On Friday, Bloomberg reported that Adobe's AI image generation tool Firefly included competitors' AI images in its training data. This is an example of AI learning from AI.
This report brings a new dimension to Adobe's claims about the ethics of Firefly. In a public article, Adobe differentiated his Firefly from competitors such as DALL E 3, Stable Diffusion, and Midjourney by highlighting Firefly's “commercially safe” training data.
“Adobe has established AI ethical principles of accountability, responsibility, and transparency,” Adobe wrote in one post.
Related: Getty Images begins lawsuit against AI generative art company for copyright infringement
Firefly was drawn from public domain images in addition to licensed Adobe Stock images, the company said. Adobe has also created a bonus compensation plan for artists whose work is included in the first release of Firefly.
Some of these artists submitted images generated by Midjourney and other rival AIs and were paid by Adobe for their input, Bloomberg reported.
Ouroboros, or a snake eating its own tail, generated by Adobe Firefly in response to the entrepreneur's prompt “Snake eating its own tail with an infinity symbol against a background of mountains, trees, and cloudy sky” symbol. Credit: Adobe Firefly
Artists needed to know that their work was created with AI, but they didn't have to tell Adobe which generator they used.
This means that even if Firefly isn't actively scraping the internet without your permission, an AI image generator like Midjourney might be. And Firefly could be trained on those images of his Midjourney.
Related: JPMorgan CEO Jamie Dimon says AI has the potential to impact 'every job'
Multiple Adobe employees spoke to Bloomberg about internal conflicts over the ethics of training Adobe's AI using competitors' AI-generated images. Even in the early stages of Firefly's development, some employees disagreed with the company's decision to include his AI images in Firefly's training data.
According to Bloomberg, Adobe Stock, the database used to train Firefly, has 57 million labeled as AI-generated. This represents 14% of all images in the database.
Adobe responded to this claim by saying that only 5% of the images used to train Firefly were from AI images created on other platforms.
“To ensure that all images submitted to Adobe Stock, including a small portion of AI-generated images, are free of IP, trademarks, recognizable text or logos, and the names of referring artists. goes through a rigorous moderation process,” a company spokesperson said. told Bloomberg.
Related: OpenAI reportedly used over 1 million hours of YouTube videos to train its latest AI model
The race to develop the next big AI is putting increasing pressure on companies to find new sources of training data. OpenAI may have trained its latest text-to-video AI generator Sora on YouTube videos, and Google may have done the same, according to a report in The New York Times.
According to Statista, Adobe products account for the majority of the global market share for major graphics software, with Adobe Photoshop, InDesign, and Illustrator accounting for more than 80% of the market. According to the same source, Canva accounts for his 10.26% of the global market.
Bloomberg reported earlier this week that Adobe has started paying networks $2.62 to $7.25 per minute of recorded video for fresh data to use with competitor Sora.
Related: Authors are suing OpenAI because ChatGPT is too 'accurate'
