EU Mandates AI Companies Stop Using Pirated Data and Respect Copyright

EU's New Code of Practice for AI

• 🇪🇺 The European Commission has released guidelines for AI safety, copyright, and transparency, specifically targeting companies developing advanced generalist chatbots like OpenAI, Google, and Anthropic.
• ⚖️ The code of practice has drawn criticism, with big tech lobbies arguing it's too burdensome, while civil society groups claim it's been watered down, potentially allowing dangerous models to reach users without adequate scrutiny.

Transparency in AI Training Data

• 💡 AI models rely heavily on training data, and the new code requires signatories to be transparent about their training methods, the data used, and how it was acquired.
• 🧐 Companies must provide evidence of obtaining rights for third-party data and allow independent external evaluators access to their AI models and relevant training data.

Addressing Copyrighted Material

• 🎨 Artists and authors have raised concerns that their copyrighted material has been used without permission to train AI models.
• 🚫 The code of practice asks companies to commit to not using pirated databases and to allow rights holders the ability to opt out of their work being used for training.
• 🏛️ This initiative follows recent US court cases that have begun to address AI training data, with judges generally leaning towards fair use for copyrighted material but not for pirated content.

The Exploding Market for High-Quality Data

• 📈 The demand for high-quality, proprietary data is surging, driven by the need to respect copyrights and avoid pirated content, as well as the competitive advantage better data provides.
• 💰 Investments like Meta's $14 billion into Scale AI highlight the market's intensity, pushing some companies towards alternative data providers.
• 🌍 Companies like Turing leverage a global workforce of freelance contractors to generate proprietary data, offering a different model than scraping internet content.
• 🚀 Startups may face challenges affording high-quality proprietary data, potentially relying on pirated content to compete, though academic datasets and initiatives from companies like Turing aim to provide solutions.