“There are things known and there are things unknown and in between are the doors of perception.” — Aldous Huxley

I’m Huxley Westemeier (26’) and welcome to the 24th and final “The Sift” article of the 2024-25 school year. I’ll be back in August for a summer recap.

______________________________________________________

I just returned from the Regeneron International Science and Engineering Fair (ISEF) in Columbus, Ohio, and even with all of the incredible experiences and new connections I made, one moment has stuck with me that I knew I had to write about. ISEF offers a series of Symposium sessions throughout the week, including judging preparation workshops, creativity, and information about how to apply to college as a researcher.

My personal ISEF project was focused on detecting AI-generated photographs without using machine learning. I’ve learned to recognize the powerful role deep fakes and AI-manipulated images can have in promoting disinformation online in contexts such as social media. Therefore, I was particularly interested in attending a Symposium hosted by David Holz, Founder and CEO of Midjourney. If you’re unfamiliar with Midjourney, it’s a subscription-based AI-art generation company. A user can type in a prompt, and Midjourney’s website will allow them to ‘create’ an artificial image and tweak it however they like.

Holz is an interesting character. His uniform for the week consisted of a beanie, cardigan, and headphones around his neck- even for the awards ceremonies where most judges wore business attire. During the symposium, I was surprised by how often Holz referred to Midjourney as a collaborative approach to imagination. In a 2022 interview with Forbes, Holz described Midjourney’s goal as trying to “expand the imaginative powers of the human species.” I found myself wondering precisely what that expansion encompassed. If we use AI tools like Midjourney even for the “concept-art stage,” something creative professionals are already considering with storyboards or even rough drafts of videos, is it truly our work? Can a picture be beautiful or encompass human emotion if a human didn’t create it?

Additionally, AI-generated images in their current state have two major downfalls that Holz has failed to address: the environmental cost and the potential for misuse. Consider the following picture, from Midjourney V6.1, prompted to look as photorealistic as possible.

It’s undoubtedly a “beautiful” image from a distance. It could convince unknowing social media users unfamiliar with the nuances of AI-generated images, even if the ripples in the water are distorted and the wood texture is off-putting. Creations from AI-generators that claim to focus on imagination and art (including Midjourney and Stable Diffusion) have already been used to promote misinformation, with little to no guardrails in place, especially as Meta recently ended their third-party fact-checking program. AI-manipulated images aren’t simply creative fodder anymore; they can promote real harm. I’d like to see Holz focus on watermarking technologies and other methods to improve the detection of Midjourney images.

According to MIT Technology Review, using AI to generate an image requires around 2,282 joules of energy. That’s equivalent to “riding 250 feet on an e-bike, or five and a half seconds running a microwave.” While image generators like Midjourney are more efficient than large-language models like ChatGPT, they still drain the environment of crucial energy. In March, OpenAI announced that users were generating over 78 million images per day (during the Studio Ghibli portrait phase), burning through the two-year energy consumption of an American household in less than a day.

All of that energy for AI-generated photographs.

If a new trend emerges in the coming weeks where everyone who’s anyone is generating “AI-slop” images of themselves in some random style, pause for a moment to ask yourself if it’s worth sending that request to Midjourney or ChatGPT.

You’ll contribute to saving a sliver of the environment, and you won’t have to look at your fake self dressed up as a cursed action figure. It’s a win-win!