April 27, 2017
At ETC’s conference on machine learning/AI at NAB, director of data and analytics Yves Bergquist talked about the work ETC@USC is doing to understand AI, storygraphics and audience intelligence. At the heart of the question, he said, is why we like or don’t like a movie or TV show. Getting an audience member to describe why she liked her favorite movie, he responded that the people who made that movie don’t know why she liked it. “Not because they’re stupid, but because it is a very complex, multi-faceted question,” he said.
Bergquist pointed to today’s challenge rooted in the “exponential complexity across every market and industry” as well as “high velocity change.”
“Media and entertainment has a unique challenge,” he said. “Your customers know a whole lot more about your product than you know about them.” The solution is three-fold: think of audiences as systems; develop and apply more data, intelligence and flexibility; and have a space to test, experiment and fail.
With regard to a new method to approaching audiences, Bergquist noted the need to rely on sub-systems of influencers and social relations, channels and platform behavior and an ecosystem of stories. He noted that in yesterday’s top-down broadcast model there was a very high signal-to-noise ratio because there wasn’t a lot of content.
“Now we have a very low signal-to-noise ratio, and curators help us extract the signal from the noise,” he said.
“Entertainment is a social contract, with the delivery of emotional states and algorithmic stories,” he said. “What drives my decision to play videogames or go to the movies is about whether the content is a fit for me.” More specifically, he said, stories help us boil down the enormous complexity of data into a cognitive script we use in the world. He pointed out that Bill O’Reilly and Jon Stewart are examples of media personalities that have delivered distinct narratives.
“That’s the power of narrative curation,” said Berguist. “Their story is extremely strong, structured and distinct. They boiled down the complexity of the world to fit what people want.”
Bergquist went over a storygraphics research project that considered 72 variables, including scene-level story and box office, for 300 films. Though the sample was small, it revealed several obvious things, proving the methodology was sound. “We found that story and character mechanics best exploited visually do very well, and internal states perform really poorly in films,” he said. “We also found that characters (who can be male or female) that are more popular with males do 25 percent better at the box office.”
“We are honing in on a grand theory of what makes things interesting,” he continued. “If you’re in a familiar environment, you don’t need much data, but if it’s not familiar, your brain needs more information. There’s a perfect ratio of expected and unexpected elements.”
Bergquist introduced his company’s AI-enabled tool Corto that allows the user to ask questions; Corto and a dozen AI plug-ins then “develop deep insights from complex data.” “We’re trying to go from Big Data to Big Knowledge,” he said. “This is a semantic organization of data, and I believe we’re headed to a semantic future, with an ecosystem that really boosts the signal-to-noise ratio.”