Meet Gabi Zijderveld
Ethics and Privacy in Machine Learning
When collecting the large amount of personal user data required to make these kinds of machine learning solutions work, ethics and privacy become an important topic of conversation.
First, Gabi explains that Affectiva’s solution is not facial recognition technology, because it does not identify or authenticate the individual. Additionally, she shares that the company has made a purposeful decision that they do not want their technology deployed where there is no opt-in or consent. While they’ve been asked by security and intelligence government agencies who are interested in the technology for surveillance and security purposes, they pass on those opportunities because they do not think it’s an ethical use of their technology.
But what if the technology could do a lot of good without giving users the ability to opt in — to keep airports and retail stores secure, for example? The issue for Gabi is where to draw the line.
“We've seen things go horribly wrong with some facial recognition systems, especially in terms of bias,” she says. “I think that's a huge concern, where these systems are supposed to accurately recognize people regardless of appearance, age, gender, or ethnicity, and we've seen plenty of things in the news where some systems have failed miserably.
“Those are basically lousy design decisions. In terms of machine learning, it's a bad sampling of data. It's not focusing on data that are representative of the use case. It's about how the technology is being used and who your users are. These are things that, if you really focus on them and prioritize them, can be avoided, because the AI algorithms are trained with data. If suddenly your facial recognition system cannot identify black women, then you'd better go back to your data set and make sure you have data that represents people in all populations and focus on retraining your algorithms.
“So in our company ethos, we believe that there needs to be a focus on ethical development of AI, and ethical employment of AI.”
How is the COVID-19 Pandemic Affecting Emotion AI?
How is technology that analyzes people’s faces affected by large populations now wearing masks to prevent the spread of COVID-19?
Gabi says that this is actually something Affectiva was thinking about pre-pandemic because they do automotive business in countries like Japan, where people commonly wear face masks. If you want to detect a smile, of course, tracking the mouth is very helpful, but Affectiva also analyzes the entire face. There’s a lot of information they can access from the eye and forehead region. This approach necessitates having massive amounts of data, which is one of Affectiva’s strengths —they’ve analyzed over nine million faces in 87 countries, including specific studies of people wearing face masks.
To learn more about the story of how and why Affectiva was founded, Gabi encouraged listeners and readers to explore “Girl Decoded,” a book by Affectiva CEO and co-founder Rana el Kaliouby released in April 2020.
“It is a story of self-discovery and perseverance, and people who are generally interested in technology, innovation, or leadership will find it interesting as well,” Gabi shares.
To learn more or order the book, click here.