The Importance of Diversity in AI
Hi there!
Many have explored the link between corporate diversity and performance, notably including 2015 and 2017 studies by McKinsey. These studies, which looked at 366 and 1,000 public companies, respectively, across a range of countries and industries, found that gender, ethnic, and cultural diversity correlates to greater company profitability.
Now, I’m not aware of any formal studies exploring the diversity of data science teams and the performance of their models, but I’d expect to see similar results, particularly in cases where models must be robust to societal and dataset biases.
The importance of diversity in the machine learning and AI community came up in a recent conversation with Moustapha Cissé, a research scientist at Facebook AI Research.
According to him, “you are what you eat, and right now we’re feeding our models junk food.”
To address this, companies working in the field should strive for their teams to be as diverse as possible, “because only in this way will we notice the problems.”
More broadly, Cissé emphasizes that “it’s important that we become more open and more diverse so that every [community] has the tools and techniques required to solve its own problems.”
My interview with Moustapha kicked off a series of podcasts we debuted earlier this month called Black in AI. Many of the guests in this series are folks I met at the inaugural event by the same name at last year’s NIPS conference.
Moustapha helped organize Black in AI along with Timnit Gebru, and others, to support and increase the representation of blacks in machine learning and artificial intelligence.
Groups like Black in AI, Women in Machine Learning, and numerous others, several of which Moustapha mentioned in our conversation, are working to increase the diversity of the field as a whole, and for this are worthy of our support.
