Published Oct 8, 2020

Joaquin Candela — Definitions of Fairness

Joaquin Quiñonero Candela, a key figure in responsible AI at Facebook, delves into the complexities of establishing fairness in AI, sharing strategies for stakeholder engagement, deployment challenges, and the imperative of technical governance and diversity to foster accountable systems.
Episode Highlights
Gradient Dissent - A Machine Learning Podcast logo

Popular Clips

Episode Highlights

  • Defining Fairness

    discusses the complex nature of defining fairness in AI, highlighting that there are multiple, often conflicting, definitions. He explains that fairness can be interpreted differently based on cultural, legal, and philosophical contexts, making it a challenging concept to standardize 1. In the context of AI applications, such as content moderation in India, fairness involves ensuring that all groups receive equal protection against harmful content, despite the absence of data on sensitive attributes like caste or religion 2. Joaquin emphasizes the importance of involving local teams and policy experts to navigate these complexities, as fairness is deeply intertwined with societal norms and values 3.

    Fairness is a bit of a social construct in a way. It depends a lot on context, and it depends a lot on how a particular society has decided to govern itself.

    ---

       

    Technical Governance

    In exploring AI governance, underscores the necessity of frameworks that incorporate diverse perspectives and external advisory boards. He argues that responsible AI is not solely an AI problem but requires collaboration with ethicists, lawyers, and local communities to address fairness and accountability 4. Joaquin highlights the importance of transparency and accountability, suggesting that open sourcing and public consultations can help improve AI systems by inviting external feedback 5. He also points out the challenges in governance, noting that technology companies must integrate public deliberation mechanisms to ensure fair decision-making processes 6.

    Transparency implies accountability in a way. If I declare to the world that I have methods and processes for testing my stuff, then it better be that I have them.

    ---

       

    Promoting Diversity

    emphasizes the critical role of diversity and inclusion in AI teams to foster fairer AI systems. He shares insights on building inclusive teams, stressing that diversity encompasses both innate and chosen characteristics, which contribute to better decision-making 7. Joaquin warns against tokenism and highlights the importance of equal opportunities and support for all team members, advocating for strategies like the Rooney Rule to promote diversity in hiring 8. He also discusses inclusive hiring practices, such as writing inclusive job descriptions and ensuring diverse candidate slates, to break barriers in recruitment 9.

    Diversity really means heterogeneity. It means having a team that has people from different, both innate but also chosen characteristics.

    ---