AI & Screenwriting
Can AI write a movie?
What was the last movie you watched that moved you?
Do you think AI could have written that movie?
The idea might sound bizarre, but writers are exploring the intersection of AI and human creativity. AI is no longer a replacement for mundane, repetitive tasks or super-powered pattern recognition. (Yes, those avenues are still popular and constantly being improved). But the newly emerging domain of AI for creativity is fascinating and worth knowing about.
With the emergence of tools like GPT-3 and DALL-E, AI is entering new exciting territories. Generative Pre-trained Transformer 3 (or GPT-3) is an autoregressive language model created by OpenAI that uses deep learning to produce human-like text. DALL-E (also created by OpenAI) is an AI program that creates images from textual descriptions. It uses a 12-billion parameter version of the GPT-3 Transformer model to interpret natural language inputs and generate corresponding images.
The journey of using AI for creative endeavour sounds very appealing, so a team at AIcrowd decided to go on a little adventure.
The Arc Project
Introducing Hussain Currimbhoy, filmmaker, producer and curator for international film festivals. Hussain has written and directed eight short films, many of which have played at international film festivals. He was in the process of developing a new story and wanted to explore an unfamiliar approach to creating a movie script.
While screening his latest short film at Cineglobe Film Festival, he ran into AIcrowd CEO and art enthusiast Mohanty. At the event, Mohanty presented a mix-medium music performance using AI-generated melody. This inspired Hussain and Mohanty to discuss the idea of developing a script by merging the efforts of a human writer and an AI tool.
“With ideas and shards of inspiration coming from left field, a duel with the writer would perhaps generate something enjoyable, maybe moving, out of a story idea”, says Hussain. “I felt convinced such a process could only push a writer creatively but also take them a step towards escaping that prism of a single perspective”.
Hussain and some local filmmakers, Fison Ali, Andreas Climènt and Måns Thunberg, met AIcrowd’s team in an online AI & screenwriting workshop.
AI & Screenwriting Workshop
In this workshop, experienced film practitioners and technologists got together to understand perspectives on AI in the industry, how it can be used as a tool and confront some myths & fears around AI head-on.
The workshop participants first learnt about GPT-3. As mentioned earlier, this is a language model that uses deep learning to generate human-like text. Since it’s launch, GPT3 has been embraced in many ways. It can be implemented as a chatbot, offering answers to some basic questions like ‘What is the average life expectancy in Sweden? Or it can behave like a philosopher, offering inspiring quotes. The group started to play around with this tool, asking some questions. You can read more about the experiments over here.
Florian, from AIcrowd’s explained the “temperature” switch on the model. He explaines that the model would deliver more random answers to your questions if you switched up that temperature. An AI model will offer deterministic and repetitive answers at a low temperature. Products like Siri and google assistant play it safe and give predictable responses. Hence, our associations with AI can be a bit mundane because AI applications are products for the most part.
This inspired everyone to turn up the “temperature” switch to bring a level of randomness and provide plot prompts and character outlines to the GPT-3 model. With only a few prompt lines, the model returned and produced a half-page of the formatted film script in a few moments, creating characters with names set in a location.
Måns had a fragment of a script idea and explained he was a bit stuck with a detail in the story. We entered this paragraph of his idea into GPT3:
A little kid gets run over by a car during a hit and run. In the car involved, one person is good, and one person is bad. They are on the run from something. They dump the dead child somewhere. The ghost of the kid wants to figure out what happens.
Remarkably, the model produced a half-page of the formatted film script in a few moments, creating characters with the names, (Rachel and Daniel) who are now driving in a car at night.
Curious to see what story is made? You can read the story over here.
The model operates by scanning the internet for specific information. AI locates these patterns in constructions common in language and, therefore, common in culture.
While doing various experiments, the group also realised that the question of where the model gets its information from has to be considered more deeply. “Writing by women, and from people outside the Western screenwriting system, will be necessary to create a fairer engine”. “On a creative level, this was key. To avoid repeating the past patterns, we had to be mindful to include a kind of internationalism, or at least a broader scope of the inputs to the AI”, Hussain concluded.
As a filmmaker, Hussain thinks the role of AI in screenwriting is the next big step. “Technology is being used to produce films the world over. The streamers embrace it, distributors pay for data, and visual effects houses run on AI. As writers, if we do not come to terms with and grasp the role technology plays in movie making, and thereby culture making, we might find ourselves much more vulnerable to progress than we like to admit”. You can read Hussain’s notes and observations on his journey in using AI for writing over here.
You can read about his experiments of using AI for screenwriting over here
Hey reader,
If you’ve made it till here, you deserve a cookie 🍪
This article is very different from our other posts. Did you find it interesting? Would you like to see more articles covering AI and its use in various domains? Let us know in the comments below. :)
