AI Video Tools Promise Much, but Can They Deliver?
Artificial intelligence (AI) is an important new technology for the telecom sector, drawing considerable investment and generating significant buzz in the past year. And it’s not just telecoms, it's going to affect almost every industry and sector that comes to mind.
The media and entertainment industry, particularly video, is going to be one of the sectors where AI is expected to grow rapidly. The transition to digital has resulted in huge quantities of data being generated by consumer interactions, as well as by traffic and network performance metrics.
It has also resulted in a lot of non-linear content being made available, which has multiplied the choices available to consumers exponentially. Helping consumers navigate those content choices and using the wealth of data to optimize consumer experiences is now a daunting task, because generating data is only a first step -- it then has to be analyzed to identify and implement the best possible approach. That is where AI comes in.
Let's take post-production as an example: Typically editors have to go through all of the footage, identify the best scenes and then put them together. If you are producing a documentary, that might mean hours and hours of footage to first review. But AI tools such as Microsoft Azure Cognitive Services can now go through the footage, generate metadata at a more granular level than ever before, and allow editors to hone in on what they want much faster.
Similarly, promotional cuts -- short trailers and promotions -- can be automatically generated by AI tools. That cuts down production and post-production time and --at least in theory -- costs. Dutch startup Media Distillery's Snackable Content tool automatically creates short clips based on a customer’s favorite person, topic or interest from a TV show or movie.
While that is more of an issue for broadcasters and content creators, AI has a number of video-related solutions for service providers as well. Perhaps the biggest challenges for service providers today is to maximize the value of expensive content they are licensing. When Netflix spends $7 billion a year on programming, it needs customers to view and appreciate the content. Therefore, navigation, search, discovery and recommendations have become extremely important to all video service providers.
AI and machine learning allow platforms to track and understand user habits and content consumption trends, and then serve viewers with the content that they are most likely to enjoy. The basis for this is to look at patterns and relationships across users to better profile individuals. So if people who liked Show A also liked Show B, and you liked Show A, AI platforms like Ooyala Discovery will recommend Show B to you.
This helps viewers find content that is of value to them, which reduces search frustration and increases the perceived value of the service. In turn, this will cut churn and boost word-of-mouth support.
Finnish start-up Valossa is another example of a video-centric AI solution. The company analyzes video content, generating descriptive tags, categories and overviews automatically. It then uses these to provide insight into the content, facilitate search and deliver targeted advertising. Valossa's cloud-based platform can recognize faces, visual and audio concepts (objects, voices, settings, etc.), spoken keywords and objectionable content (nudity, profanity, etc.) to better help search and recommendations.
The aforementioned Media Distillery also offers AI tools for content discovery and targeted advertising, among others. But for me, its most interesting features are the Binge Markers and EPG Correction tools. Binge Markers allow binge viewers to skip opening and closing credits and simply get on with the next episode. EPG Correction provides real-time adjustment when the EPG doesn’t correctly display the start and end times of a show. That's a huge issue for people who regularly watch shows time-shifted on DVRs.
Another key area that media-AI solutions are targeting is the localization of content, particularly as OTT takes shows global. In many countries, piracy is rampant, so if a new show is not included in the local legal streaming service within hours after airing in the US, it will be available on pirated sites before the legitimate ones. Most local streaming services want to localize the content, i.e., add subtitles or dub the dialogue. That usually takes a couple of days today, which could be too long given the high level of piracy in some regions. Again, automated tools using AI can turn this around considerably faster, and could be a huge help for streaming services in several Eastern European nations, for example, where this is a serious problem.
Targeted advertising has been a dream for service providers of all sorts for decades. The idea that highly relevant advertising could be delivered to each household is a great idea, but it’s been difficult to implement. Again, the ability of AI to sift through streams of user data and match advertised products to individual consumers could be a huge shift from the traditional "spray-and-pray" approach TV advertising is often accused of. This kind of attention to ad insertion also extends to concerns about placing advertising next to inappropriate content, which has been a concern particularly where user-generated content is concerned. For example, a baby care brand wouldn't want to be placed in a video with adult content, and using AI the brand can specify that can't happen.
Piracy is another area where AI could help, automatically identifying content placed on video sharing sites illegally and informing local authorities, site owners and the rights holders immediately.
And path optimization for content delivery is yet another area for AI, helping content providers monitor the state of their network and identify the best path to the end-user. This includes selecting the best network paths, the best caches and CDNs, the best bit-rates and video profiles. It also includes monitoring traffic levels on the network or at particular cells (on a cellular network) to deliver the best QoE to their subscribers.
Video-oriented AI solutions promise a lot, but all of this is at a very early stage. AI is far from the first revolutionary technology to hit the video industry, and it’s likely that a lot of evolution will be required before most of this gets widely deployed. But it has begun: particularly in the area of search and discovery, AI is increasingly being utilized today, and path optimization for content delivery has been in place for several years.
— Aditya Kishore, Practice Leader, Video Transformation, Telco Transformation