Geena Davis Institute’s Post

Transcript

(upbeat music) (computer keys clicking) - Yes, Scully was a badass. She was levelheaded, she used logic, she used science, and she certainly could save herself. - Dana Scully from "The X-Files" was breaking stereotypes, and I thought of her to be my hero. And then later on, I was exposed to this study called "The Scully Effect," which basically revealed that 63% of women working in STEM identified Scully as a role model while they were growing up. I am a byproduct of the Scully effect. And I think that's the power that characters in media hold. (upbeat music continues) (upbeat music continues) (computer keys clicking) - For 20 years the institute has looked at onscreen representation, and what we've always done is we've taken the data directly to the leading content creators, studio by studio, network by network, company by company. When people are presented with the data, it really has moved them to make change. - In the past few years, Google Research has developed Project MUSE, which stands for Media Understanding for Social Exploration. It is basically a project that is aimed at using AI to study representation in media. (machine clicking) - Media is composed of what we call modalities: the visuals, the sound design, and the language. What AI does is try to automatically understand how people are being portrayed along these various dimensions. - [Komal] So it can be computer vision systems that can infer the perceived age, perceived gender, perceived skin tone, to natural language processing models. - These are all perception attributes, and we need scales that are inclusive. For example, we often have scales like light or dark, which is not enough variety of skin tones that people have. So the Monk Skin Tone Scale was developed in collaboration with Dr. Ellis Monk This 10-point scale captured diverse types of skin tones, which helps us understand the media content better. - So we were really excited to partner with Google Research on the American TV study looking longitudinally over 12 years. - The scale of the study was huge because it was looking at top 10 broadcast TV shows of every year. The data we looked at was about 480 episodes. That accounted to about 440 hours of data. We were able to process this in under 24 hours. - And that's something that would not be possible with using human annotators, which has been the traditional way of doing such studies. - So what we found is, even though there's been progress over 12 years, male characters still had much more screen and speaking time than female characters. (machine clicking) - After we conducted the American TV show study, we partnered with the Geena Davis Institute and with the University of Southern California again to study representation in Indian television because of the usage of multiple languages, the skin tone diversity, the cultural nuance. - Part of this work relies on a lot of technological advances made by Google Research. For example, the Universal Speech Models allow us to automatically transcribe speech, not just in English, but different Indian languages we used in the India study. And Monk Skin Tone Scale helped us measure skin tone on a much granular scale. - It was really interesting to see that a country where there's so much skin tone diversity, even now, more than 80% of screen time is given to characters with lighter skin tone. - Google Research has always been a great and supportive partner, and the insights that they've been able to help us garner really pushes the research much further. (upbeat music) - What technologies are good at doing right now is understanding these individual modalities well. And once we start understanding these modalities together, we can then start understanding the stories being told computationally, ask questions like: When does this person show agency? - I grew up in India, where I was very impressionable by the media that I consumed. A lot of these roles were not played by people who looked like me. And then, to be at a place in my professional life to be able to do something about it was really a full-circle moment for me. (upbeat music continues) (air whooshing) (upbeat music continues)