Thomas Crul - Algorithmic bias is everywhere (especially at Breeze) - what can we do about it?

Ai

Learn how dating app Breeze tackles algorithmic bias in their recommendation system, exploring ethical solutions while balancing user privacy, legal compliance & business goals.

Key takeaways
  • The Breeze dating app discovered bias issues in their recommendation system, particularly regarding ethnicity and racial bias

  • Key challenges stem from predominantly European user base and potential algorithmic discrimination against users of non-European ethnicity

  • Dutch Human Rights Institute confirmed that interventions aimed at addressing ethnic discrimination in dating apps are legally allowed

  • Current approach uses the Monk skin tone scale to measure diversity metrics while maintaining user privacy and GDPR compliance

  • Team organized consultations with 50 experts and ethics committees to properly address the issue

  • Business model differs from traditional dating apps by only charging when matches occur, not through subscriptions or ads

  • Four main intervention options identified:

    • Self-identification audits
    • Introducing test profiles
    • Using paid agents
    • Anonymizing existing data
  • Technical implementation requires careful balance between measuring bias and protecting user privacy/data

  • Need for complete voluntary consent and transparency when collecting ethnicity data

  • Recommendations:

    • Start with measuring and understanding bias before implementing changes
    • Get diverse perspectives when designing solutions
    • Focus on leveling the playing field without introducing new biases
    • Consider intersectionality with other demographic factors