Any production company working with a content library will know that it is worthless without some means of knowing what is in the archive. Increasingly, that means much more than titles and an outline of content: as automation becomes ever more prevalent, you need metadata which can provide actionable insight.

Over the last decade there has been significant advances in artificial intelligence systems, and specifically machine learning. This is driven by the growing effectiveness of deep neural networks – algorithms that mimic the structure of the brain, and which can then be taught to learn new things without explicitly programming them.

Machine learning algorithms have won cultural acceptance, not least through the widespread use of recommendation engines. We have all experienced the way that Amazon, Facebook, Netflix and others use machine learning to predict behaviour based on past activity – you liked that, so you may like this.

In many ways recommendation systems have paved the way for…