New Study Shows Increase in Diversity in Hollywood

The recent study, conducted by researchers at UC Berkeley, used AI technology to reach its conclusions. 

In an inventive effort to analyze diversity in Hollywood, researchers at UC Berkeley have turned to artificial intelligence (AI) to record how often actors are presented on screen in over 2,300 films. This AI-driven study, led by David Bamman, an associate professor at UC Berkeley’s School of Information, offers an expansive look into Hollywood’s evolving representation, spanning over four decades of films from 1980 to 2022.

Using facial recognition technology, the team evaluated the top 50 box-office earners each year, categorizing them as “popular” movies, and films nominated for “Best Picture” by, at minimum, one of six major organizations like the Academy Awards, Golden Globes, categorizing them as “prestige” movies. The findings revealed a significant increase in diversity since 2010, showing more screen time for actors who are female, Black, Hispanic/Latino, East Asian, and South Asian. The study indicates that progress isn’t limited to ensemble-driven projects like Black Panther, but also in Hollywood overall and individual films. 

This advanced analysis became possible thanks to recent changes in copyright law. Historically, the Digital Millennium Copyright Act of 1998 restricted researchers from bypassing digital controls on DVDs made enforcing against piracy to study them with advanced tools. But in 2021, researchers helped secure an exemption that permits higher education institutions to decrypt the digital controls for the purpose of exhaustive academic research. This change enabled Bamman’s team to purchase and analyze a collection of 2,307 DVDs for their research, furthering data-intensive studies in film analysis.

Despite its transformative potential, the researchers emphasized that their automated approach complements, rather than replaces, traditional film analysis. Manual viewing allows for richer, more nuanced analysis of leading roles and dialogue, but such AI technology lets scholars examine a far greater range of films and non-leading roles from a quantitative point of view, leading to more precise measurable results.

The findings indicate an upward trend in the representation of women and minority groups on screen. For instance, the research discovered that from 1980 to 2010, the screen time of females was consistently around 25%, but by 2022, this figure had climbed to about 40%. The research additionally identified gaps in representation, revealing that diversity remains less prevalent in lead roles compared to supporting ones, and that Black actors are less represented in award-nominated films, particularly in works produced between 1980 and 2010.

In terms of methodology, the researchers referenced Wikidata and performed customer surveys to assess the nuanced perceptions of race, gender, and ethnicity and align with viewer perceptions. Bamman explained that focusing on perceptions offers insights into how representation resonates with audiences, rather than making assumptions about an actor’s identity.

Bamman’s study doesn’t just stop at tracking representation; it also makes reproducibility a priority. The research team has made public its non-copyrighted portions of data, such as metadata on each actor’s screen time and frame location. This openness, according to Bamman, is intended to encourage further studies into the cultural dynamics of Hollywood films. He expressed hope that the data will fuel more nuanced inquiries, including how actors are portrayed on screen and how these portrayals intersect with stereotypes and biases.

Beyond the mere presence of diverse actors, the study hopes to examine how portrayals might perpetuate or challenge cultural stereotypes. This project points to the potential for AI to aid in tracking cultural trends on a larger scale than previously possible, as well as possible future collaborations with scholars and studios.