Rethinking Discrimination Law in AI Era

Aug 30, 2024

Lecture Notes: The Theory of Artificial Readability and Discrimination Law

Introduction

  • Lecture by a fellow invited to Berlin to discuss a paper titled "The Theory of Artificial Readability: Protecting Other Groups and Discrimination Law."
  • Speaker's background: Central and lawyer, part of Oxford research group "Governance of the Merch of the Conscious."
  • Group's focus: Emerging technologies, legal responses, and societal impacts.

Overview of AI and Decision Making

  • AI is used in significant decision-making areas like loans, insurance, education, and criminal justice.
  • AI can cause harm, often discriminating against already marginalized groups based on gender, ethnicity, ability, sexual orientation, etc.
  • AI creates new group classifications based on non-traditional metrics (eye movement, mouse movement, heart rate) for decision-making.

AI and Non-Traditional Grouping

  • Explores how AI forms new groups that may not align with traditional bases for discrimination, such as being a dog owner or based on behavioral data.
  • Raises concerns about AI's ability to influence decisions without human bias but still perpetuating unfairness.

Case Studies and Real-Life Implications

  • Austria (2019): Employment agency's algorithm discriminated against women, people with disabilities, and older people.
  • UK (2020): Algorithm for A-levels favored private school students, harming students of color and low socioeconomic backgrounds.
  • Social Media Moderation: Discrimination against LGBTQ+ content.

Challenges with Current Discrimination Laws

  • Current laws protect against discrimination on immutable traits (e.g. gender, race, age).
  • AI's groupings challenge this by focusing on mutable or non-traditional characteristics.
  • Examples: Obesity, chronic sickness, and intersectional discrimination cases have failed in court.

Philosophical and Legal Considerations

  • Immutability and Choice: AI groups don't fit neatly into immutable or protected choice categories.
  • Relevance and Arbitrariness: AI can find relevance in seemingly irrelevant traits.
  • Historical Oppression: Traditional groups have a history of oppression, which AI groups lack.
  • Social Saliency: AI groups often lack cultural or social cohesion.

Proposal for New Legal Theories

  • Suggests a need for a new legal theory to address AI-created groupings: "Artificial Immutability."
  • This new theory would consider arbitrary and uncontrollable factors that AI uses in decision-making.

Implications for Society and Law

  • Calls for a shift from material definitions of groups to formal ones, focusing on how groups are created.
  • New guidelines for determining acceptable criteria in AI decision-making.

Conclusion

  • Emphasizes the need to rethink discrimination law in response to AI's capabilities.
  • Urges for more engagement and understanding of AI's impact on fairness and decision-making.