Just a few weeks ago, OpenAI announced the release of its most sophisticated image-generating chatbot to date — GPT-4o. It’s a single system that combines the power of ChatGPT and OpenAI’s previous image-generating chatbot, DALL-E. GPT-4o is now one model that is able to learn from not only text but also images. This is a system that will not merely create images from familiar concepts - a zebra with black-and-white stripes standing in a field of red tulips, for instance - but will also generate hyper-realistic images based on a series of complex prompts or unfamiliar, irrational requests, such as a toothbrush with serrated teeth rather than bristles. It’s considered to be a huge step forward in artificial intelligence technology, but at what cost?
There is a grave danger lies in the unfettered ambiguity, the blurring of the visual distinction between reality and fiction, that comes with the ability to generate ever-more-complex and startlingly realistic images. According to a review in Arstechnica, “Shortly after OpenAI launched 4o Image Generation, the AI community on X put the feature through its paces, finding that it is quite capable at inserting someone’s face into an existing image, creating fake screenshots…and much more. It seems like we’re entering a completely fluid media ‘reality’ courtesy of a tool that can effortlessly convert visual media between styles” (arstechnica.com). We must consider what it means to live in a fluid media reality. Fluid, when used as an adjective, describes something that shifts or changes shape, something that is not stable or fixed. Ergo a fluid media reality is one in which media - in this case pictures designed to impersonate photographs - offer an unstable, shape-shifting view of reality. Is this really what we want?
As a high school teacher, I cannot help but consider the effect of all of this on the young women that I teach. What safeguards are being put in place to prevent the abuse of this new “fluid media reality” at their expense? Jackie Shannon, multimodal product lead for ChatGPT, “emphasized the system includes robust safeguards to prevent misuse…the tool prevents watermark removal, blocks generation of sexual deepfakes.” (theverge.com) It’s worth noting that images generated by 4o have no visual watermark indicating that they are AI-created. Shannon downplays this by saying that “all of our generated images will include standard C2PA metadata to mark the image as having been created by OpenAI.”
OK, so the tool can’t be used to create fake-porn images of Taylor Swift. But could a vengeful teenager thinking with her amygdala use GPT 4o or one of its competitors to easily create, say, a very realistic picture of herself cozying up to her arch-rival’s boyfriend? Absolutely. And could that image then be texted, or Snapchatted, or Instagrammed without any acknowledgement that it was AI generated? Yes. It could, and it will be, and it already has been. And in the middle of the prom, or lacrosse practice, or the family trip to Disneyland, are my students going to stop to check the metadata on that photo and verify whether or not it was created by AI? They’re not. Most of us wouldn’t. Is this the world we want to live in? One in which we have to second-guess every image, one in which nothing can be taken at face value, one in which we always have to check the metadata?
This is what concerns me the most - that the young women I interact with every day, who already live in a distorted reality that leads to constant questioning of body image, will be further harmed by inhabiting a world that has moved beyond filters and retouching to complete falsification. How can they compete with that? How will reality ever match AI-generated images of their peers on yachts in the Mediterranean, the valedictorian showing off an acceptance email from USC, a snapshot snuggled up with the projected top-10 draft pick? It’s not only my students who are comparing their lives to this complete falsification who will suffer. It’s those who are partaking in the ruse. When we normalize the creation of completely imaginary photographs, and normalize passing these images off as real - when, in fact, we invite people to test their intelligence and try to suss out what’s real and what’s fake - we are tacitly endorsing this guessing game of fiction vs. reality. We aren’t calling it out for what it is - frightening, abnormal, deserving of strict guidelines - and instead are welcoming it into the fold. We are embracing it, making it a part of our lived experience. Consider whether this is wise for the average teenager, whose prefrontal cortex is nowhere near fully developed.
OpenAI founder Sam Altman has his own thoughts about ChatGPT 4o. This is what he posted to X: “People are going to create some really amazing stuff and some stuff that may offend people; what we’d like to aim for is that the tool doesn’t create offensive stuff unless you want it to, in which case within reason it does.”
Well, then.