(Subscribe to our Today's Cache newsletter for a quick snapshot of top 5 tech stories. Click here to subscribe for free.)
What can an automated computer analysis say about Bollywood? To begin with: its idea of beauty has been consistent over the past seven decades, according to a team a researchers at Carnegie Mellon University (CMU).
The $2.1 billion film industry continues to associate beauty with fair skin, making it one of the many social biases Bollywood has remained consistent with, the team at CMU’s Language Technology Institute said.
The team collected 100 Bollywood movies spanning 70 years, and juxtaposed them against 100 top-grossing Hollywood films made during the same period. They also analysed subtitles of 1,400 films for gender and social biases in their statistical language models. And finally used artificial intelligence (AI) to make the observation.
They used a fill-in-the-blank exercise to assess specific statements like “A beautiful woman should have _____ skin”. The model predicted ‘fair’ after being trained on Bollywood subtitles data. The bias was less pronounced when the AI model was used on Hollywood subtitles, the team noted.
The CMU team used the model to also understand gender diversity in the films. They used ‘Male Pronoun Ratio (MPR)’ metric to compare occurrence of male pronouns with total occurrences of male and female pronouns. Right from 1950 to now, the rate for both Bollywood and Hollywood movies ranged from 60 to 65 MPR, the team noted.
The AI-powered method is said to be able to analyse 2,000 movies in a matter of days, compared with manual cultural studies that can consider up to 10 movies at a time, according to the team.
The tool could also be used to rapidly analyse hundreds of books, magazine articles and social media posts to understand biases that may occur in them, the team said.