Microsoft Scientist Calls AI Training the "Largest Theft of Labor in Human History"
Microsoft's Director of Applied Science, Dr. Brent Hecht, has described AI training as perhaps the "largest theft of labor in human history", according to documents unsealed on Thursday. Another Microsoft file refers to the result as a "doom loop". Its own data shows that Copilot significantly reduced news click-throughs, cutting them by up to 94%.
Brief Unsealed in OpenAI Copyright Litigation
A 92-page brief, unsealed in the consolidated OpenAI copyright litigation before Judge Sidney H. Stein, opens with quotes from Hecht. The brief sits in the hands of plaintiffs, including The New York Times, Digital Content Next, the Daily News group, and the Center for Investigative Reporting, among others.
Key Findings from the Brief:
- Microsoft's AI content strategy has created a "doom loop" that threatens both its own models and the web as a whole.
- Click-through rates for news websites, including those of The New York Times, the Daily News group, and Ziff Davis, dropped by as much as 94% when compared to traditional Bing search.
- Hecht states that large models replicating patterns in their training content pose an "astonishing theft of unprecedented proportions".
- The brief argues that the economic foundations of Microsoft's suppliers are at risk, as the end product threatens the economic value of the content creators.
Microsoft's Concerns:
- A ruling in Microsoft's favor would undermine the concept of "fair use".
- AI technology risks "significantly disrupt[ing] the employment of the very people who generated the data" on which the models were trained.
- One Microsoft document warns of a "real risk" that the technology could "destroy[s] its supply chain".
OpenAI's Perspective:
An OpenAI software engineer, whose name is not disclosed, states that "users won't click" on links, regardless of how prominent they are, highlighting the substitutive nature of AI-generated content.