact with our phones (we quickly reverted to default app icons after Kermit got all blurry and made our devices harder to use). It’s no wonder social media companies are pushing generative AI. It feels like the technology offers both a way to melt time and a shortcut to the kind of numbers-boosting posts that make these networks so compulsively usable. As my colleague Charlie Warzel wrote last month, the plug-and-play nature of generative AI imagery has given it some utility for MAGA crowds, who routinely embrace outrageous lies for political gain. They can illustrate and post memes in seconds that they use to rally supporters on any given day. Similarly, spammers have discovered that they can make a profit by flooding Facebook with attention-grabbing AI garbage images.
The use of generative AI is as oil for a broken algorithmic machine. Pour this into a social network and, hopefully, the alchemy will keep the gears turning. This is the internet’s synthetic maximalist moment, where fake content easily leads to superficial interactions. I soon noticed that many of the typed replies to my posts seemed to follow a script and were sent from anonymous accounts that hardly followed anyone (or were followed back by anyone). Many were bots, no doubt interacting with JPEG files, though this was at my naughty prompting.
The information environment is hopelessly trashed, and its mechanics can be discouraging to even the most cynical people on the internet. But I have to say it was fun watching Kermit’s post go viral, and I’m sure many of the people who responded to me felt the same way. I was amused. Perhaps when we look back on the generative AI revolution, we’ll realize that chasing this feeling is the ultimate goal of many of these programs, especially as they make their way into social apps designed to prioritize engagement.
Neil Postman’s famous 1985 book, “Fun Enough to Kill,” which argued that television encourages the masses to prioritize spectacle over substance, is a long way from the original. But it’s clear that Postman had the right perspective. Many prophets have spoken at length about the existential risks of AI, stating that AI technology could be used to create biological weapons or for god knows what other uses. Meanwhile, with the help of other sophisticated machines and sometimes jaded parents turning to iPhones, AI has become a brain softener of the highest order. Use with caution.
