algorithmic bias

Meaning

Systematic and repeatable errors in a computer system that create unfair outcomes, such as privileging one arbitrary group of users over others. It often arises from the prejudices of the developers or the historical data used to train machine learning models.

Examples

• The hiring software showed clear algorithmic bias by consistently ranking female candidates lower than male candidates.

  The hiring software showed clear algorithmic bias by consistently ranking female candidates lower than male candidates.

• Researchers are working hard to detect and mitigate algorithmic bias in facial recognition technologies.

  Researchers are working hard to detect and mitigate algorithmic bias in facial recognition technologies.

• If the training data is historically prejudiced, the AI will inevitably inherit that algorithmic bias.

  If the training data is historically prejudiced, the AI will inevitably inherit that algorithmic bias.

• Advocates argue that algorithmic bias in risk-assessment tools can lead to unfair sentencing in courtrooms.

  Advocates argue that algorithmic bias in risk-assessment tools can lead to unfair sentencing in courtrooms.

• We must establish strict regulations to audit automated systems for potential algorithmic bias.

  We must establish strict regulations to audit automated systems for potential algorithmic bias.

Common Mistakes

• ✗ The computer has an algorithm bias.

  ✓ The computer has an algorithmic bias.

  Use the adjective form 'algorithmic' to modify 'bias', rather than the noun 'algorithm'.

• ✗ The AI tool is full of algorithmic biases.

  ✓ The AI tool exhibits algorithmic bias.

  While 'biases' can be pluralized, when discussing the systemic issue as a concept, the singular uncountable form 'algorithmic bias' is preferred in formal registers.

Related Expressions

Practice

• When facial recognition systems fail to identify diverse faces, it is often a direct result of ___.

  💡 The technological term for unfairness built into computer code or training data.

  Show answer

  algorithmic bias

• To prevent ___, developers must ensure their training datasets are diverse and representative.

  💡 A two-word phrase starting with 'a'.

  Show answer

  algorithmic bias