How Media Companies Are Using Artificial Intelligence to Connect With Consumers

Media Companies Turning to Artifical Intellegence
Matthieu Bourel for Variety

Killer computers, robot uprisings: Hollywood has long had a deep fascination for artificial intelligence. Even off screen, AI is increasingly a key part of the media business — but thus far, the reality isn’t nearly as dramatic as movies like “Her” or “Ex Machina” make it out to be.

Case in point: You’ve probably been contacted by AI today without even knowing it. That push notification on your mobile phone, the email newsletter of your favorite website, or the videos recommended to you while binge-watching are being powered by machine-learning algorithms that rely on huge amounts of data to make smart decisions about the media you’d be inclined to consume.

One example for this is YouTube’s content recommendations. In its early days, the video-sharing website simply used something called collaborative filtering to tell its users what to watch. If 10 users all had selected the same five videos, then the site predicted that the next user who watched four of these videos would also like the fifth.

Over time, YouTube refined these recommendations, relying on more complex forecasting models, but that still failed to account for the billions of videos available on the service.

“YouTube is just so diverse,” explains Cristos Goodrow, the service’s VP of engineering for search and discovery. “We believe that for every human being on the planet, there are already 100 hours of videos on YouTube.”

The tricky part is to correctly identify those 100 hours. That’s why YouTube has embraced deep learning powered by neural networks — which essentially means that the company is using algorithms that simulate the way the human brain works. These algorithms are being trained to make informed decisions about which videos to recommend.

Goodrow developed similar neural networks for big enterprise clients 20 years ago. Back then, the technology was a bit hit or miss, he recalls. “Today, they suddenly work really well.”

The key difference is data, he explains. YouTube’s algorithms are being trained against hundreds of billions of data points of viewing behavior — something that simply wasn’t possible just a few years ago.

YouTube parent Google has identified AI as a key part of its business going forward, with Google CEO Sundar Pichai recently telling journalists at a press event that the shift to AI is as fundamental as the invention of the web or the smartphone.

“We are at a seminal moment in computing,” Pichai believes. “We are evolving from a mobile-first to an AI-first world.”

Alex Holub, CEO of San Francisco-based AI startup Vidora, agrees. “In the next three to five years, every business will have an AI engine at its center,” he says.

Vidora is helping media companies like News Corp., Yahoo Japan, and Walmart’s Vudu video service to kick-start their AI efforts. The startup’s technology allows these companies to customize their communications with consumers to reduce churn, be it on their websites, in apps, or via mobile push notifications.

AI-based systems like Vidora’s can learn not only which videos to recommend, but also how to market each video, and even find the best price and value proposition for any given customer. “Everybody is getting their own experience,” says Holub.

Key to this are, once again, vast amounts of data that help the system to learn from user behavior, then train the AI engines. “In the online world, you can get data unique to each user that you couldn’t get in the past,” Holub says.

In time, these systems will improve and be able to develop complex strategies, maintains the Vidora chief. For example, a video-subscription service could, in the not-too-distant future, let AI decide which movies to license.

Goodrow also banks on AI for YouTube’s next big challenge: understanding customer satisfaction. The service already knows which videos you like to watch, but it doesn’t really understand which of those you truly love. The site is experimenting with surveys to get feedback on individual videos. Over time, it may use those insights to train AI algorithms on the subtler signals
of satisfaction.

In other words: Instead of killer robots, AI may simply bring us killer content recommendations.