This is a continuation on from our previous post The Arc Project Update #2 – Tell The Model Workshop, part 1 so check that out first if you haven’t seen it yet.
The Arc AI workshop continued, albeit with fewer participants. Måns stayed on with me and Mohanty and his team to dive deeper into this exciting but mysterious new world.
At the suggestions of Florian, Mohanty and Nimesh, we continued to explore AI tools that already existed to experiment and imagine applications that could be useful to a writer.
There was a scene-generating model from NVlabs that we loved. This transforms textual descriptions and transforms them into 3D landscapes. Yes, you can build your own world and add whatever details you can imagine. Much like architects use 3D models to imagine how their buildings would behave in real world environments, I started to think about how this could help me build a character. I remembered so many marvelous films where scenery is integral to the journey of a character (the long tracking shot in the restaurant in Goodfellas, the entire ranch in Power of the Dog, for example). The protagonist of my story hails from a vast mountainous region of India. By all accounts it is elegant and filled with natural beauty. It is a landscape that formed my leading man so I want to imagine it on screen to inform his movements within the story. Then I imagined putting my academic leading man inside of an upscale New York restaurant: he might feel out of place and awkward, but this also started to get me thinking that it is the perfect mode to experiment with.
The more alien the setting, the more psychological pressure you put your character under, the more opportunities become apparent with a character. You could create any kind of landscape or interior in a few moments and let that fuel some ideas. The AI draws from what it already knows. You have to be specific if you want to create David Lynch – style world that is less tied to reality. So after toying with the AI, we realized that this too felt a bit limiting. The group sensed that it did not address a fundamental problem of screenwriting. Creating landscapes and scenery are not key to solving your story puzzle, it’s true. But I would recommend this tool if as a writer you have a period piece to write, or want to plan scenes that are location- specific. You never know what you will find.
‘Imagine you are kind of writing a movie about World War Two, and you somehow have no access to these digitized documents of conversations between, for example, some generals. Now it will take you ages to kind of analyze all of that. But if you have clarity on the kinds of questions you want to ask and you have that information in text, this model can go over all this text. It can organize this information. It’s like having an army of research assistants and that becomes very powerful, especially if you are doing this background research for some personality traits or character profiles, or historical details. If you’re an independent filmmaker depending on how perfectionist you are, if you are going on with any kind of research, it can be extremely expensive, right? Large media houses would have the staff. As an independent you would not.’ –
The mind mapping models do have potential. I imagine this would serve a useful purpose for those writing factual based scripts, like… me! Regarding the example of a war film, there are millions of films about WWII, and while the beauty is that there is a lot of information that is easily available online. For me using a mind map tool would be exciting to recognize gaps in a story that might have been overlooked in the past. One of the great rules of writing is to keep in mind that your work should be a good use of the reader’s time. And this makes a good use of a reader’s time because they are getting something new from a story they thought they already understood. (Like a war story for example)
Perhaps this AI model does not push me to write ‘harder’ but it helps me ‘see’ a storyworld clearer.
The real world use of a mind map like this could be for arranging a timeline of an event or a story. You could locate specific events in a historical timeline and look for discrepancies. But what I really liked about this part of the conversation was when Nimesh said:
‘Another potential use case that I can think of is like finding discrepancies in text. Imagine if you had this huge script and the timeline is quite messed up. This tool could neatly extract that information and kind of make a linear timeline showing a clear chain of events. I can imagine this model could help a writer identify potential places where you can inject plot lines, for example. Like it formats the data in such a way that you can analyze the whole story in a single linear setting.’ – Nimesh
Time is one of the best layers of film to experiment with. It is also one of the hardest. But when a story can move an audience through sequential time differently, and really keep a viewer wondering where they will end up next (like Pulp fiction, Rashomon, Memento) you have something unique on your hands because then you are approaching what lived experience of time is like. Using this tool with the idea of time opens up possibilities to a story in a real way. For sure a mind map tool does require some preparation, a significant amount of research and specificity in the questions you pose, but it could be worth pursuing, especially if you are writing a series. How would you keep track of the plot lines and story worlds within worlds, I wonder.
CHATTING WITH EINSTEIN
From this point our conversation fell into the very real fact that, in reality, AI is not aware of what it is writing. A frustrating fact. (As you will learn in my future interview with the director of a theater play written by an AI in Czech Republic, sometimes an AI displays its moods when you write with it!) So we thought to use this fact and start to play with other AI applications, many in beta mode but still very impressive and entertaining.
We had a great time chatting with this Stoic Today. I absolutely suggest you try it. It offers ‘profound’ philosophical quotes: the user had to guess if they were AI written quotes or genuine pearls of wisdom from a real philosopher. When we played with this AI, there were some great moments when the group debated core subjects like the nature of love, what is happiness, and even the essence of poetry. But soon enough we (mostly Mohanty, let’s be honest) cracked the code and we worked out that the longer the philosophical quote, the more likely it was made by a real person. (ha!)
What we learned was that, unlike a real philosopher, or even a philosophy student, AI can easily contract its own values. It can sound profound but it will not (or can not) expound an idea based on a foundational principle. This is what we as humans want philosophy for. To know and understand principles to live our lives by. And every great film character does exhibit a defined approach to life, whether that is a nihilist or optimistic belief system. It is their belief in life, or lack of it, that gives dimensionality and relatability to a film character. So how useful would a model like this be for creating a fictional character? Yet to be seen. To test out clever sounding dialogue, for sure.
Our final exploration of the day was with an AI model that mimicked a conversation with Albert Einstein. It had presumably absorbed many writings by and about Einstein to edge very close to an accurate prediction of what he might say. It was fun to converse with Albert, one of the greatest, most imaginative minds to have walked the Earth. It was not a rote conversation either. This would be a great tool if you were working with an archetype, and could base a character from a figure who is a public figure or is well known in some way. If your villain was based on a political figure you could, with some help from our tech savvy friends, feed information to a model that was specifically about this villainous character and eventually you would be engaging in a conversation with it. We spend a lot of time in the heads of our characters, speaking with them and asking them questions almost subconsciously. In fact one of the most fun writing exercises I know of is to imagine having a drink at a bar with your character. With enough journaling and informational inputs, you could have this drink with your character and test its reactions. I found that we grew close to the Einstein AI quicker than I expected. I would not say there was a bond but we did find ourselves leaning into the conversation as we anticipated what the AI Einstein would say next.
By the end of the session, we realized that for an AI to work for a writer there is a degree of management that must take place. We have to work closely with the AI, manage expectations, and dream together with it. But there are opportunities for us to work with AI researchers to create an AI model that would be simple to use, and simple to feed information into, in order for a writer to engage their imaginations.
This is just one part of the puzzle. We found that AI likes specifics. It loves hard information. Lots of it. It likes to know and understand a process. This is where it can be useful for people or for a company. While screenwriting is not a surgical process, Mohanty and I want to identify the many problems associated with screenwriting, and be detailed in how the journey of a script comes to life once it leaves the writer’s mind. A big task but we will continue to try and crack this, because it is something worth understanding. I think we will get there, and hopefully write something that is worth sharing very soon.