Artificial intelligence is likely one of the most used buzzwords of our time. No matter whether you are in the automotive industry, healthcare, retail or something else, everyone is talking about AI. The media industry is no exception; from NAB to IBC, from Europe to the US, this phrase is resonating on every industry event’s floor.
We can argue for a long time whether AI will boost the economy or kill industries, how many new innovative businesses will emerge, how many people might become unemployed, how secure AI-based solutions might be, and so on. I’m pretty sure this could be a great topic for a live debate. But let’s leave aside these “philosophical” aspects and focus more on potential and capabilities.
Most media industry players already boast some kind of AI-based solution implementation within their business workflows. Netflix can be easily called a pioneer here because of its intelligent cast compilation and viewer data analytics1 (e.g., “House of Cards”); the sophisticated deep learning and computer vision algorithm that is applied to its recommendation engine, and which is far beyond the industry standard and is still evolving2; and all the way back to its video encoding, which analyzes each shot in a video and compresses it without affecting the image quality, thus reducing the amount of data it uses.3
Another example of AI in the media industry is when 20th Century Fox and IBM used Watson APIs and machine learning techniques to analyze hundreds of horror and thriller movie trailers. After learning what keeps audiences on the edge of their seats, the AI system suggested the top 10 best candidate moments for a trailer from the movie Morgan, which an IBM filmmaker then edited and arranged together4. And this happened almost two years ago already! Since then, 20th Century Fox use AI and deep learning models to predict which audience will most likely see a film based on the film’s movie trailer. They can accurately predict audience type and attendance for existing movies, as well as to-be-released movies.5
The same is true for other media and entertainment giants. Disney is working hard on mixed reality and augmented reality projects, robotics and human-computer interaction, computer vision, etc. On the AI front, Disney and the University of California used a deep learning approach to denoise Monte Carlo-rendered images, which produced high-quality results suitable for production. For the film “Finding Dory,” a special convolutional neural network was trained to learn the complex relationship between noisy and reference data across a large set of frames with varying distributed effects, and then produce noise-free image quality. Now it can be applied to other films, as well.6
Comcast uses machine learning models (among other solutions) to predict customer issues right before they occur. According to Adam Hertz, VP of Engineering at Comcast, their technology is 90% accurate in terms of predicting if a technician needs to drive to a subscriber’s home to fix a connectivity problem.7 Major tech vendors like Amazon, Microsoft, IBM, and of course Google are also working hard in this direction. Let’s take for example Microsoft and its Azure AI platform. From pre-built AI, to customizable ML and deep learning services and tools, you can find all you need to upgrade your solutions and services with cognitive capabilities, natural language processing, etc.
Even without thorough scientific research activities — and just by observing industry itself — it would be quite easy to spot the key areas where artificial intelligence is changing the media business.
Bill Gates’ phrase, “Content is King,” is relevant like never before. Industry incumbents, as well as new-gen video services, spend billions of dollars on original content.8 Of course, AI is not currently in a position to create content on its own, even though we do have such examples available. (In the movie “Zone Out,” AI was responsible for everything from script writing to video editing9). But intelligent technology solutions can help human directors produce pixel-perfect videos. AI can analyse your entire footage to select the best possible shots per specific needs: proper color scheme, right actor emotion, best place to cut or merge scenes, etc. Things that are not possible for the human eye to detect are simple to do for AI-based software. An engine that is capable of putting together the best scenes can create custom ads on the fly, more engaging movie trailers, etc.
If content is a king, then I would call user experience a queen. Apart from original programming, the way in which you interact with customers will define whether they will stick with your service or not. The most common way to do it right now is through targeted content and sophisticated recommendations. Which is good, but not enough. It is important not only what you recommend, but also how you recommend it. Good examples include custom pages or screen layouts, banners tailored based on your profile data, payment workflows specific to your habits and preferences, etc. Or imagine an intelligent ad system capable of not only serving relevant advertising based on content, but also defining a user’s relevant emotional state and the proper timing to insert this ad.
The following items could be easily defined as those that also affect user experience in the long run, but they still deserve separate category. We can start here with tagging and video indexing. It used to be quite a slow and labor-consuming process. Recent computer vision development progress can save you a lot of money with automated metadata extraction, while making your insights about footage deep and niche like never before.
The way that your video is streamed is also very important. And again, AI already contributes here, such as ensuring the best possible image quality while optimizing network usage, utilizing intelligent fault diagnostics during video delivery instead of manual alerts configuration, making your content accessible to international or hearing-impaired audiences by means of subtitling and captioning, and so on.
There’s one other factor that may bother you about AI: the cost of implementation. Depending on your goals, available basis, interactions with third-party tools, the complexity of specific workflows, etc., the cost of your project could in theory skyrocket. But in reality, purpose-built AI is inexpensive. The market is full of various tools, frameworks, libraries, and datasets that are ready to be leveraged. For example, Tensorflow, Keras, Microsoft Cognitive Toolkit, MXNet, Torch, Chainer are only some of the available open source frameworks for deep learning. Training datasets is not an obstacle for machine learning any more, either, through facial recognition, object detection and recognition, sound data (e.g., speech and music), text data, etc. Just take what you need!
In a nutshell, if you have specific, well-defined tasks that consist of repetitive and not creative work, then it might be a good case to consider the power of AI. The same is true if you have tons of data and use more than a few people to “play” with it.