Big AI's content problem: Take the work, keep the money
5 hours ago
- Big AI companies rely on ingesting vast amounts of copyrighted content to train LLMs, often without payment or respect for licenses.
- Court filings quote Microsoft's Dr. Brent Hecht calling the alleged copying an "astonishing theft of unprecedented proportions," possibly the largest theft of labor in human history.
- Internal Microsoft documents reportedly admit generative AI can disrupt the very content creators whose data trains it, risking model collapse and destroying its own supply chain.
- OpenAI and Microsoft defend training on public articles and books as "fair use," arguing it advances public knowledge and isn't an unlawful market substitute.
- OpenAI's corporate representative reportedly testified that no effort was made to detect or remove paywall-protected content from training datasets; cofounder Greg Brockman allegedly reacted "ah, nice" to a suggestion of bypassing firewalls.
- The article argues that refusing to pay creators is not a side effect but the business model of Big AI, while most users fail to click sources or verify AI-generated information.
- A recent court ruling about copyright management information (CMI) is narrow: generating new code without CMI isn't the same as removing it from existing work, and it doesn't exempt AI code from copyright or open-source licenses.
- Legal experts worry open-source licenses are increasingly litigated as contracts rather than copyright licenses, which could bypass foundational questions like ownership, protectable expression, and actual copying.
- AI-generated code may import unknown legal obligations and supply-chain compliance risks, especially when provenance and license terms are absent.
- Big AI's calls for coordinated AI safety and pace-setting could also be seen as competitors deciding market development, raising antitrust concerns despite the need for responsible deployment.
- The central issue is whether AI developers should be allowed to treat all external work as free raw material, then use the resulting products to erase creators' traffic, revenue, and bargaining power.