Why AI is losing the creativity fight (literally)

Matt Murdock (Charlie Cox) expresses how I feel when an AI bro sounds off. Still from Daredevil series 3

Recently my social media feeds started showing me clips generated by an artificial intelligence (AI) engine called Seedance 2.0. These were all clips of fight scenes that the poster was breathlessly claiming needed ‘just one prompt’. Wolverine fighting Superman, Godzilla versus King Kong, Terminator fighting Robocop, there were plenty, all of which proclaimed that this video generator was ‘the end of Hollywood’. Except, they’re all wrong. None of these clips had any actual creativity, storytelling, or well, any soul.

Computer-generated imagery (CGI) has been around for a couple of decades now. I remember the mess that was the Scorpion King in The Mummy Returns, but a year earlier Ridley Scott had made use of the same technology to produce the vistas of ancient Rome in Gladiator (as well as the avatar of Oliver Reed). In more recent years people have complained that the CGI used in blockbuster movies from studios like Marvel/Disney has become a bit dull, with action scenes often taking place in featureless landscapes like deserts (Deadpool & Wolverine, The Flash) or similar. Even the final battle of Avengers: Endgame took place against dark grey rubble and dirt (compare that to the first film’s battle for New York). 

The reason for this is simple: as CGI has become more and more ubiquitous, the workload of the studios has increased exponentially and so they have to cut some corners. The thing is, a well-produced CGI-heavy action scene will always win out over the clips being lusted over by thirsty AI bros online for one simple reason: people are involved. There has been some creative input and quality control, checking for continuity errors and similar, something that the AI clips have failed to account for. 

So, seeing as the AI cheerleaders are focusing on fight scenes to support their ‘this will end the movie industry’ argument, allow me to retort. First, let’s take a quick look at one of their vaunted fight sequences. This is listed as Daniel Craig vs Matt Damon, so it’s Bond vs Bourne.

I know it’s only a few seconds long, but there is so much to pick apart here. Neither ‘fighter’ is displaying any kind of individualism. In fact they both advance on each other like characters in Street Fighter except with less style. If we look back at the fighting on display here, it bears virtually no relation with the fighting styles displayed in either of these actor’s action franchises. The AI just has two avatars throwing clumsy haymakers like two drunks who think they’re Rocky Balboa.

I think that ‘Matt Damon’ (to be spoken Team America-style) is the better example to use. In the Jason Bourne films, Damon’s fighting style is based almost entirely around the Filipino martial arts systems, often referred to as kali or escrima. These are highly distinctive fighting styles with both armed and unarmed techniques. Here’s a quick example of Matt showing off his moves in The Bourne Identity. Can you spot the difference?

But, I’m not done with the AI fight bros yet. Your engines, LLMs and generators cannot compete with human creativity and ingenuity. Your digital lumps throwing awful punches with the inconsistencies in style and presentation etc lack some key components of any good action scene. They don’t tell any kind of a story or even contribute to the overall tale of a piece and most of all they don’t make the audience feel anything. I would like to take some time examining a fight sequence that does all of these and more. Ladies and gentlemen, may I present Matt Murdock, the Devil of Hell’s Kitchen.

The single-take sequences have become a hallmark of the Daredevil series, and while the stairwell and prison battles are superb, as is Daredevil vs Bullseye at the start of Born Again, I have to go with the hallway fight from season 1. I’m sure there will be some AI fanboys out there who will try to argue that the scene can at least be improved with AI, but my question is, how exactly? What can the technology find here to make it better? I’m not sure there is anything wrong with this sequence. 

First, it tells a story. Daredevil, still in his early days as a superhero, is searching for a child that has been kidnapped by Russian gangsters as bait for him. This fight scene occurs at the end of the episode when Matt finally catches up to them. We know the episode has been building to this, but the film makers still take their time setting the scene for the final confrontation, letting us know where everyone is, so the audience is fully aware of the gauntlet Daredevil must run before he even appears.

The choreographers also find a simple and effective workaround for a typical issue with one-on-many fight scenes. Usually, in these sequences, the bad guys tend to stand off respectfully while the hero takes on one or two of them at a time. Here, the gangsters are picking themselves up off the floor, or spilling out of other rooms, confused by what is going on in the hallway. Daredevil himself is injured and weakened, and it is his compromised state that generates such an emotional reaction to the sequence. He’s exhausted, slumping against walls, barely able to stand at points but still battling on. 

The scene highlights Daredevil’s varied and improvisational fighting style, disarming opponents to use their guns as clubs and magazines as projectiles (and then there’s the microwave!). Then there's the mixture of hand-to-hand styles, as he uses a combination of punches, knees, throws, headbutts and some typically acrobatic kicks to take out his attackers. Now look again at the AI sequence from earlier in this piece (or any of them really). I dare you to tell me that they are in any way superior to this or the fight from The Bourne Identity (or any piece of quality fight choreography from, well, ever). They're just slop.

The day may well come when the algorithms and engines become so advanced that these problems are removed, but there is also a good chance that we will have bled the rivers dry to power data centres by then. So, with the massive environmental impacts of this technology in mind, why do we need to use it like when we not only have the ability to do it ourselves, but do it so much better than the machines?

It simply does not compute.