You may have seen this in the News: London premiere of movie with AI-generated script cancelled after backlash
After the premiere was canceled, the filmmakers decided to release it for free on YouTube. I was curious enough (and had a perfect repetitive task to do while it played) that I decided to give it a go.
The premise is that every word of the script was written by ChatGPT 4.0 with minimal input from the writer. In fact, the LLM is fully credited as the screenwriter. It's directed by Peter Luisi, a Swiss filmmaker of some moderate renown, whose previous quirky films have consistently received more positive than negative reviews. It is made with a decent budget and professional talent, and it could be a sign of things to come - it certainly portends to.
But is is any good?
First the good: It is surprisingly coherent, likely helped along by human input and editing. A scroll at the start of the film reads: "'Write a plot for a film where a screenwriter realizes he is less good than artificial intelligence' - This was our first prompt and only creative input as screenwriters. Based on that plot we asked chatGPT to generate characters, a step by step outline, and to write these scenes one by one. Not a single word was changed or rearranged. The only thing we were allowed to do was shorten." I don't really know what all that means, but it makes it sound like there was likely a lot of prompt engineering and human editing that happened to make the script we ended up with. The dialogue in the individual scenes is, unsurprisingly, fairly natural sounding - after all, that is really the LLM's wheelhouse.
Taken as a whole - that's where it starts to fall apart. Every scene, every dialogue, every conflict centers around lengthy and repetitive philosophizing around this central question. Whether at a cocktail party or in bed with his wife, or talking to the puck-shaped AI device, every line of dialogue concentrates on whether a machine can "truly" represent human emotions. It's as blunt and repetitive as being jumped in an alley by a little league team. It just hits you over the head repeatedly.
As if in answer to its own question, every conflict is contrived and resolved at the mechanical convenience of the plot. The screenwriter is initially resistant to using the device at all, but his producer insists and he is immediately onboard (apart from the lengthy and - as I mentioned - repetitive laments he offers about giving up his career to a machine. The screenwriter's wife leaves because he becomes too obsessed with the device, but ultimately takes him back in a dialogue that sounds like it was generated by an algorithm. Because it was.
The character's mentor dies abruptly, seemingly because the story needs an emotional beat and is done with him anyway. He only shows up a couple of times before, and the previous scenes are so unremarkable that I had to go back and remind myself who he was and if there was some allusion to an illness or accident (there is not). It's only 72 minutes, so it's telling that an entire character could be so unmemorable.
Unfortunately, the acting and the sound design are likewise robotic. I found myself wondering if the film was scored by a machine, too. This is the kind of fare you might expect from any low budget drama. On that note, it was not the worst film I've seen. This is the type of writing that could certainly compete with Hallmark romances and low-budget cop dramas. It hits all the beats (with some likely basic human intervention), and delivers a story that is adequate. Not good, mind you. Adequate.
There are strange plotholes, like after the success with the first script using the AI as an assistant, his producer asks him to submit a script written entirely by the device. That fundamentally makes no sense. If it is written entirely by the device, then what exactly is the producer asking him to do? Press the "go" button? The scene makes sense in an LLM-logic kind of way. It resembles an actual conflict if you squint and don't think about it for more than 2 seconds.
Like most of what LLMs and other generative algorithms creates, it is a fair approximation of human creativity. It outperforms half of the film school students hawking scripts in Hollywood, and indeed I think it is an omen of how films will be made (and already are). No need for studios to bother with writers for basic rewrites or to pay humans to churn out another low quality, background noise Netflix series. The central question of the film: can chatGPT make art? No. Obviously not. The film attempts to take itself seriously, but it is fundamentally drivel.
It is not good.
I'm planning to come back as Fermi's Paradox.