AI Is Not Just A Differentiator In Media - It's A NecessityAI Is Not Just A Differentiator In Media - It's A Necessity
AI Is Not Just A Differentiator In Media - It's A Necessity
December 3, 2018
NEW YORK - Media as we know it has changed irrevocably in the last ten years - and with it, the world. Mass-marketed, focus group-tested visual and print media is a thing of the past. Today, all media consumers are also producers, generating countless volumes of data which reflexively inform the way media is then delivered and consumed.
Always-on connectivity has disrupted businesses, political processes, and the way we live our lives. As media continues to decentralize across myriad platform services, spurred on by technological change, it's up to the world's media giants to either innovate or fall behind.
Founded in 2015, NBC Media Labs is the internal innovation team within NBCUniversal - and was created with a specific mandate to anticipate, incubate, and implement new technologies across the network's myriad business units. In other words, it is designed to deliver technological innovation necessary to remain competitive.
To better understand what disruptive technology can achieve for media brands, we caught up with Media Labs' Vice President, Sowmya Gottipati. Sowmya and her team work closely with various businesses across NBCU to apply technology to key challenges - with AI and machine learning, she says, playing a pivotal role.
Why are emerging technologies so important to media brands?
Speed is of the essence in media. Take breaking news. Today, our TV team at NBC are really focused on live broadcasts operating around particular news segments. You already have the content—you aired it live on TV—but to take that content from the back-end and put it on a digital platform like Twitter or on a mobile app can take an hour or two.
Guess what? That doesn’t work in this day and age. You can’t publish breaking news two hours after the fact. That’s where we think machine learning technologies can help. With the proof-of-concepts we’ve created, the machine can ingest and collect video metadata in real-time and make it available for our editor, who just needs to push the content out onto the digital platforms of their choice. You can literally make this happen within minutes after it’s live on TV.
This is not only a differentiator—it’s necessary in this day and age. You should be able to create just one piece of content and have it appear everywhere. This speed and efficiency is what I think machine learning can make possible.
"Some people seem to think AI is a magic wand you can just apply and everything is covered. As we know, that’s not the case."
How do you achieve that?
There are many different ways one can apply machine learning algorithms. One of the things my team has been focused on is applying AI and ML to video. This is a unique field within itself. Today, we have minimal metadata around the videos we deliver to viewers—episode number, season, the list of actors, and maybe a two-line summary of what the episode is all about.”
With most of the audience now engaging with us on the new digital platforms, this is no longer enough. We all search for content in a way similar to how we speak. Unless you have a deeper level of metadata, it’s hard to separate and search for the videos in this way. So that’s where we’re focused: understanding the video at a frame-by-frame level so we can generate keywords for the important topics, characters, faces, objects, and logos that appear at any given moment. We want to then separate that in-video metadata for different applications, such as ads.
So what are the challenges of deploying that much metadata at scale?
An algorithm might produce 4000 datapoints within a video that you wouldn’t know what to do with. We have to therefore work with content owners to understand what actually matters for their use case. Every use case is different, so you have to apply your own internal knowledge, taxonomy, and algorithms on top of the metadata, and then extract maybe 100 datapoints that actually matter. I think that’s where it’s critical to have machine learning expertise on your team so you can then develop custom models for your own needs.
What are the key challenges in terms of getting the business teams to grips with AI?
AI and machine learning are very new technologies. Some people seem to think AI is a magic wand you can just apply and everything is covered. As we know, that’s not the case. One of the challenges we encounter in the beginning of working with the business teams is to explore what AI and machine learning actually means—what you can and cannot do with it.
This includes an explanation of accuracy levels. You need to educate enterprises and explain that, hey, you’re not going to get 100% accuracy on day one. It just won’t happen. It’s a model that constantly learns. There’s work involved—it’s not as if the algorithm can instantly process the video and give you beautiful results.
Another issue is that business teams often don’t always know what is meaningful for them. They may understand that AI can provide them with some interesting data, but they might not know why how that data can be leveraged to solve a business problem or enhance consumer experience. This is really a hand-in-hand process, in which you have to work in partnership with the business team [to explain] the technologies and their limitations, and also help them arrive at use cases with which they can actually solve a business problem. You have to be there for that journey.
This interview has been edited for brevity and clarity. As told to Ciaran Daly.
Sowmya will be joining 300+ senior enterprise decisionmakers and CxOs speaking at The AI Summit New York, December 5-6 at the Javits Center. Find out more