Scriptwriting and artificial intelligence: towards new subjects?

Since ChatGPT in 2022, generative AI has been transforming our ways of creating and thinking. These technologies, linked to artistic forms, are reshaping our worldviews. I invite you to explore, in an unconventional and perhaps slightly unsettling way, their impact on screenwriting. The goal is to prepare for profound changes.

Anthropological Changes Induced by Technologies

Generative artificial intelligence technologies have been advancing at an incredible pace since the release of ChatGPT in November 2022, the first general-purpose conversational agent powered by deep learning. I believe it is important and interesting to date this article to January 2025, as it may have several future iterations and updates.

In my view, it is essential to preserve traces of our relationship with artificial intelligence, as AI tools are deeply intertwined with the artistic forms we produce. It is crucial to create a dialogue between working methods and the techniques at hand. Understanding the inner workings of cultural productions—films, performances, books, music, etc.—is important because it is through these that our culture and worldviews are constructed.

Here, I address the creative implications of artificial intelligence for screenwriting, as of January 2025, from my perspective, which is partial, like any viewpoint on this subject. It seems impossible to think exhaustively about this, as there are countless parallel experiments happening worldwide; let us embrace our partiality.

A few days ago, I spoke with a senior executive at a major tech company who told me that, in her opinion, the changes brought about by AI today are comparable to those caused by the widespread adoption of mobile phones in the early 2000s, especially after 2005 with the advent of 3G, which brought internet access to mobile devices. This technology, and particularly its widespread use, has induced anthropological changes—alterations in lifestyles and even living conditions. I find this perspective quite accurate. Even though these are two completely different technologies, applied to distinct aspects of human phenomena, I believe it is enlightening to consider AI in this way and how it integrates into our lives, rapidly transforming their nature.

Toward a Culture of Change

This comparison helps us realize that with AI, we are only at the beginning. Just as mobile phones, as they became ingrained in our lives, increasingly altered various aspects of our daily existence, the same will happen with artificial intelligence. We could also draw a parallel with ecological concerns: mobile phones are highly dangerous to health, and AI is extremely resource-intensive, raising broad ecological challenges.

Let’s return to screenwriting. How can artificial intelligence influence screenwriting? What is happening beneath the surface? I propose to examine this topic carefully, as there is a superficial way to approach it, which is to think that AI simply allows for faster research than Google. Indeed, Google and the internet already exist, and AI appears as a new function of the internet, a new way to access human data in a structured and formulated manner. While a Google search provides a list of unformulated results, AI relies on the same statistical data but produces a synthesized formulation.

Yes, the increased processing speed may seem significant, but it is more of an evolution than a profound change. Thanks to AI, we will make fewer mistakes, save time on automatic script formatting, etc. All of this is important, but it remains within the realm of evolution. What seems more significant to me is understanding how AI could replace the author. This is a contentious issue for screenwriters, as it raises the question: what is our purpose if a machine can replace us?

Of course, one might argue that machines will never replace humans, that imagination is uniquely human, etc. I apologize to those who might be shocked, but I will present perspectives that I find rather encouraging. I invite you to read to the end, even if what you are about to read may unsettle you.

What strikes me as most important about AI are the changes it will bring to the world. We cannot predict these changes, but we can prepare for them. Of course, the future always surprises us, but I believe that the more we prepare, the more capable we become of adapting to the surprises that will arise. If I do not prepare for change, I will believe or convince myself that things will remain as they are, and when change comes, I will be utterly unprepared and destabilized. However, if I can imagine that the world will change, even profoundly, then even if the changes that occur are not exactly what I had envisioned, I will have prepared myself for change. Future changes will surprise me just as much as anyone else, but I will have cultivated a mental agility that will allow me to adapt, anticipate, and perhaps even thrive.

Big Blue, AlphaGo, and Deep Learning

Let’s take an example outside the audiovisual field to lay the groundwork for the concept I propose in this article. In 1997, the machine Deep Blue, an IBM computer, defeated a human at chess. The world chess champion at the time, Garry Kasparov, was beaten by a computer! It would be difficult to call this artificial intelligence, although that was the term used at the time, because it was what is known as “brute force.” There are a very large but finite number of moves in chess. Thus, the machine could anticipate all possible moves from a given point in the game and, among all possible moves, choose the one that gave it the best chance of success—statistically, at each step, selecting the optimal move to win the game. This brute force approach is not very interesting, even if it is highly effective. Not everything can operate on brute force.

A later example occurred in 2016 with AlphaGo, a Google software that tied with the world’s top Go player. The difference between Go and chess is that in Go, there are an infinite number of possible moves. Thus, brute force cannot be applied to Go. So how did the machine manage to win? At the time, no one believed a machine could match or defeat a human at Go; it was predicted that this might happen, but perhaps 30 years later. Yet, it happened, thanks to deep learning, which is used by today’s generative AI systems.

Deep learning works as follows: we give a machine input data and a result to achieve. For example, two photos, and the result to achieve is the two names of the people in those photos. How does the machine learn to recognize them, that is, to build a “reasoning” that allows it to associate an image with a name? This concept is based on the Perceptron, updated, which was conceived and programmed in 1957 by Frank Rosenblatt—showing that our current reality has been in the making in research labs for a long time! We start by dividing the photo into a certain number of pixels. Each pixel is connected to a logical “neuron,” which has a “weight” in its relationship to others. There is a first layer of neurons that represents all the information in the photo. We assign a certain weight to each pixel. Then they interact to produce a higher layer of fewer neurons, connected differently, in a more “qualified” way, which in turn produces another layer, and so on. These are called abstraction layers. We end up with a symbolic representation, very synthetic, that makes this vast complexity of neurons correspond to a word, a name.

Training machines is therefore necessary. This is the most resource-intensive and costly operation in AI, which is why we are told that ChatGPT, for example, does not account for the present but only information up to a certain date, as the time and cost of training are immense. How does training work? We ask the machine, through the layers of neurons, to produce a sequence of letters. It will take many paths through the layers to produce different sequences of letters. But how does the machine find the right path to the correct sequence of letters? This happens through what is called “backpropagation of error.” To summarize, we make it go back to correct the path and refine it step by step until it reaches the correct sequence of letters, which is the correct name (provided during training). Inside, it’s as if there are two interlocutors: one moves forward, and the other asks it to go back and correct.

Once training is complete, we, as humans, do not know the path. It is not an algorithm we have mastered; it is this “double machine” that has patiently forged it itself. We do not know exactly what the reasoning is, and can we even call it reasoning? But we can observe that, starting from an image, the traversal of layers leads to a name. And then, if we give the machine another image of the same person, taken from a different angle, under different lighting, or at a different age, etc., it will still be able to recognize the same name by traversing the layers, because it “understands” what is common between them. To create this “reasoning,” there was a confrontation within the learning system, and it is this confrontation that produces the reasoning. I will revisit this topic when I return to screenwriting.

Let’s return to AlphaGo. This machine, through the deep learning system I have just simplified, imagined moves, sequences, and openings that were unique due to its inhuman reasoning modality. We humans do not reason in the same way. We do not spend months or years in feedback loops. We do not fully understand how we learn, but we assume it happens differently. Thus, AlphaGo invented openings that humans had never conceived, because AlphaGo has a “reasoning modality” different from human reasoning. And today, there are Go players who use moves, particularly openings, that were imagined by AlphaGo! The machine’s unique capacity for imagination has enriched our human knowledge and abilities.

Machine Screenwriters

Now let’s move on to the main topic of this article: “Machine Screenwriters.” If we revisit the principle of confrontation that enables reasoning, we can criticize today’s generative AI for producing very conventional results. For example, if you ask it to write a romantic movie script, it will draw on everything it can find on the subject and produce a sort of statistical average of a film, with no originality. Thus, at first glance, there is no fear of competition!

However, we know that the more detailed our requests (the “prompts” and conversations with the machine), the more AI provides relevant, detailed, and unique results, opening up imaginative possibilities we would not have had without it.

So, if we imagine two generative AIs collaborating to write a script, and we take two generative AIs with very different algorithms, developed by different labs, they already have internal confrontations, but by making different tools interact, this creates a form of otherness. And if we instruct them to push each other toward novelty, toward less explored ideas, well… they will do it! Let’s go a step further: to enrich ourselves through our differences, as humans, we need to create spaces for dialogue and collective intelligence, which works well with two people but becomes increasingly difficult as the group grows larger.

What if we developed screenwriting software that contains these two different engines within it? These are just machines, so we can scale from 2 to 4, 8, 16, 32, 64, 128, 256, 512, etc. We can exponentially multiply the collective intelligence of machines with computing power, which is impossible with human intelligence. Thus, we can easily imagine the construction of new imaginaries, ways of telling stories different from our conventional approaches, new rhythms, narrative logics, and ways of working with our perceptions of time and space. We can fully anticipate, just as AlphaGo did in its time, that machines, on their own, through deep learning, will invent forms of expression that are entirely enriching for us humans, forms we could never have conceived ourselves, but which will enrich us and become part of our palette of worldviews and ways of life.

What I find most interesting about AI is the scripts it will write on its own, because they will be non-human and thus fascinating to us! These scripts will transform the ecosystem of films, bringing new ways of telling stories. Just as music videos influenced cinema, machine-written scripts will enrich our palette of worldviews.

I believe that the role of human screenwriters in the future will be to engage with scripts written by AI, to draw inspiration from them to create new works. Machines will become interlocutors, very different from humans, capable of enriching us through their difference. In my view, it is essential, starting today, to explore these new possibilities and adapt to the changes they will bring.