Bloomberg’s recent report highlighted that Adobe’s AI image generator, Light, incorporated AI images from competitors in its training data, showcasing a scenario where AI learns from other AI systems.
This revelation added a new dimension to Adobe’s assertions regarding Firefly’s ethical standards. In its public communications, Adobe distinguished Firefly from rivals like DALL·E 3, Stable Diffusion, and Midjourney by underscoring Firefly’s utilization of “ethically sourced” training data.
In an article, Adobe stated, “Adobe has established ethical guidelines encompassing responsibility, integrity, and transparency.”
Getty Images has initiated legal action against an AI generative art company for alleged copyright infringement.
Adobe clarified that Light sourced images from licensed Adobe Stock content as well as images from the public domain. Furthermore, Adobe introduced an additional compensation program for artists whose work contributed to the development of Firefly.
As per Bloomberg’s investigation, some designers were indeed compensated by Adobe for their contributions to the images used in Firefly.
The artists were obligated to disclose the tools they used to create the images, although this information was not mandatory.
This situation implies that AI image generators such as Midjourney might be actively scraping the internet for images without explicit consent, unlike Firefly. There’s also a possibility that Firefly could have been trained using images sourced by Midjourney.
Jamie Dimon, CEO of JPMorgan, expressed his belief that AI could impact virtually every industry.
Several individuals at Adobe raised concerns internally about the ethical implications of using competitor AI-generated images to train Adobe’s AI, even during the early stages of Light’s development.
Within Adobe Stock, the dataset used to train Light, 57 million images are tagged as AI-generated, accounting for 14% of the total graphics in the collection.
Adobe contested these claims, asserting that only 5% of the images used to train Firefly were sourced from AI-generated content from various platforms.
A spokesperson from Adobe informed Bloomberg that every image submitted to Adobe Stock, including a minute fraction generated using AI, undergoes a rigorous vetting process to ensure compliance with guidelines, excluding internet links, trademarks, recognizable individuals, logos, or artists’ identities.
The intense competition in the AI domain has made it challenging for companies to secure novel sources of training data. Reports suggest that Google may have engaged in similar practices as OpenAI.
Adobe’s suite of products has dominated the global market for graphic design software, with Adobe Photoshop, InDesign, and Illustrator commanding over 80% market share, according to Statista. Canva holds a 10.26% share of the global market, as per the same source.
In a recent Bloomberg report, it was revealed that Adobe has started compensating its network between \(2.62 and \)7.25 per minute of recorded video to acquire fresh data for its Sora project.
Authors are reportedly taking legal action against OpenAI, claiming that ChatGPT’s accuracy is too high.
