AI and Factual Television 2: Truth & Trust

As the Allies landed in Normandy in June 1944 and fought their way eastwards, their progress was filmed by British and American cameramen under the direction of Sidney Bernstein (later founder of Granada TV) from back in London at the Ministry of Information. When the concentration camps were liberated at Majdanek, Poland (the first major one, in July 1944), Auschwitz (January 1945) and Bergen-Belsen (April 1945) the unknown horrors revealed to these cameramen were sent back in rushes (dailies) to Bernstein. As soon as he saw them he realised that the way these inconceivable scenes should be filmed was important and he issued precise instructions to his teams. The filming was to be as incontrovertible testimony, using slow pans/camera movements and with a view to minimal editing, so that no one could accuse the footage of fakery.

Cut to 80 years later and the age of fake news and deep fakery. What role can footage created through generative AI play in factual filmmaking and documentary? The same issues surface – trust, authenticity and truth. AI-generated footage by definition contains no genuine truth, despite being the product of various source truths. Since documentary is fundamentally about documenting reality AI footage can by definition have no major role to play in it.

Where AI can contribute to documentary is in anything from the pre-film-camera age. For example, a thoroughly researched, historically accurate shot of the Roman forum in the reign of Nero.

It is also useful for resurrecting dead people. In the 2022 feature doc ‘Gerry Anderson: A Life Uncharted‘ director Benjamin Field deployed AI-generated images of the Thunderbirds creator’s talking head synced to archive audio recordings of the great man.

What about B-roll and GVs? Sunset over the jungles of Vietnam? Waves lapping the beaches of Normandy? This must be the wrong side of The Line because they document nothing. They are ultimately fantasy. Fine for scripted. No place in documentary.

When Stephen Lambert went down in 2007 for faking the Queen’s behaviour at a photo session, ‘Crowngate‘ proved an accidental act of public service by putting ‘Viewer Trust’ firmly on the TV agenda where it remains to this day. The issue has spread to other media, most recently in the doctored photo controversy. People want to be clear about what they are actually looking at.   

How can documentarians indicate material in their films that were created by generative AI? Benjamin Field did this by putting the Gerry Anderson talking heads into vintage TV sets to differentiate them from the regular factual footage. One way to do this as a standardised practice would be to create an AI ‘watermark’ to make clear what is not actual documentary.

Another way would be to establish a certificate that indicated ‘nothing in this film was made by AI’. At the moment a scheme of this sort is being discussed by PACT, Equity and BAFTA.

The 1 Habit of Highly Effective Factual Filmmakers is: “Trust is the glue of life. It’s the most essential ingredient in effective communication. It’s the foundational principle that holds all relationships.” (Stephen Covey)

Sidney Bernstein, 1st Baron Bernstein of Leigh by Howard Coster (courtesy of National Portrait Gallery, London)

AI & Factual Television

“The future of Television” created by Microsoft Designer’s AI image generator

Just back from CPH:DOX documentary festival in Copenhagen where, apart from running a workshop on stress-testing nascent documentary ideas, I have been exploring the interface between factual filmmaking and Artificial Intelligence.

I had an interesting conversation with multiplatform specialist Simon Staffans about the current crisis in Television in the UK, Europe and beyond. His observation is that, although there is tons of money knocking about in the world in the wake of the pandemic, quantitive easing, etc., the vast bulk of that is being invested in applications of AI in all aspects of life, and not the very 20th century technology that is TV.

What I am seeing at the interface of TV and AI is a strong focus on what AI can or soon will be able to do to speed up, improve or enhance the processes of video content making – from idea generation to pitching to editing to creating voiceover to cleaning up picture/audio to optimising distribution to upping discoverability.

Most training/CPD and briefing sessions at present seem to be largely catalogues of the latest software presented in broad categories representing production stages – that is, the What and a bit of the How. But the frame of the Why is for the most part absent.

AI can help speed up processes and reduce the human resources required – so quicker and cheaper to produce.

It can help fix dodgy pictures and degraded audio – so higher technical quality.

As the arms race for TV sizzle reels continues apace, it can generate impressive visuals, moving and still, to help get you in a room with a money-person. You can add a cloned celebrity voice-over and some in-the-style-of music for a very polished pitch.

But who are you pitching to? If the business models of TV collapse further, where will the funding come from to enable production companies to make good use of these amazing tools?

To what extent will AI give rise to new low-cost forms of content? For example, where text-to-video apps do away with the main costs of production. What’s the relationship between puppies playing in the snow and 20 Days in Mariupol?

Meanwhile today the BPI (British Phonographic Industry) went into legal battle with London-based AI start-up Jammable (formerly Voicify AI), which creates voice clones from BPI-represented and other music artists. The knotty legal/rights and ethical questions AI is throwing up are fascinating and watching them play out over the next couple of years will be as interesting and meaningful as all these other questions about what this major ground-breaking, era-defining, future-shaping technology signifies for our industry.