How Microsoft is fixing AI bias in blind representation
Microsoft new model policy mirrors the 2002 Trustworthy Computing memorandum by Bill Gates
Microsoft purchased over 20 million minutes of video from Be My Eyes, a nonprofit accessibility platform, to retrain its AI models after discovering they were generating offensive and inaccurate depictions of blind people.
Microsoft's new Head Jenny Lay-Flurrie consolidated Trusted Technology Group, discovered the problem while overseeing accessibility initiatives, which was that AI-generated imagery showed blind people wearing full blindfolds stereotypical caricatures drawn from biased training data reflecting societal prejudices rather than reality.
While they could have simply accepted the flaw, Microsoft decided to go an extra mile in gathering videos of the life experiences of people who are visually impaired through walking around with canes, working with guide dogs, and accessing things in their houses after removing any visible faces to train their model for realistic representations of blindness.
Microsoft has opted to centralise responsible technology into a top-down policy introduced in early 2025 called the Trusted Technology Group.
This model mirrors the 2002 Trustworthy Computing memorandum by Bill Gates, who put reliability above speed in terms of philosophy, a complete antithesis to Silicon Valley’s mantra of "Move fast and break things."
Nonetheless, Reliable CEO Annie Brown cautions that data diversity will not be enough to fix algorithmic biases. Metadata, the way data is labeled and categorised, presents another level of bias that goes unnoticed through superficial audits.
While Microsoft’s investment in data sets is geared towards improving data quality, it does not guarantee that the processes of labeling are also fixed for downstream users.
Early access to Copilot was provided to employee groups with disabilities. Deaf employees received captioning and sign language recognition; neurodiverse employees benefited from lower cognitive loads.
Lay-Flurrie says AI is "leveling the playing field for previously marginalised workers," though disability advocates emphasise including disabled people in decision-making roles, not just as users of solutions designed for them.
-
OpenAI and Microsoft quotes on AI threaten news copyright defense, media outlets argue
-
Russia’s Max app has powerful ‘backdoor’ to spy on users, researchers find
-
OpenAI, Anthropic open their doors, watchdogs want full access
-
Samsung Gallery now syncs with Google Photos: Here's how
-
Only 25% of enterprise AI pilots reach production, Deloitte finds
-
New York now requires utilities to report their AI use
-
Salesforce CEO calls Nvidia ‘exquisite’ as AI race heats up
-
France probes smart glasses in sexual harassment and privacy cases
