Artificial intelligence (AI) is affecting every business which has data and is changing the way they operate. TV is no exception. We already see AI start to change the way workflows operate at the backend. We’re now beginning to see exciting and innovative applications around discovering what viewers want to watch and how they can find this. AI is attracting hype, and there are some areas where we tend not to think that AI provides a useful solution (more about that in a future blog post). Here are three AI applications we are excited about.
Better content discovery with TV data
While a few customers will stay loyal to a particular channel, the desire for content they want to watch is what guides the viewing decisions of most. However, users can be overwhelmed with choice and end up viewing nothing, with a third of viewers complaining that they often cannot find anything worth watching. The average viewer says they want to spend 45 seconds deciding on starting to watch something, and this is a critical and tiny time-window where well-organized content and smart recommendations can direct a potential viewer to content they view.
AI can help viewers to rapidly discover the content they want by managing the data and metadata offering smart and personalized recommendations and brilliant search results that swiftly take them to what they are most likely to want to watch, regardless of their taste in genres. There are applications like this already, such as the now ubiquitous recommendation engines that MVPDs and operators use to surface content.
Where AI can add value is in offering a much more consistent approach to presenting and describing this data. Poorly organized metadata can be very visible to the baffled viewer, e.g., tagging a soccer game as drama, wasting the valuable 45-second customer window. Soccer may be full of high drama but meta-tagging it with this tag is simply incorrect from the perspective of TV genres. AI can help by being better and much faster at tagging accurately.
Meta tags are particularly useful in an AI-powered search. Metadata from third parties are notoriously disparate. They are often not standardized and can be used differently by different providers. For example, tagging soccer as football may mean World Cup matches appear in searches for viewers looking for American football but don’t appear in results for soccer. AI can be used to reduce these errors, by validating (and if necessary modifying) inconsistent meta tags.
Voice commands and TV
The integration of AI-powered assistants into TV services will provide a new level of data for the TV industry to use. OEMs like LG have integrated Google’s voice assistant into devices. We’re excited about the potential of these systems, and the data science that powers them, providing a much clearer picture of not just what people do when watching TV, but why.
Let’s imagine a family starts watching a new drama, but they start to get bored after ten minutes. They tell the remote to change the channel where they watch an old cop show for ten minutes. When the ads come on, they tell the Assistant to lower the volume, an indication they are watching the ads but responding to their tendency to be a little louder. We can insinuate that the family is watching the ads, but responding to the fact that ads are usually a little louder than the broadcast. This is a simple example of where voice assistants can add color to the data.
Maybe after the show, they search for another show to watch by asking for TV schedules or setting reminders for future programs. We know that more of this type of data will become available as the adoption of smart homes and smart cities technology becomes mainstream. The challenge for the industry will be to understand how to classify and harness this data and use the reporting and modelling to make better decisions
Producing content using data analytics
The use of any kind of algorithm to help to commission or produce great content is controversial, and one that strikes to the heart of how science can impact the artistic and creative side of the TV industry. We don’t see a future where robots are writing the latest episode of The Big Bang Theory, but we think AI can be used to help to create great TV content.
If writing for a comedy, a data science consulting team could build an AI application that could access an extensive database of previously written successful comedy scripts as well as another database of comedy pilot scripts for shows that failed. Perhaps all the successful shows averaged 10-15 jokes or funny moments per hour where the failures averaged either more or less than this. So whereas AI is not capable of writing jokes that we would find funny, it can help the human writer to structure the new script based on its insights into what has worked in the past.
AI could also help by suggesting characters and plots although it is the human that will have to convert these suggestions into a viable script. We already see some companies moving in this space. Vion Labs, for example, has built an AI system that analyzes in detail the content of a movie or TV, down to the level of the number of gunshots fired, or movie kisses. Any patters it derives can be used to understand what people like in a show, in a way that is more granular way than humans can themselves express. Adding these kinds of insights to the content production process can ultimately help content creators to make better shows.