The power of AI means that fully automated TV may well be drawing near. With the right materials and in the right context, machine learning might even take charge of content creation, explains Masstech CEO George Kilpatrick.

George, not Samuel

“When a man knows he is to be hanged … it concentrates his mind wonderfully”

Samuel Johnson (1709 – 1784)

Samuel Johnson lived in different times, but understood the human condition better than most. If he was here to see the onset of COVID he would not be surprised at the speed of change in industries reacting to the pandemic, driven by a need to survive. The media industry is no different.

Halving advertising budgets, an inability to produce new programs due to social distancing, and a complete curtailment of live events and sport has left channel operators desperate for content to fill emptying schedules. The good news is that there is plenty of archive content out there, coupled with a huge capacity for innovation.

Content management tools like Kumulate allow content owners today to fully utilise their archive – but what if we were to take this further to create full automated TV channels? The aim would be to create a very low-cost offering designed for social broadcast channels,  potentially serving niche markets with suitable free-to-air business models or even subscription.

The power of AI is such that this is now perfectly possible. Using sentiment analysis from multiple news and social feeds and advanced personalisation techniques,  we could automatically restore from archive to a schedule determined by algorithms to meet that users wishes, conscious or unconscious. When the content is not of a suitable format,  AI techniques are now available to upscale the content either before broadcast delivery or even within the watcher’s TV set. Samsung use AI to optimise the filter on their TVs and employ such techniques as detail creation and edge restoration. 

CRAFT explicitly predicts a temporal-layout of mentioned entities (characters and objects), retrieves spatio-temporal entity segments from a video database and fuses them to generate scene videos. Image © Allen Institute for AI. (https://prior.allenai.org/projects/craft)

You don’t have to go as far as training models to recreate new episodes of old favourites such as the Flintstones† – you just need to ensure sufficient metadata is associated with the assets and a suitable content management tool can use those hooks to automate workflows. OTT channels have already started to exploit the concept of micro personalisation, and can deliver content and targeted advertising either down to the individual or to a set of user categorisations that suit the broadcaster’s need. This is just one step further.

Clearly such low cost automated channels are only suitable in certain circumstances, but for operators desperate to replace missing revenues,  this might fill a need. Expect more experimentation like this as the pandemic concentrates our innovative minds, wonderfully.


† The Allen Institute for Artifical Intelligence, The University of Illinois Urbana-Champaign, and The University of Washington developed the AI, called Composition, Retrieval and Fusion Network (CRAFT). It was trained on a database of more than 25,000 Flintstones videos which had been painstakingly annotated.
https://prior.allenai.org/projects/craft


Find out more about how Kumulate incorporates AI and Machine Learning services into media workflows at www.masstech.com/kumulate, or email info@masstech.com for further information.


Follow by Email
Twitter
YouTube
LinkedIn