Buy your pass

Beyond trade-offs: How AI can be fairer than we think

OnlineDiscovery - From Molecules to Models

  • Register

    Watch

    * Register (or log in) to the Neural Network to add this session to your agenda or watch the replay

    • Date
    15 October 2025
    Timeframe
    17:00 - 18:00 CEST
    Duration
    60 minutes
    Share this session

    AI is often seen as facing unavoidable trade-offs between accuracy and fairness. The well-known “impossibility theorem” in algorithmic fairness suggests that multiple fairness metrics cannot be satisfied simultaneously—except in unrealistic scenarios such as perfect prediction or identical base rates. But does theory really translate to practice? In this talk, I will present insights from our 2023 FAccT paper The Possibility of Fairness: Revisiting the Impossibility Theorem in Practice. We demonstrate analytically and empirically, across five real-world datasets (including Ukrainian educational data), that when fairness constraints are relaxed in a way that reflects practitioners’ perspectives, large sets of feasible models emerge. These models achieve fairness across multiple criteria simultaneously, challenging the view that fairness is inherently impossible. 
    Beyond theory, I will discuss practical lessons for AI practitioners and policymakers: when fairness is possible, how to evaluate trade-offs, and what tools can guide the design of responsible AI systems. I will also briefly introduce the newly established Center for Data, AI, and Society at the Ukrainian Catholic University, which advances research and education at the intersection of AI, equity, and social good.

     

    Learning Objectives
    By the end of this session, participants will be able to:

    • Understand the implications of the impossibility theorem in algorithmic fairness. Analyze how relaxing fairness constraints changes the feasibility of fair AI models.
    • Apply practical guidelines to assess fairness across multiple metrics.

    Recommended mastery level or prerequisites
    Recommended Mastery Level: Intermediate to Advanced – suitable for graduate students, early career researchers, and professionals in climate science, oceanography, or environmental data science.

     

    Share this session
      Add to Calendar
      Discover more from programme stream
      Discovery - From Molecules to Models
      Convolutional networks beyond the Euclidean space

      Convolutional networks beyond the Euclidean space

      12 November 2025 - 17:00 to 18:00

      Related stories

      Bringing AI into human rights: why dialogue matters

      At the 2025 AI for Good Global Summit, the workshop titled “AI for Human Rights: Smarter, Faster, Fairer Monitoring”, organized...
      Global Summit

      Ending data disparity: Snowflake’s vision for AI and tri-sector partnerships

      My first job was at a summer camp in the mountains, where I worked as a camp counselor. The camp...
      Global Summit

      Equipping Civil Servants for the AI-Era: UNESCO and Oxford Launch...

      According to the World Bank the Public Sector employs one third of the global workforce. Yet fewer than half of...
      AI Skills coalition

      Are you sure you want to remove this speaker?