Desperate_Yuppie, who goes by @Iron_Yuppie on Twitter or X or whatever you want to call that website, is a long-time Epsilon Theory reader who participates actively in our online community of thousands of paid subscribers – The Epsilon Theory Forum. The Forum is honestly one of the best things on the interwebs today, a safe space for full-hearted people to voice their concerns and hopes and observations about today’s world without the emotional cudgel of cancellation or the emotional crutch of an echo chamber. You know … what social media should be! And yes, it’s for paid subscribers only, which is how we keep the trolls at bay. Enjoy this sample of what you’ll find on the Forum (there are hundreds of active threads just like this), and if this sounds like your kind of place, please join us!
As with all of our guest contributors, Desperate_Yuppie’s post may not represent the views of Epsilon Theory or Second Foundation Partners, and should not be construed as advice to purchase or sell any security.
Last week Bing launched their AI image creation engine, powered by Dall-E 3, and it was a big hit. Hilarity, followed by panic, followed by even more hilarity, ensued.
For those of you with healthy and productive lives you almost certainly missed this as it happened on TwitterX. Here’s the synopsis:
Dall-E 3 is a generative AI that creates art (but not Art ) in multiple styles based on your prompts. If you want a digital art picture of a capybara trick-or-treating in an American suburb it’ll make it for you. See:
Neat, right? There’s just one tiny, minor, hardly-worth-mentioning problem…
The people at Bing have no idea how the internet works.
Almost immediately people started having some fun with this new toy and Twitter was flooded with pictures that were meant to both entertain and offend. SpongeBob apparently did 9/11. Kirby (a Nintendo character) also was responsible for 9/11, according to photographic evidence I saw. This particular trend started who knows where, but a small time tech journo heard about it and that’s when things really took off. This hectoring scold was so incensed by people having fun that he decided to start prompting Bing to make offensive images, then he wrote about it. He then went so far as contacting Nintendo to ask them how they felt about these sorts of images. For those of you keeping score at home here was the progression:
Hears people are making these offensive images
—> Goes and makes his own offensive images
—> Gets offended
If you think political journalism has been captured by activist hacks you should see the state of tech journalism.
Anyway, Bing engineers worked to shut down the potential trigger words in the prompts so Kirby could no longer do 9/11.
But of course the internet was undeterred by such modest barriers. What followed was exactly what any sentient observer would have expected.
Taylor Swift and Travis Kelce wearing stereotypical Native American headdresses. Taylor Swift in an SS uniform (many of these), Taylor Swift as a Roman centurion at the crucifixion, OJ Simpson jumping out of the woods at Taylor Swift…while she’s in an SS uniform. You’re getting the idea, right?
Swift is nothing if not careful about her NIL, and she probably doesn’t love the idea of having pictures of her as a Nazi circulating the internet. So who precisely does she talk to about getting this sort of thing pulled down? Who’s responsible when this all goes from harmless trolling to something a lot darker? Because remember, you can choose the style of your pictures and photorealistic is one of those choices.
That lovely young lady is not real. She’s the result of the prompt ‘young Latina woman working in a physics lab, photorealistic’. We should pause for a moment and reflect on how unbelievable it is that this technology exists and at how rapidly it went from glitchy and weird to simply working well. It’s quite something.
But see this is where it gets very messy very fast. All those trollish pictures I mentioned above were done in the style of digital art, basically a Cartoon. I suspect the people making them knew better than to ask for photorealistic images of actual, recognizable public figures. But not everyone will be so restrained.
Three days before an election a local newspaper runs a picture of the incumbent Congressman and his young female staffer looking a little too close. He loses the tight race and his reputation is ruined. A week later it comes out that the image was AI generated and had been leaked to the paper through various opposition groups. Where does he go to get his job back? How does he rebuild his reputation and his standing in the public eye?
The answer, according to Bing and every other company involved in this work is basically the shoulder shrug emoji because nobody has bothered to actually game this all out.
In the rush to get a product to market nobody thought about what happens when said product is in widespread, uncontrolled use. [Insert Ian Malcolm quote here] What are the laws governing this stuff? If I ask AI to make me a photorealistic picture of some celebrity doing something wrong-but-believable, who does he/she sue? For sure not me, because I didn’t make the image. In fact I can’t use the image for commercial purposes because it is by definition not mine. So does the celebrity sue Microsoft? What about the half dozen other companies that are doing this work? If Bing locks down prompts maybe I find a loophole on Stable Diffusion XL or Midjourney. This is chaos being unleashed on the public and all because the leaders of government and tech couldn’t be troubled to think five minutes into the future. Sound familiar?
Bonus points for anyone who gets this