How to prevent the application of AI in the film industry leading to the delivery of the same films again and again and again

That AI’s trained with historical data might, unless care is taken, repeat the biases of the past is a well understood issue.

Garbage in, garbage out.  It’s a problem for many AI use cases, and the film industry is no exception.  Generic data analytics in the industry has already contributed to the rise of sequels, prequels and universes presented to audiences, the percentage having risen from 9% to more than 30% of the top 100 grossing US movies over the past decade.  Unless the industry is careful, AI will only make that problem worse, and in more ways that you might at first imagine.

Predicting the future may be more like repeating the past

The industry would like to better predict success, because making movies is very risky.  Less than 20% of films make their budgets back and even in Hollywood with easy access to talent and large production and marketing budgets, less than 50% of films are profitable.  The case for selecting projects with an AI fed with big data is compelling, and start-ups are stepping into the breach.

LA-based start-up Cinelytic, for example, aims to help predict the success of a production based on analysis of similar films, talent involved, and other criteria.  Belgian start-up Scriptbook aims to predict success at an even earlier stage by having their AI read your script and predict its chances of success.  And, inevitably, Scriptbook and others are working to develop AIs that can credibly generate stories.  OpenAI’s GPT-2 large-scale unsupervised language model delivered headlines in February, its makers declaring that the stories it is capable of writing are so good that it was too dangerous to release.

The problem is that unless developed with significant care, since each are fed on the stuff of yesterday, these AIs are going to deliver audiences yet more of the same along with all the gender and diversity biases that yesterday’s world contains.

Making actors younger means they never ever die 

In recent years, AI has most prominently been revolutionising visual storytelling, enabling incredible special effects with ever-increasing realism.  ‘Deep fake’ face-swapping AIs have become embedded in the movie industry in the form of face cloning and de-aging effects, recently giving us younger versions of Will Smith in Gemini Man and of Robert De Niro, Al Pacino and Joe Pesci in Martin Scorsese’s The Irishman.

Of course, the industry’s use of AI doesn’t stop there, with the producers of the forthcoming Vietnam War drama, Finding Jack, announcing in November that they are using AI to resurrect James Dean to play the lead role (Apparently they wanted Elvis, but somehow or other the King politely said no).

Another more subtle case in which the application of AI in the film industry means audiences are going to get more of the same: more of the same talent, for ever and ever and ever.

Dealing with mundanity may lead to more mundanity

In filmmaking there are a surprisingly large quantity of manual, mundane tasks, and this is another natural place for AI to offer its services, not least because the industry would rather see less names at the end of their movies than more.

Editing is one place in which AI tools are being developed.  Mackenzie Leake’s work at Stanford and that of Reduct.Video aim to help editors assemble and cut films swiftly, using NLP to identify scenes and cut points that a human would otherwise spend hours piecing together.  An opportunity for more creativity?  Well, not if easy editing algorithms are based on yesterday’s movies.  That may lead simply to more of the same kind of editing, and the same kind of visual storytelling as we’ve seen before.

The industry needs today’s as well as yesterday’s humans in the loop

It would be near-sighted to suggest that AI-driven analysis of movies already made, AI-powered delivery of advanced audio-visual effects, or AI-enabled freedom for creatives from mundane tasks cannot lead to innovation in the film industry, but – as with many other industries – there is a real need to ensure today’s humans are in the loop, and not rely wholly on the analysis of yesterday’s world.

Human-in-the-loop machine learning is a branch of AI which recognises cases in which there is a need for humans to seed, calibrate and test machine learned outcomes.  In cases when innovation and diversity are sought, regardless of whether a concept was dreamed up by a human or a machine, being able to assess the reactions of today’s humans to those concepts can be a crucial part of the loop.

For the film industry, this reveals a very particular use case for AI: the ability to assess an audience’s emotional reaction to a movie.  In fact, being able to understand in detail how audiences are reacting to the story and execution of a film from the very beginning and all throughout the development, production and marketing of a movie would enable the film industry both to take and to measure risks; to experiment, discover and iteratively improve their output.

Being able to prototype audio-visual content earlier in the development process is itself being helped by the application of AI as more and more film-making essentially becomes digital animation, lifting the fidelity for test audiences beyond the words of a script to a story they can watch and react to naturally.  Pairing prototype-generating AI tools with AI-driven emotion and engagement detection, such as the automatic discernment of emotions via real-time facial analysis, provides writers. directors and producers with the tools of experimentation that enable both the taking of risks and the measurement of them: the core components of an innovation loop.

Of course, it’s not only in the film industry that there’s a danger that the application of AI based on yesterday’s humanity can lead us into cul-de-sacs.  For AI to get us to a better future, we need to break the tyranny of yesterday by putting today’s humanity in the loop.  Repeat ad infinitum.

About the Author

Simon David Miller is a seasoned filmmaker and technology entrepreneur. His first feature film was the BAFTA nominated Seachd – The Inaccessible Pinnacle, the first Scottish Gaelic film to be released in cinemas. Over the past two decades, Miller has set up and sold numerous technology startups from Peoplesound during the first dot com boom to, most recently, Zeebox and Beamly. Under Miller’s leadership New Forest Film Co are producing a slate of TV shows and films with Agile, the first of which to be released will be the feature film, Invisible.

Featured image: ©Metamorworks