Heinz wanted to see how much its name had become synonymous with ketchup in the past few months, when AI services that created images based on text prompts exploded across the web.
Heinz commissioned DALL-E 2, an AI image maker, to create a series of images for them in a variety of artistic styles (including "renaissance," "impressionism," "street art," and more) through the agency Rethink. While the outcomes varied stylistically, they were all unmistakably Heinz-centric, with many works specifically referencing the label's distinctive shape. While entertaining, the stunt also served as a powerful flex, underscoring the dominance of the Heinz brand.
However, it also raised a potential problem for trademark holders, which is getting more attention as of late: most of these AI systems appear to have been developed without much thought to intellectual property.
This is an issue that has been discussed for some time by outsiders. DALL-E 2 and other text-to-image converters are "trained" by scouring the Internet and "learning" from hundreds of millions of pairs of words and pictures. A report from earlier this year on TechCrunch mentioned that during the learning process, pornography and duplicates were "filtered out" by DALL-E 2's developer, OpenAI. However, the program can create works that feature logos, trademarked characters, and other forms of intellectual property, such as a Caravaggio-style "angry mob" of Ronald McDonalds protesting working conditions, a Homer Simpson in Psycho, a Spider-Man from ancient Rome, Santa Claus making a purchase on Amazon, and so on.
Midjourney, a research lab, and the open-source project Stable Diffusion, which is similar to DALL-E 2, are both making their products more widely available and popular. The next iteration will focus on video, with steadily improving image quality. The majority of the online content we encounter in the next few years may have been created by artificial intelligence, according to a recent prediction by Kevin Roose, a tech columnist for the New York Times.
Consistent with past experiences, any sort of regulatory strategy to establish guidelines for managing this technology is lagging far behind its development. In fact, stock image websites like Shutterstock, iStock, and Adobe Stock are already rife with AI-generated images. (Meanwhile, Getty has banned AI imagery due to "unaddressed rights issues.") The questions of who created and owns these images and how they should be used still need to be answered.
Brands face slightly different but no less frustrating problems. However, as Heinz showed, AI-generated visuals are yet another toy with which forward-thinking businesses can experiment. Some other instances can be cited as well. Specifically, Stitch Fix used the technology to "surface the most informative characteristics of a product in a visual way, ultimately helping stylists find the perfect item that matches what a client has requested in written feedback," as the company put it.
Others have looked into how AI-generated images could be used in advertising. One line of thought is whether or not artificial intelligence will eventually pose a threat to fields like graphic design and the arts.
Marketers may find that AI-generated videos are significantly less expensive to produce than more conventional methods in the near future. With the same budget, an advertiser will be able to create 10,000 different commercials, each tuned to a cluster of like-minded viewers at a moment in time, rather than spending $100,000 on a single television commercial targeting millions of people, as has been argued by Brett Winton, director of research at ARK Invest. It's possible that such scenarios are still in the far future, but in the meantime, it's interesting to note that both Meta and Google are actively experimenting with basic AI-generated videos.
In the near future, IP and branding issues brought up by AI imagery will likely be resolved in real time. The most straightforward example would involve artificially created works that incorporate well-known intellectual property, such as a Disney character or a company's trademarked logo. One prominent IP attorney told TechCrunch that if a Disney princess is recognizable in an image generated by DALL-E 2, then it is likely that The Walt Disney Co. will assert that the DALL-E 2 image is a derivative work and an infringement of its copyrights on the Disney princess likeness. While transformative and fair use arguments might be made, IP owners are likely to take a hard stance on the matter. The question of the legality of allowing these AI systems to "learn" from copyrighted imagery goes one level deeper.
There is also the immediate, mundane concern that logos and other recognizable brand identifiers could be co-opted by amateur AI experimenters and used in jokes and viral content creation for no financial gain. Imagine this as the complete opposite of the Heinz study. These are the unavoidable outcomes due to the fact that brands are a form of cultural material. Janelle Shane, on her "machine learning humor blog," AI Weirdness, has already experimented with using DALL-E 2 "to mess up corporate logos," albeit in a fairly benign way. The end result is a criticism of the relative strength of any given brand along with elements of accidental satire.
It's unclear whether or not a company could use intellectual property law to stop the spread of such content, and it's even less certain that doing so would be wise. When compared to its competitors, DALL-ability E's to visualize our wildest dreams is already astounding. The repercussions, however, will be very tangible.