AI and Machine Learning are rare industry buzzwords being concepts with real staying power. Still, there is a lot of hype around how “AI” – the term we use when we mainly mean “Machine Learning” - can be applied across the media value chain.
With so many AI-driven solutions on the market, and a certain lack of understanding about this technology, how is it possible to get started, prove the value and manage expectations all at the same time?
This was the topic addressed at the recent Broadcast Projects / IET Media webinar, “How to Prove Value with AI & Machine Learning”. Our panel, moderated by David Short, Vice Chair of IET Media included top AI/ML experts and practitioners from WIREWAX, Al Jazeera and Fraunhofer IDMT.
Some of the key takeaways included recognition of the fact that video is different and more complex by an order of magnitude. If we compare the well-known use-cases for AI, such as autonomous vehicles and medical diagnostics, video is far more complex in terms of the number of cognitive elements needed to make meaningful sense of it.
AI and ML are often misunderstood. Stories abound of those who have tried but have been very disappointed with the results. Disillusionment settled in early and has placed some vendors in a difficult spot, being on the back-foot before they even get started with a pitch. They must be very careful when it comes to expressing how effective they will be at solving problems.
At one of our previous events, Fraunhofer IDMT’s Head of Semantic Music Technologies, Hanna Lukashevich, very clearly stated: “AI is not magic, and if something is not working there is usually a good reason for that”. Clearly defined user requirements are a major factor in mitigating disappointment in AI/ML services. Machine-assisted solutions will not solve all your video workflow problems, yet. Knowing what you can expect – and what you cannot - in terms of extracting value from your video archive is a key starting point, too. There are two options to consider in terms of getting started, you can either go with an off-the-shelf solution from a specialist vendor, or, you can get expert advice to customize individual components to address a really specific use case giving a more bespoke solution.
Fraunhofer IDMT offers a vast portfolio of components that can be customized for bespoke solutions.
A large number of use cases can be classed as winning propositions. AI and ML services can automate tedious tasks, saving time for humans and enabling them to be refocused on tasks that can only be done by humans. In this way, they become faster and more efficient at their jobs.
Machines can also reveal the invisible elements buried in your content. AI and ML can make the invisible, visible, enabling you to extract value from your archive, even when it’s poorly managed, full of duplicates and missing metadata. Simple tasks too can be speeded up greatly, such as aligning video content in multiple files, or detecting the “intro” of a program for automating a “Skip Intro” function.
Example of automating the alignment of content in multiple video files, from Wirewax.
Empowering smarter production processes is another powerful use case, especially in busy newsrooms. The state of the art of Natural Language Processing (NLP) is such these days that “metadata-less” search is possible – enabling the quick location of clips or images to illustrate a story.
Al Jazeera uses cases for news production.
Combining technologies enables a whole range of applications that support content moderation, bias and fake news detection at speeds that humans could never manage manually. AI and ML can also be deployed to automatically generate extended metadata based on facial recognition, NLP, location, content and contextual detection, providing a Knowledge Graph that can be interrogated for a variety of purposes.
Al Jazeera uses cases for news production.
This insightful discussion covered a range solid use cases and efficiencies that AI and Machine Learning can bring to your media operation including suggestions for getting started.
Watch the replay to find out:
What are the biggest learnings about AI/ML?
Which applications of AI/ML are people getting started with first?
What AL / ML means for the future of content platforms?
How AI / ML will impact recommender systems over time?
Why AI / ML are powerful tools for pushing intelligence back to the production side
Overall, AI and ML can help you make a series of small incremental gains to your business performance and give you an edge over the competition.