Broadcasters should learn to walk before running with AI

Broadcasters mulling over bandwidth optimization, personalization, metadata creation, dynamic ad insertion, audience targeting, and more can do so by embracing AI, according to network experts over at the radiocommunications sector of the ITU (ITU-R), who have taken a stab at arguing the case for using artificial intelligence and machine learning to improve production workflows in the broadcast sector.

You read that right – broadcasters. Not telcos, or ISPs, or cable operators, or streaming pioneers. Unfortunately, the report’s conclusion comes over as a little confused and therefore perhaps it could’ve learned a thing or two from arguably more advanced sectors like mobile networks.

It’s a bold move from the ITU-R, we’ll give it that, considering a common theme already this year at Faultline Online Reporter has seen us discuss the complex transition to IP being carried out broadcasters which – surprisingly to some – is a process many broadcasters are either struggling with technically or approaching much too tentatively. It makes the ITU-R’s conclusion, from a new report published this week, that the broadcast industry is ready to inject AI into the workflow process a contentious one.

Let’s start by being succinct and flipping the 20-page report on its head. The ITU-R essentially concludes that while AI and machine learning algorithms can positively influence program and production workflow as well as audience experiences, the development of a training framework is the fundamental element needed to enable AI in broadcast workflows. This framework must be fed real task data to build applications with sufficient accuracy for broadcasting. So, while the report details a number of relevant applications and categorizes these into seven distinct topical descriptions of technological benefit, it’s this aforementioned framework we were eager to drill down into.

Rather inconveniently, the word “framework” appears only once in the entire Artificial Intelligence Systems for Program Production and Exchange report, found in the conclusion we just mentioned – making the job in hand a little trickier. We have contacted the ITU-R for clarification, but in the meantime, automated program content creation jumps out as a particularly intriguing topic, one of ten subsections under the report’s broader automated content creation umbrella.

The ITU-R cites a project from BBC Four in the UK last year, when it carried out two full days of programming scheduled entirely by AI algorithms developed to optimize accordingly to the user demographic. BBC Four, which we associate as the BBC’s guinea pig channel, used extensive data from the broadcaster’s archive to train an algorithm to create segments aired with content directly generated by AI. The report doesn’t actually say whether the project was a success or not and it doesn’t appear the BBC ever publicized anything resembling a full post-study report. What the report also doesn’t reference is that the BBC developed most of its algorithms in-house, following an exclusive conversation with Faultline Online Reporter, while we have seen similar projects use AWS Rekognition technology.

A BBC spokesperson told us, “I can confirm the AI technologies are a combination of bespoke in-house software and existing open source/free software – we always build on top of existing standards and systems where possible (although there are too many to list here).”

But has progress been made nearly a year on? A drawback from the BBC Four project is that it focused purely on viewer behavior from a single channel to form trends using AI, rather than expanding to other data-rich realms such as integrating social media activity or looking at viewer behavior on the BBC iPlayer, for example.

It just so happens social media is a prominent topic in this week’s AI report. In a section called Automated Video Digest, the ITU-R outlines how public viewer comments on social media regarding the program, performers and other program features, may be taken into account for content modification. For example, Japanese broadcaster NHK has developed image analysis technologies to identify characters and performers in a specific program which can be used to provide program previews to viewers, as shown in the figure below.

Automated video digest comes under the Workflow Optimization umbrella, which also includes compliance tracking and content creation, giving the example of how companies have created AI and machine learning-driven workflow implementations to improve compliance in production and delivery as mandated by the FCC. It cites live IP technology vendor TVU Networks as one company with a transcriber service deployed for use by Call-letter stations – integrating AI algorithms to detect the need for closed captioning in content and automatically transcribing absent dialog as closed captions. All-in-all this promises that all video content is FCC compliant prior to on-air broadcast or delivery through any other digital platform – avoiding potentially hefty fines.

Speaking of promises, we will continue to pressure the ITU-R for a response to our query about the somewhat ambiguous conclusion and provide an update as soon as we can.

In total, the report details 27 sections and thankfully doesn’t dwell on each too long. Clearly the potential for AI in broadcast is far-reaching – as with many industries – going well beyond basic workflows.

“Netflix, for instance, estimates that its use of AI to automate workflows and reduce customer churn saves the company around $1 billion annually,” states the report. If that doesn’t persuade broadcaster industry players to follow the streaming industry’s lead then nothing will.