How I'm fighting bias in algorithms | Joy Buolamwini

Algorithmic Bias and the “Coded Gaze”

Joy Buolamwini explains algorithmic bias, which she calls the “coded gaze”: unfairness embedded in algorithms that can spread widely and quickly. She describes repeatedly finding that facial-analysis systems could not detect her face, even while detecting others. In one project, she had to wear a white mask to use generic facial-recognition software.

Why Bias Occurs

Buolamwini traces the problem to machine-learning training data. A system learns to recognize faces from examples. If those examples are not diverse, it may perform poorly on people whose faces differ from the patterns represented in its training set. She argues that more inclusive, “full-spectrum” datasets could help address this issue.

Risks and Consequences

The stakes extend beyond inconvenient errors. Facial-recognition tools are used by police, sometimes without accuracy audits or adequate regulation. A mistaken identification can have serious consequences for civil liberties.

Buolamwini also notes that machine learning is used in decisions about hiring, loans, insurance, education, pricing, policing, and prison sentences. Biased systems in these areas may produce unfair outcomes.

The “Incoding” Approach

Buolamwini proposes three principles:

• Who codes matters: Build diverse teams that can identify one another’s blind spots.
• How we code matters: Consider fairness during system development.
• Why we code matters: Make social change a priority, not an afterthought.

Proposed Actions

She suggests collecting people’s reports of bias, auditing existing software, and creating more inclusive training sets—for example, through a “Selfies for Inclusion” campaign.

She also announces the Algorithmic Justice League, where people can report bias, request audits, and help test systems via codedgaze.com.

Speaker and Sources

Joy Buolamwini is the speaker. She cites a Georgetown Law report on facial-recognition networks and refers to data scientist Cathy O’Neil’s book Weapons of Math Destruction.

Reviews, Guides, or Tutorials

None are presented. This is a talk analyzing algorithmic bias and proposing ways to address it.