Microsoft Buys 20 Million Minutes of Data to Fix AI Bias as Jenny Lay-Flurrie Leads Trusted Tech
Updated
Updated · CNBC · May 23
Microsoft Buys 20 Million Minutes of Data to Fix AI Bias as Jenny Lay-Flurrie Leads Trusted Tech
2 articles · Updated · CNBC · May 23
More than 20 million minutes of multimodal data from Be My Eyes were bought by Microsoft after its models misrepresented blind people and AI-generated code often missed accessibility needs.
Jenny Lay-Flurrie, who took over Microsoft's Trusted Technology Group in February, said responsible AI requires both building systems correctly and continuously testing, iterating and correcting them with human oversight.
Microsoft anonymized the Be My Eyes material to better train models on blindness, though Reliabl CEO Annie Brown said bias can still persist if metadata and labeling are not handled carefully.
The effort sits inside a companywide responsible-tech structure launched in early 2025, consolidating accessibility and other safeguards even as Microsoft shifts hiring toward AI infrastructure after about 15,000 job cuts in 2025.
Lay-Flurrie and disability advocates argue AI can widen access for deaf, neurodiverse and disabled workers, but only if disabled people are included in decision-making rather than treated as an afterthought.
As Microsoft spends $80B to win the AI race, can its ethical promises keep pace?
AI empowers disabled workers but replaces thousands of jobs. What is the true human cost?
Microsoft Acquires 20 Million Minutes of Blind User Data to Tackle AI Bias and Lead Inclusive AI Innovation
Overview
Microsoft identified a major flaw in its AI models, which often misrepresented blind people due to training on non-inclusive societal data. Led by Jenny Lay-Flurrie, the company took decisive action by acquiring over 20 million minutes of real-world video data from Be My Eyes, a platform serving blind and low-vision users. This unique dataset, captured by blind individuals themselves, was carefully anonymized to protect privacy. By intentionally adding this representative data to its AI training sets, Microsoft aims to correct bias and ensure its AI systems provide more accurate and respectful depictions of blind and low-vision people.