Netflix and machine learning is one thing - but for the rest of TV, AI is just hype
Artificial intelligence (AI) hype is everywhere. But does AI truly have a place in the TV industry?
Electronics company, LG, is testing smart assistant integration that could streamline TV interaction, meaning it may soon be time to say goodbye to your remote. When it comes to content, the BBC used AI to create programming; with two nights of AI-assisted content for BBC “4.1”. While over in China, in a world first, AI TV hosts joined humans on stage for China Central Television’s (CCTV) Spring Festival Gala.
But don’t be fooled, the AI revolution hasn’t quite reached your living room screen. Enthusiasm for smart technology may be high – even the government has pledged to invest £300 million in AI projects as part of a £1 billion sector deal – but capability is limited and we are still some way from futuristic viewing and bot-powered planning. It can be artificial, but is it intelligence?
The hype cycle
A key reason behind the surge of interest in AI is simple: it’s on trend and we’ve been waiting a long time for those flying cars, robots and all. Research shows 85% of UK businesses plan to boost AI investment by 2020 and recruitment for those who have an “AI skillset” is increasing.
The demand for “data scientists” spiked when the term entered the mainstream, causing a 57% global rise in the number of data scientist roles. However, that role is now ambiguous, incorporating roles such as operation researchers and statisticians but the need is still for good data analysis – not AI.
Despite the vast power and capability of AI, most technologies are still in early stages of development. Just take a look at the positioning of AI innovations on Gartner’s latest Hype Cycle; on the whole, dreams still exceed reality. But to prove AI is worth it, you have to first make sure it works and this is a difficult proposition.
From sophisticated algorithms to super AI
In the TV landscape, current AI applications are limited. The nearest use is Netflix’s or in OEM content recommendations — essentially deep learning — that analyses past viewing data to inform tailored suggestions. This technique is filtering through to the publisher side. Here in lies the dreaded filter-bubble as human interest really peaks when discovering something new and unexpected the New York Times has deployed multiple algorithms and statistical models to suggest a mix of related and random content for its readers.
Beyond this, other use cases remain relatively hopeful. Machine vision is making advances in semantic analysis of video content that could improve contextual targeting. Digital video manipulation is opening up possibilities for easier creative adjustment: with graphics processing used to quickly adapt ad elements such as the featured product. Other initiatives aim to employ Bayesian or regressive econometric approaches to determine the likely impact of media and external effects.
Even this type of machine learning is still not what many would truly consider AI. But there’s a lot to be gained from analytic techniques that don’t quite deserve an ‘AI’ moniker, for example; signal processing, exposed or control attribution and multi touch attribution can show exactly what type of media and which segments provide the highest likelihood of return on ad spend (ROAS) and this can be used to rapidly and repeatably make TV campaigns better which is good for both the buy and the sell side.
TV needs effective measurement, rather than AI
When it comes to TV, there are three key factors to note. Firstly, the majority of AI tech actually applies to digital, which restricts advertising scope to video on demand (VOD) or over-the-top (OTT) content. Secondly, these tools aim to provide a faster way of conducting the statistical analysis people have been doing for years. And finally, the quality of AI output is only as good as the data it feeds on — and this data is often low quality – hampering the ability to provide a “faster result.”
The TV industry can’t afford to get distracted.
The problem to solve isn’t a quest for a futuristic sci-fi ideal, what we actually need is measurable TV that works for the challenges the industry faces today. A human (as opposed to programmatically driven) media planner, for example, has an enormous compendium of information in their heads and they can easily simplify this for the task at hand. Whereas an AI algorithm starts with absolutely nothing, it has to learn everything from only the data it’s given, with no external context beyond what it initially is programmed. The priority should be giving those in the TV industry the tools that elevate the knowledge already in their possession, rather than constructing something from scratch which needs to be trained how to be “clever.” This can be done by looking to machine-learning engines as a first step into the world of AI.
The decisions and subsequent actions will be made by people, utilising automation to ingest bigger datasets, and then analysing every possible marketing scenario to find the appropriate media mix and tactical placements.
Clearly, there are areas of TV where advanced machine learning can be very beneficial, but it’s about cutting through the hype and clutter and not getting distracted by expectation versus reality. What we really need is accurate, timely reporting that’s actionable.
Find out more about Digital Marketing World Forum (#DMWF) Europe, London, North America, and Singapore.
- » Learning social best practices from industry experts: Find your voice and make ‘clever’ use of data
- » Getting started on the data-driven content journey: Machine learning models and upskilling SEO
- » Examining the second cycle of programmatic – and the effects for end users and advertisers
- » Salesforce dives into blockchain with low-code CRM offering
- » The power of propensity: Understanding this measure's role in your marketing efforts