There has been a heated debate surrounding the training methods of AI models since the emergence of generative AI technology. On one side of the argument are tech giants like OpenAI, asserting that training AI models without utilizing web-scraped copyrighted data is unfeasible. Conversely, artists raise concerns about AI companies appropriating their intellectual property without appropriate compensation.
In a departure from the norm, Adobe has taken a distinctive approach by supporting the former camp. Adobe introduced Light, an image-generating tool seamlessly integrated into its renowned Photoshop video editing software, as a prime example of creating conceptual AI products without relying on copyrighted internet data.
In an exclusive discussion with MIT Technology Review, Adobe’s AI leaders emphasized the necessity of this approach. They highlighted that the development of ethical technology should not undermine commercial viability or the rights of creators. David Wadhwani, senior vice president of Adobe’s digital media business, expressed concerns about the tech industry’s tendency to prioritize capability over consequences, stressing the importance of mindful innovation.
The genesis of Light was influenced by these apprehensions, particularly in response to the backlash from creative circles during the conceptual image surge of 2022. This surge led to legal conflicts over trademark infringement and fair use due to attempts to mimic artists’ styles using conceptual AI models. The proliferation of deepfakes and misinformation further underscored the potential misuse of advanced AI technologies.
Adobe’s strategy with Light diverges from conventional AI practices by eschewing web scraping for data collection. Ely Greenfield, Adobe’s chief technology officer for online advertising, highlighted the company’s commitment to using licensed content for training AI models, primarily sourced from Adobe’s stock photo library. This contrasts sharply with the prevalent industry practice of indiscriminate web scraping, which often results in the inadvertent inclusion of personal or copyrighted materials in AI training data.
Greenfield emphasized Adobe’s meticulous content moderation process, ensuring that objectionable or copyrighted content is not integrated into the training data. This meticulous approach not only enhances content moderation capabilities but also distinguishes Adobe from its peers in the AI landscape.
Adobe’s conscientious stance on data sourcing and content moderation aligns with broader industry discussions around responsible AI development. By prioritizing ethical considerations and user safety, Adobe sets a commendable standard for relational AI applications. The company’s collaboration with industry partners and active participation in initiatives like the Content Authenticity Initiative underscore its commitment to transparency and accountability in AI-generated content.
Overall, Adobe’s approach to AI development not only addresses critical ethical concerns but also resonates positively with its user base. By prioritizing responsible AI practices and user safety, Adobe has not only enhanced its business standing but also contributed to the broader discourse on AI ethics and accountability in the digital age.
