Lightning talks: Advances in fairness in AI: From research to practice

Опубликовано 8 февраля 2022, 16:42

Over the past few years, we’ve seen that artificial intelligence (AI) and machine learning (ML) provide us with new opportunities, but they also raise new challenges. Most notably, these challenges have highlighted the various ways in which AI systems can promote unfairness or reinforce existing societal stereotypes. While we can often spot fairness-related harms in AI systems when we see them, there’s no one-size-fits-all definition of fairness that applies to all AI systems in all contexts. Additionally, there are many reasons why AI systems can behave unfairly. In this session, we explain the diversity of work taking place on fairness in AI systems at Microsoft and in the broader community. We highlight how we’re applying fairness principles in real-world AI systems by measuring and mitigating different kinds of fairness-related harms in vision, speech-to-text, and natural language systems.

Introduction
Speaker: Amit Sharma, Senior Researcher, Microsoft Research India

Fairness in speech-to-text
Speakers:
Michael Amoako, RAIL Program Manager - Quality of Service Fairness Lead, Microsoft
Kristen Laird, Program Manager, Microsoft Cognitive Services Responsible AI

Representational harms in image tagging
Speaker: Solon Barocas, Principal Researcher, Microsoft Research NYC

SAVII: Measuring fairness-related harms in NL services
Speaker: Chad Atalla, Applied Scientist & Tech Lead, MSAI Responsible AI VTeam