On Friday, Bloomberg reported that Adobe’s AI image generator Firefly had included AI images from competitors in its training data – an example of AI learning from AI.
The report adds a new dimension to Adobe’s claims about Firefly’s ethics. In publicly available articles, Adobe differentiated Firefly from competitors such as DALL·E 3, Stable Diffusion and Midjourney by highlighting Firefly’s “commercially safe” training data.
“Adobe has established the AI ethics principles of accountability, responsibility and transparency,” Adobe wrote in a post.
Related: Getty Images has filed a lawsuit against an AI generative art company for copyright infringement
Firefly relied on licensed Adobe Stock images as well as public domain images, the company said. Adobe even created a bonus compensation plan for artists whose work was included in Firefly’s first release.
Some of these artists submitted images generated by Midjourney and other AI competitors and were compensated by Adobe for their contribution, according to Bloomberg’s report.
Symbol of an ouroboros, or snake, eating its own tail, generated by Adobe Firefly in response to the entrepreneur’s prompt: “A snake eating its own tail in an infinity symbol against a background of mountains, trees, and cloudy sky.” Image credit: Adobe Firefly
While artists had to indicate that their work was created using AI, they did not have to tell Adobe which generator they used.
That means AI image generators like Midjourney could do this even if Firefly isn’t actively searching the internet without permission. And Firefly could be trained on these mid-journey images.
Related: JPMorgan CEO Jamie Dimon says AI could impact “every job.”
Several Adobe employees told Bloomberg about an internal conflict over the ethics of training Adobe’s AI with AI-generated images from competitors. Early in Firefly’s development, some employees disagreed with the company’s decision to include AI images in Firefly’s training data.
In Adobe Stock, the database used to train Firefly, 57 million are marked as AI-generated, according to Bloomberg. That’s 14% of all images in the database.
Adobe responded to the claims saying that only 5% of the images used to train Firefly came from AI images created by other platforms.
“Every image submitted to Adobe Stock, including a very small subset of AI-generated images, goes through a rigorous moderation process to ensure that it does not contain any intellectual property rights, trademarks, recognizable marks or logos, or names of reference artists,” a company spokesperson said Bloomberg.
Related: OpenAI reportedly used more than a million hours of YouTube videos to train its latest AI model
The race to develop the next big AI has increased pressure on companies to find new sources of training data. According to a New York Times report, OpenAI may have trained its latest text-to-video AI generator Sora on YouTube videos, and Google may have done the same.
According to Statista, Adobe’s products have captured the majority of the global market share of leading graphics software, with Adobe Photoshop, InDesign and Illustrator accounting for more than 80% of the market. According to the same source, Canva has 10.26% of the global market.
A Bloomberg report earlier this week revealed that Adobe has begun paying its network $2.62 to $7.25 per minute of recorded video for new data for use in its own Sora competitor.
Related: Authors sue OpenAI because ChatGPT is too “accurate.”