Building User Trust in Recommendations via Fairness and Explanations

Explore techniques for building user trust in recommender systems through fairness and explanations in this 27-minute conference talk from UMAP'20. Delve into the ethical implications of AI-driven decision-making, examining fairness aspects such as non-discrimination, diversity awareness, and bias elimination. Learn about explanation approaches that provide human-understandable interpretations of complex systems. Discover how fairness and explanations can work together to promote trust, considering the perspectives of various stakeholders with different technical backgrounds. Gain insights into the taxonomy of fairness in recommendations, white box vs. black-box models, local proxy models, and counterfactual explanations. Understand the importance of fairness-aware explanations and their role in building trustworthy AI systems that impact everyday life decisions.

Intro
Trust in Al Systems
How do people feel about Al systems?
Ethical principles in Al systems
What is Fairness?
Let's break it down
For Whom?
When is the distribution fair?
Of What?
Fairness of Utility
Fairness of Exposure
Taxonomy of Fairness in Recommendations
Explaining Recommendations
Why Explain?
White Box vs Black-Box Models
Local Proxy Models
Counterfactual Explanations
Towards Fairness-Aware Explanations
Take Away