November 22, 2017
The Dawn of AI in Broadcast and Media
Exploring current and near-future use cases of Artificial Intelligence in Broadcast and Mediaby Francisco Fronda, MediaPower Marketing Director
These days, we can’t get around technology news without hearing about bots, AI, or machine learning. Every time I read up on industry news and technology, AI is always a pervasive subject. A good number of the articles we share over our social media accounts are on the topic of AI. But compared to other industries, there seem to be fewer articles that focus on AI's application to TV & Broadcast, or how it delivers unquestionable and immediately realizable benefits. So, I did some further reading to see how AI is progressing in Broadcast & Media.
AI in TV used to be only seen as part of fictional TV programs or documentary shows. Obviously, this is no longer the case. However, it is not easy to determine which segments of the Broadcast and Media industry get the most significant impact from developments in AI. We are also still far off from applying AI to create programs that are consumable by regular audiences automatically. But we're not being left behind. We see AI creeping into many aspects of TV and media through some mainstream applications and in several not-so-distant future use cases.
Content Discovery and Personalization
While there are many general use cases for AI in content creation and delivery, the two most viewer-centric applications would be on content discovery and content personalization. The aim is to deliver personalized and targeted experiences to multiscreen audiences to keep them coming back. Whether done through big data analysis or in combination with machine learning methods, the underlying objective is the same - that is to display curated content based on user information, preferences, watch history, and context. One such example would be Netflix. Netflix has long been making viewing recommendations based on previously viewed or searched videos. Netflix algorithms are sophisticated and complex. Over time, they have made huge strides in adding AI into the mix. Their recommendation engine 1
already employs machine learning to deliver user recommendations.
Ad Targeting / Filtering
The same also applies to ad targeting. Advertisements can also be tailored to the audience - based on their demographics or preferences. YouTube delivers ads to viewers based on information that is known about the viewer and selected based on advertiser target selection. YouTube also uses AI to ensure content is "safe" and not offensive. In fact, just recently, YouTube has intensified its efforts with the help of AI 2
to filter ads and protect the brands of advertisers. This came after a major backlash a few months ago when advertisers pulled out their ads after they appeared in offensive content.
Content Clustering and Metadata Building
Beyond a simple watch history or audience analysis, there is also the promise image recognition for audience analysis as a potential mainstream application. On content, recognizing images from live or recorded videos can lead to metadata tags or descriptive information of a movie or a program beyond its usual essential attributes (genre, language, actors, etc.). Such additional metadata information can then be used to cluster content by context or also as a basis for content search and selection. This automatic metadata building from image recognition is also what MediaPower's own M.ART.IN. (MediaPower Artificial Intelligence) upcoming offering aims to do - that is to use machine learning to carry out image recognition and automatically convert recognized images to metadata.
Voice-assisted Content Search & Playback
With growing prevalence of voice-assistants like Amazon's Alexa, Apple's Siri, Google Assistant, Microsoft's Cortana, Samsung's Bixby, it is only a matter of time before voice assistants are "fully connected" to TV sets and other viewing devices to deliver voice-assisted content search and playback. Pretty soon, you'll just be voicing out context-sensitive commands to play a particular movie title, or search for movies released within a specific time frame, with your favorite actor in a starring role, and much more.
Content Curation through Audience or Scene Recognition
Another possible use case under image recognition would be facial or scene recognition. A camera on your viewing device (TV, computer, or mobile device) can open the possibility of audience recognition. For example, through the use of the camera, the system can determine whether a child is part of the audience and ensure content shown is within the bounds of appropriate TV ratings; or determine the audience’s age and gender to tailor content further, according to pre-built assumptions or carefully analyzed audience data.
Content Production through Cognitive Analysis
Further beyond image recognition for audience analysis, there is cognitive analysis - that which looks deeply into individual viewer consumption habits or even emotions, and then predict intent, tailor content selection, or even produce unique real-time content. One such example of cognitive analysis in the works is an AI generated trailer 3
last year for an upcoming movie Morgan (an IBM creative project through its primary AI platform Watson). Using machine learning, Watson was trained to recognize scenes from trailers of hundreds of horror movies, and then shown the full-length film for it to create its own trailer.
There is also network optimization - analyzing available bandwidth and delivering the best balance between video quality and streaming bandwidth. Network optimization can include real-time optimization of codecs, network paths and data transfers, all to minimize the amount of data and time to reach a viewer while ensuring optimum video quality. Again, as an example, Netflix is already doing this 4
- it uses AI and image recognition to analyze scenes on videos and compress them without affecting video quality - reducing the amount of data that needs to be streamed.
Other Use Cases
There are also other smaller use cases for AI that can improve user experiences such as on-the-fly captioning through dialogue analysis, scene analysis to determine logical and least disruptive points for inserting ads, or AI-powered QC to check for black frames or color bars and automate actions. While these don't seem too huge compared to other use cases, they all could contribute to better viewing experiences.
While a major AI disruption in TV and Broadcasting still seems to be far off, we see more evidence of its application and benefits to many aspects of the way TV and media content is produced and delivered. The list of applications I cited is by no means exhaustive or definitive. But one thing is certain. We will see a growing adoption of AI-powered solutions in different aspects of TV and broadcasting - from content creation to management, delivery, and consumption. The possibilities are exciting.References:
(1) What Netflix teaches us about using AI to create amazing customer experiences
(2) Google Is Using Artificial Intelligence to Make Sure YouTube Content Is Safe for Brands
(3) IBM Research Takes Watson to Hollywood with the First “Cognitive Movie Trailer”
(4) Netflix’s new AI tweaks each scene individually to make video look good even on slow internet