Fear and loathing in artificial reality

When I first started messing around with generative AI in 2022, my reaction was mostly fear and, if I’m honest, a fair bit of loathing. 

Terry Gilliam’s 1998 film Fear and Loathing in Las Vegas is about counter-culture disillusionment set in a warped, unreliable reality. Early generative AI felt a bit like that. The world suddenly felt rubbery. People, places and moments could be invented, mangled and remixed in seconds, with no physical reality behind them. 

If something is clearly a style, anything goes. If it’s trying to be real and fails, it fails.

It was fascinating and mildly horrifying. But you could see where this was going; If a machine can do this now, what could it do in a few years time? My professional assessment was simple: we’re fucked. 

But as the tools have evolved, so have my feelings towards it. 

Above: Famous early AI footage of Will Smith eating spaghetti.

We don’t hate artifice. We hate seeing the joins. 

Film-making has always involved artifice. From casting to final grade, every choice is made in service of an idea, not the objective truth. Even our visual language for “authenticity” is constructed. An imperfect handheld camera. 'Natural light'. Non-actors. Vignette. Awkward framing. All directorial decisions that audiences have learned to read as 'real'. 

But AI felt different because there was no perceivable decision maker in sight. Technology and moving images are inseparable, and revulsion at new 'automated' advances is nothing new. 

Even straight up terrible animation seems valuable because it was made by humans. 

CGI had a similar pathway, personified by the 'uncanny valley' issue of the early 2000s. If something is clearly a style, anything goes. If it’s trying to be real and fails, it fails. Remember the ridicule and backlash aimed at the more than creepy characters in the Polar Express and Beowulf. Ouch. 

Above: The Polar Express left viewers feeling uneasy with its realistic, 'uncanny valley' characters. 

Audiences happily accept abstraction, and they’ll happily accept enormous amounts of invisible manipulation; they object when they can feel the manipulation. As if they’re suddenly aware someone is trying to trick them, the illusion is broken and suspension of disbelief comes crashing down. 

Back then, creating perfection was hard. Now, in an age of immediate artificial perfection (mostly), we crave things that feel human and imperfection is becoming evidence of authenticity. 

Authenticity isn't necessarily about provenance. It's about whether the thing being communicated is true. 

For decades, technology has been moving towards sharper images, cleaner skin, flawless VFX, etc. And suddenly we're deliberately asking for mistakes. We’re teaching machines wabi-sabi - to value irregularity and imperfection as signals of life. 

Concepts that could only be real, or less than perfect film-craft, become valuable because imperfection has become a proxy for authentic 'realness'. Even straight up terrible animation seems valuable because it was made by humans. 

Above: Despite its amateur animation (created entirely by a mother and son duo with no animation experience), online Chinese movie Niu Lai (牛来) gained meme status and quickly became popular online.

Where does authenticity live? 

Gilliam’s film deliberately uses extreme lenses, warped perspectives, lurid colour, unstable framing and distorted performances. It is highly artificial filmmaking being used to express an authentic, subjective experience. You could argue that there are really three different kinds of real: 

Physical truth - did this person/place/event actually exist in front of the camera? 

Representational truth - is what we're showing an honest representation of the person, product or experience? 

Emotional truth - does the moment feel recognisably human and true? 

Advertising has always played fast and loose with the first while trying very hard to achieve the third. AI pushes that separation further. AI lets you quickly create something with almost zero physical truth that potentially contains a great deal of emotional truth. 

If AI makes filmmaking ten times easier, that doesn't mean we need a tenth as many filmmakers. Maybe it means we make ten times as many films. 

If an entirely artificial moment makes you feel something real, is it authentic? It's something we've loved for over a century in animation. 

Authenticity isn't necessarily about provenance. It's about whether the thing being communicated is true. You can create a completely synthetic image that expresses a genuine human observation. Conversely, you can point a camera at a real person and manipulate their story until it says something they never intended. 

Authenticity isn't a property of the image. It lives in the intent, performance and emotional truth of the idea being expressed. 

Above: Despite being an entirely CGI movie about robots, Disney and Pixar's WALL-E succeeded in evoking real, human emotion.

It's taking our jobs! 

Like many people, a big part of my fears about AI were the disruption implications. But automation doesn't always produce the simple outcome you might expect. In 2016, AI pioneer Geoffrey Hinton famously declared that we should “stop training radiologists now”, predicting deep learning would outperform them within five years. Ten years later, radiology is still very much a profession in demand where technicians increasingly work alongside AI rather than giving way to it. 

When technology dramatically reduces the cost of something people value, we don't necessarily consume the same amount more efficiently. Often we consume vastly more of it. This is called Jevon's paradox. 

There’s also an uncomfortable truth beneath the current AI boom: much of it is built on enormous web-scraped datasets acquired without consent. 

Digital cameras didn't eliminate photography as an art; smartphones created billions more photographers. YouTube didn't shrink filmmaking because video became cheap; it created an enormous new moving-image economy. If AI makes filmmaking ten times easier, that doesn't mean we need a tenth as many filmmakers. Maybe it means we make ten times as many films. 

While specific mechanical tasks in film and advertising will face displacement, my bet is that the overall market for visual storytelling grows with it. We may lose some familiar tasks and gain entirely new ones, but cheaper image-making could ultimately mean more things being made, not simply fewer people making the same number of things. 

Above: AI-generated Denzel Washington articulates the democratisation of media through new technology.

It's taking our stuff! 

There’s also an uncomfortable truth beneath the current AI boom: much of it is built on enormous web-scraped datasets acquired without consent. That matters. But there’s precedent for messy technological beginnings eventually becoming more legitimate systems. 

Ultimately, methods of production are replaceable. Authorship isn't. 

Napster blew apart the old music model through mass unauthorised sharing; today most of us pay for Spotify. AI will likely go through its own version of that transition, towards licensed data, provenance and clearer rules around what machines are allowed to learn from. 

Above: Ben Smith's AI commercial Invite The Horse for New York Lottery.