AI: Feeding the virtuous circle

Cable Satellite International

Feeding the virtuous circle Artificial Intelligence for media is still in its infancy but already has a critical role to play helping broadcasters leverage broadband/ broadcast convergence across multiple workflow components.

p38 http://www.csimagazine.com/eblast/Digital_Editions/September_v2/CSI_September_2017.pdf

Artificial Intelligence will contribute as much as $15.7 trillion to the world economy by 2030, according to PwC. Global GDP, which stood at about $74 trillion in 2015, will be 14 percent higher in 2030 as a result, according to its projections. $6.6 trillion of this would come from increased productivity as businesses automate processes and augment their labour forces with new AI technology, states the report.

According to CB Insights, 658 funding deals around AI took place in 2016, worth over $5 billion in VC funds. In the last five years, 1,928 AI investments have accounted for over $12.4 billion as the development permeates all aspects of technology. The television/video industry is no different, with more and more back-end and front-end powered AI solutions slowly making their way onto the market, from recommendations systems and video analytics, to security and archiving. Eventually, even applications like playout will be affected.

However, much of what is suddenly branded as AI in media is in fact the evolution of the data analysis solution market. This is taking different dimensions. 

First, the use of unstructured and structured data, coming from multiple sources, combined with configurable data science algorithms, allows for the development of powerful prediction engines that is galvanising the data analytics and business intelligence market. New knowledge graphs and correlations can be found and refined in real time by data scientists.

“This is clearly where most value can be delivered in the payTV industry in the short term, as the players face a business transformation challenge and need to drive their business using more relevant data that they often don’t have, as legacy TV systems are not always two-way connected,” explains Simon Trudelle, senior director, product marketing, Nagra. “Social platforms and other external systems can offer extensive sources of data intelligence that can benefit service providers and other players in the TV/video value chain.”

A second dimension to consider is the development of machine learning predictive algorithms that use a feedback loop to improve the relevance of the prediction engine over time. 

Then there are AI/ML platforms that go further and leverage neural networks (using dedicated chipsets) and have been proven to outperform human beings for repetitive skill-based tasks such as speech or image/video recognition for instance.

For example, AI can play an important role in improving accuracy and reliability for placement, believes Michael Atkin, BroadView Software’s president. “BroadView’s latest Promo Campaign Management tool learns from previous passes and works to deliver improved placement results with each pass. Scheduling at both planning and presentation level can be more automated and done in a manner that improves operations by learning from past runs.”

Mark Mulready, senior director, Cyber Services & Investigations at Irdeto, is focused on the role of AI in combatting content redistribution piracy. “Combating redistribution piracy requires the proactive search and identification of illegal re-broadcasts,” he says. “AI can play a key role by enabling detection of illegal streams, through semantic analysis of social media ads and/or web page indexes, and by enabling inspection of visual elements in the re-distributed content, matching it to the original source.”

AI enables Irdeto to automatically process streams distributed by pirate aggregation sites or other distribution media, and recognise the original source of the video stream by identifying the broadcaster logo. The key benefit is the scale at which the system can operate. In January 2017, its crawling platform analysed over 700,000 unique URLs where pirated content streams were embedded on the internet. “A manual approach cannot scale for this type of volume,” says Mulready. “After all, speed is the name of the game especially when it comes to fighting live sports piracy.”

In terms of cybersecurity and the threat of malicious attack, if the system is monitoring all the logs and knows what ‘normal’ looks like, then unusual internet traffic coming from a new device on the network could be flagged as it happens, enabling early intervention with preventative measures.

“Rather than just providing intrusion detection in the network layer, our automation application software might become aware of where control commands are expected to come from,” says Pebble Beach Systems’ CTO Ian Cockett.

Another important area is in network management. Bandwidth is not limitless and is a significant portion of the expense of operating (especially for OTT). “Today’s CDNs and protocols are very wasteful of bandwidth (your device will always pick the highest available bandwidth whether it’s needed or not at that particular time,” says Tim Child, CCO and co-founder, Cantemo. “Analysing the video as it is transmitted and then using the data to determine the required bandwidth can help network operators make more efficient use of bandwidth, lowering costs and improving quality at the same time.”

Metadata the key If metadata describing the content as well as the precise timing of the start and end of each program were incorporated into the broadcast, it would be possible to use this to record content for later viewing or to start programs again both from the actual start of the content rather than when the EPG says they should start. “Most systems today provide DVR and start again capability but they are not as accurate as they could be and if the schedule changes at the last minute it doesn’t work at all,” says Cantemo’s Child.

“Signalling metadata in band at the time of transmission solves this and allows AI systems to ‘know’ the what and when so they can match with the user’s preferences/ circumstances automatically and accurately. This same metadata can also being married with rights information for the content so that it can be intelligently and automatically be substituted downstream.”

Many applications are using AI to deliver great results faster and cheaper – image processing, real time translation and speech recognition to name a few. Their success is based on training computers to process a huge amount of data in order to recognise patterns and derive meaning from the patterns.

“The first thing is to get the metadata right,” stresses Alan Young, COO at network management software developer Crystal. “All too often metadata is a manual afterthought. Automating the production of accurate metadata describing the content fully including timing for live/linear content is going to be critical for downstream AI/ML systems to be able to operate properly.”

Its Crystal Connect product automatically extracts and formats metadata so that it can be delivered to and used by AI engines downstream.

This metadata could be used by an AI engine to provide people with better search capabilities. It could also be used for contextual advertising by communicating exactly what is coming up when. “When matched with who is watching and where they are at the time this cannily make advertising more relevant and therefore more valuable,” explains Young.

There is a further discussion to be had on dynamic ad insertion and the delivery of customised commercials from the linear playout stream. Pebble Beach says its systems can help some of these legacy linear systems better compete with online services by making it easy to add SCTE signalling for downstream dynamic ad insertion, for example. “However, given that broadcast is still playing catch up and trying to implement virtualisation many years after the IT sector embraced it, the effective implementation of AI across the playout space may yet be some way off,” says Crockett.

Training datasets Netgem makes a similar point when applied to recommendations. Taking an OTT provider like Netflix, for example, there are two main limitations to AI, according to MD Sylvain Thevenot. Such providers limit recommendations based on the content available within their environment and often the recommendation is based on the user history. “Netflix won’t recommend live sports to its users because it doesn’t offer it,” he says. “There is no interaction with other data sources from other platforms that allows meaningful and relevant recommendations. This creates silos within the different environments. The data universe isn’t complete as far as the end user is concerned so the analysis of the data to provide a meaningful service is flawed once you look outside of each silo.”

Netgem says it’s helping to counter this by blending different sources of content on live, on demand, SVOD and catch up and gathering the associated data to provision a solution for operators to provide consumers with relevant recommendations across environments.

“The goal is to realise the benefits of moving from manual curation and QA, which is the current industry standard, to a hybrid model that combines the best of editorial expertise with the scalability, accuracy and dynamism of algorithmically-supported machine learning,” says Charles Dawes, Tivo’s senior director, international marketing. “For example, humans tend to be able to describe a TV show or a movie with emotional descriptors like its mood far more easily and accurately than machines. When that editorially-driven descriptive information is added to knowledge graph-based datasets, it enhances the system’s understanding of entertainment content and its ability to identify meaningful relationships to fuel more relevant recommendations.”

Josh Wiggins, CCO at automated metadata collection, curation and search specialist GrayMeta would like to see AI rebranded. “People are scared of the word. But I think there’s real opportunity to bring data together and to let human deals with the important 10% that machines can’t understand.”

Others suggest that rather than being replaced by machines, human jobs will evolve to more creative or oversight functions. “Advanced systems will change the way things are currently done and will require supervision and proactive management to make sure key business objectives are reached,” says Trudelle. “Constant benchmarking with other predictive applications is often needed to validate the performance of the new system. So, it’s clear that while automated AI/ML technology can help reduce the workload associated with some tasks and improve business performance, it does not take away the other more business-centric angles to consider when deploying any IT-based solution: scoping the business issue, defining the “why?  and what if?” questions, making sure the outputs have an actionable business impact.”

GDPR and AI Automated individual decision making, heavily based on AI and machine learning algorithms, will fall under greater scrutiny as the EU’s General Data Protection Regulation (GDPR) come into effect next May.

“OTT and TV companies are using Al-based applications for many processes including fraud management, personalised content recommendations, personalised marketing offers, and increasingly for programmatic advertising,” highlights Viaccess Orca’s SaaS director of product marketing, Ludo Rubin. “They are leveraging user profiling to predict the content piece that a customers’ will like the most; the personalised offer that will prevent churn or the most impactful ad to serve.”

 The GDPR wants, in particular, to guarantee transparency and equal rights when algorithms operate. Consequently, OTT operators are required to obtain explicit consent from customers to collect and process personal data, and be ready to share some information with them about the logic involved and the significance and envisaged consequences of such algorithms. While ML technology is constantly improving, we shouldn’t assume it is going to get it right every time.

“Although it has the potential to make media providers much more efficient and enable them to offer much more personalised services, it will always need that human touch to check it has got it right,” emphasises Child. If today’s mindset is man versus machine, “What we see as the future is that man and machine together can be better than the human,” says Anand Rao, AI researcher and principal at PwC.

Down the track Looking to the mid-term the realm of possibilities is expected to keep growing and be vast, from improving the user experience (UEX, content, advertising, platform/network performance), with more relevant personalisation, to optimising video and business operations.

“We will hear more of the ‘zero interface’ model where tools like Alexa and Google Home will be used to access automation systems,” predicts Niall Duffy, CMO, Virtual AI. “There are already examples of Alexa and IBM Watson doing ‘conversation-style’ operations for support and information checking.  The area of scheduling and rights management is also prone, again in quite basic areas such as the digitisation of contracts and then the ability to use AI to check and query contract terms.”

Child forecasts that “In future, we will be able to generate personalised video content based on audience profiles, behaviours, location. There’s already a great deal of investment from media companies capturing detailed information about viewers to enable them to serve personalised content and ads. Using video intelligence, this process can be automated and much more accurate, meaning consumers only get content that is highly relevant and interesting to them.”

Oliver Botti, head of international business development at Fincons Group, agrees, arguing that AI will increasingly help analyse vast and complex data sets spanning from specific TV content choice, to products purchased over ecommerce from a specific viewing device such as a tablet computer or even from the smart TV itself, helping to define a much more accurate content recommendation process.