Artificial intelligence (AI) may not be ready to write the next blockbuster movie, but a team of AI researchers from the Massachusetts Institute of Technology’s (MIT) Media Lab successfully used machine learning to teach computers about emotional arcs in movies.
The researchers, which collaborated for this project with McKinsey, used machine learning to analyze thousands of videos, including movies, TV shows and short films found on Vimeo. “We developed machine-learning models that rely on deep neural networks to ‘watch’ small slices of video—movies, TV, and short online features—and estimate their positive or negative emotional content by the second,” the team wrote in a blog post Monday morning.
The approach didn’t just pay attention to the general plot line of a movie, but also to more subtle aspects, including the score, and close-ups of a person’s face. Using these clues, the project’s machine learning algorithms were able to identify positive and negative emotions, and map out the extend to which each scene would provoke emotional responses — something the researchers called “visual valence.”