09/05/1982, 09/05/1982

We are going to answer ourselves, but Jorge will probably use artificial intelligence to translate, to cut… Be that as it may, we have to keep in mind that using tools to give our expression a better form does not replace the in-depth work. Since ideas, decisions, positions are ours. We are at a point where collaborating with these technologies does not matter as much for the results as they do for the thought process they emulate. In a way, this reasoning on process and result also applies to our film.

We belong to a generation of Colombian filmmakers working outside of the country on themes grounded within Colombian reality. We have known each other for years through the network of the Colombian diaspora. Our long friendship made the decision of working together on this project easier.

It all started in concrete terms after Camilo published his book Room of Shadows¹. We used this as an occasion for us to reconnect. But quickly the conversation began shifting towards issues like the disappearing of certain forms of representation, media, visual language. Facing this phenomenon, we would bring up the upsurge of new image creatio technologies, like artificial intelligence. At that time, Jorge writing the end touches of La imaginación artificial. La imagen después de la imagen², co-written with Josep M. Català.

We then realized that we had both become, almost at the same time, filmmakers who formulate an idea through [literary] writing. Additionally, that we had a thematic and aesthetic affinity regarding the way images are used to validate ideologies, justify economic tendencies, support institutional power structures and legitimate military actions. In light of this trend, our work aims at understanding through which means footage can also serve as resistance against imposed powers.

Rather filmmakers than theorists, we wanted to try exchanging these ideas directly on the ground of the footage. 09/05/1982, more than a film, is an essay, in every sense of the word.

There was no pre-established script. Our approach was pragmatic: we immediately tested our ideas by generating footage. The results drove us to make adjustments, inspiring new ideas to try. The process was based on the IA feedback information, allowing us to evaluate our reasoning. Several times, we noted that our ideas were simply getting lost as we had to remove quite a lot of footage. The material we would lose suggested that we were exploring the wrong paths.

We were aware that using artificial intelligence as a creation tool demanded a thinking process, the same way that traditional shots demand an educated outlook. In that sense, we feel that generating footage isn’t the same thing as creating footage. But it is true that there is a trend of believing that automated footage production competes with the act of creating. The difference does not reside as much in the quality of the output, as in the thought process guiding every decision at the moment of creating an image. AI, more than a world of recorded representations, is a ground on which choices operate, a language develops, references get translated. We believe that our previous experiences (that is, either in traditional cinema technologies, or digital technologies) helped us understand the ways in which we would occupy this ground.

Very quickly, we understood that the difficulty was not technical, but conceptual: how could we prevent AI from imposing us its mindset? How could we resist the temptation of the spectacular? How could we create little information in a system which, a priori, allows anything to be generated. These questions allowed us to go from simply generating images to creating dialogue forms in tension with the machine. We realized we had to force existing tools so that beyond their common use, they would work in a way that suits us. For example, Jorge imagined a way to transform Telegram into an interface coupling in one operation the generating and sending of images. Through this system, we were able to dialogue like a visual ping-pong, generating and sending images in real time. The answer of one was sometimes an attempt of the other to reformulate what each would identify as the prompt³ of the received image. But sometimes the answer was a total change from the logic of what was received. We also worked in an iconotextual field.

For the most part, these are not intentional mistakes, but rather artificial intelligence mistakes that we found very useful for our work. In fact, one of the questions we were most interested in was the possibility to generate “poor” images in the sense proposed by Hito Steyeri. Low aesthetic and technical quality footage… what these types of model does not design. Most AIs are formed to generate spectacular footage. Generating grainy, unstable footage, like analogue film footage, is not the norm. The will to work with the mistakes that arise when AI doesn’t know what to do, “artefacts” if we had to give them a name, was more of a discovery than a direct research. We consider these “artefacts” like the shivers of the system revealing its flaws.

And it is also true that the models have changed. Mistakes used to be more visible a year ago. Now everything is more perfect, more “believable”. This is why we felt it was important to capture these moments of imperfection, where the models still show their limits. This is were a great part of the meaning of the film is at play.

The issue isn’t new but it did grow. The link between image and truth has always been fragile. The idea that an image documents a fact, that it can bear witness of it, has been questioned since the start of documentary cinema. What AI is doing isn’t inaugurating this tension, but rather bringing it to a new scale, a massive scale, automated and accessible.

Before, modifying images or building plausible narratives demanded specific skills. Today, a greater number of people can do it without knowing anything about editing, photography or aesthetics. This changes everything, not so by inaugurating something, but rather by normalizing it.

So it is not only representation at stake anymore, but the trust in what is seen. This opens debate on the origin of the footage, its traceability, its authenticity and its capacity to be a reliable source. In short, it is about knowing if it is still possible to maintain a sort of pact with what we call the real.

Documentaries have worked on this flaw for years. What is new and positive with these new technologies irrupting, is that they force us to implement new ways of selecting the documents they propose. Creating isn’t only producing anymore, it is discriminating, choosing. And beyond that, it is searching the missing image in the ocean of replicas offered by visual capitalism: the image that would allow to question the others and open the gaze towards new horizons. This is where we wanted to intervene with this film.

1. RESTREPO, Camilo The Room of Shadows. In the Archivo de cineastas collection.. Two bilingual versions (Spanish/Eskera & French/English). 145 pages. San Sebastián: publisher Elías Querejeta Zine Eskola, 2024. ISBN 798-84-7907-864-5

2. CABALLERO, Jorge & CATALA, Josep M. La imaginación artificial. La imagen después de la imagen. In the Signo e imagen collection. 360 pages. Madrid: publisher Catedra. ISBN 978-84-376-4862-0

3. Prompt: instruction given to generative AI, understood in this context like the description of the desired footage.