Gender and racial bias in AI

AI feels neutral: code doesn't hold opinions. But the systems we build reflect the people and data behind them, and right now that data skews heavily toward one demographic.

Where the bias comes from

The AI field remains overwhelmingly male and non-diverse: around 80% of AI professors are men, and workforces at major tech companies are similarly skewed. Black representation is even lower: around 2.5% at Google and 4% at Microsoft.

Datasets and algorithms built by a narrow set of people naturally encode a narrow set of assumptions. Once deployed, biased systems amplify that bias at scale.

Two examples that made headlines

• ▸Amazon scrapped an AI hiring tool after it learned to favour male candidates, a reflection of a decade of male-dominated resumes it was trained on.
• ▸Zoom's virtual background feature failed to recognise Black users' faces early in the pandemic, because the underlying algorithm was never trained to.

These aren't just embarrassing bugs. Similar bias has led to wrongful arrests and misidentification, with real consequences for real people.

What actually helps

• ▸Talking openly about bias as a human problem, not a technical glitch.
• ▸Investing more in AI research and testing.
• ▸Not treating tech as a shortcut around deep-rooted societal issues.
• ▸Making the field itself more diverse.

As Princeton's Olga Russakovsky put it: “I don't think it's possible to have an unbiased human, so I don't see how we can build an unbiased A.I. system. But we can certainly do a lot better than we're doing.”