BY Paul Chan in Opinion | 05 JUN 24
Featured in
Issue 244

How Chris Marker’s ‘Dialector’ Foreshadowed Machine Intelligence

The computer programme created by the artist in 1988 functioned as a chatbot long before the development of ChatGPT

BY Paul Chan in Opinion | 05 JUN 24

After the 1983 release of his nearly uncategorizable film-essay Sans Soleil, the protean artist and filmmaker Chris Marker started developing a computer program called Dialector. He worked on the project for three years before abandoning it. In 2014, two years after Marker’s death, a small band of programmers and writers, as well as curator and artist Agnès de Cayeux, who knew Marker, res- urrected Dialector and launched it online. This year marks the tenth anniversary of the work being reborn. Dialector functions like a chatbot: users ‘talk’ to it by typing on a keyboard and the program responds with text, sounds or 8-bit images of cats or owls. It’s tempting to think that the inspiration for Dialector may have come from another project he was working on during the same period: a 13-part television series on the influence of Ancient Greek thought and culture called The Owl’s Legacy (1989). Was Marker trying to reinvent the Platonic dialogue for the 21st century?

dialector-visuel-chris-marker-ai-program-installation-view
Chris Marker, Dialector, 1988 (reactivation 2014-15), AI program. Courtesy: Agnès de Cayeux, André Lozano, Annick Rivoire and CC by SA 2.0 Poptronics

The second line of the source code for Dialector reads ‘the second self ’, which serves as the subtitle for the project. The phrase is a noteworthy choice for an artist whose fixation with privacy was arguably as famous as his work. Over the course of his 60-year career, Marker rarely granted interviews and refused to be photographed. In fact, ‘Chris Marker’ is just one of a handful of pseudonyms he went by – none of which can claim to be any more or less real than the name given to him at birth in 1921. This persistent depersonalization and need for anonymity has been interpreted variously as evidence of his media savvy, a habitual form of political protection he picked up during World War II, a personality quirk and so on. To these theories, I wish to add one more: that he didn’t endorse the sanctity of life as a single self.

Marker didn’t pursue life; he pursued lives. And it was in those pursuits that he further dissolved boundaries which distinguished selves from everything else by the sheer attention and discernment he paid to whatever enlivened and interested him, until he was nothing but who he cared about, what he cared for and all those moments he cared enough to record. If life and work evolve like this, the self broadens into an array of selves that find their development and forward momentum in all that emerges from – and survives by – the singular attention and care they give to what enlivens them into lives worth having. Here, death becomes less meaningful, since it is simply not as decisive. What ‘ends’ when living is pursued this way?

dialector-visuel-chris-marker-ai-program
Chris Marker, Dialector, 1988 (reactivation 2014-15), AI program. Courtesy: Agnès de Cayeux, André Lozano, Annick Rivoire and CC by SA 2.0 Poptronics

According to Paul Lafonta, an electronics engineer and computer developer Marker befriended during the time he was making Dialector, Marker did the programming himself. Lafonta said he offered suggestions and advice on how Marker could handle issues like memory allocation, given the hardware limitations of the Apple IIe computer he was using at the time. But Marker wrote Dialector in Applesoft BASIC, a computer programming language that was pre-installed on the Apple II. That a new world demands a new language may be part of the spirit of any utopian project. And there is an interesting undercurrent of utopianism that weaves through Dialector, of which Marker himself may not have been aware. Applesoft BASIC is a variant of the computer language co-invented in 1963 by Dartmouth College mathematicians.

That a new world demands a new language may be part of the spirit of any utopian project.

John G. Kemeny and Thomas E. Kurtz: Beginners’ All-purpose Symbolic Instruction Code (BASIC). Kemeny wanted to create a language that was easier to learn and work with for those who were not mathematicians or computer scientists: namely, the students and faculty in the humanities department. He believed that to really understand what computers at the time were capable of, fresh ways of working and thinking about what was worth being computable were vital. His intuition was that only outsiders produce new ideas.

dialector-agnes-varda-chris-marker-ai-program
Chris Marker, Dialector in conversation(s) with Agnès Varda, 2014, AI program. Courtesy: Agnès de Cayeux, André Lozano, Annick Rivoire and CC by SA 2.0 Poptronics

Kemeny has stated in interviews that, in part, he was inspired to make BASIC while attending a series of lectures given by John von Neumann at Princeton University in 1955. Best known as a mathematician and physicist, Von Neumann also made foundational contributions to what came to be known as cybernetics: the study of the communication and regulation of complex living and machine systems. In his 1955 lectures, Von Neumann detailed his thoughts on the brain as a kind of machine and, by implication, the concept of the possibility of a general purpose ‘thinking’ machine. What we today call artificial intelligence was seeded, in part, by those lectures. I imagine what most inspired Kemeny may have been the path-breaking spirit of Von Neumann’s ideas: the way he developed them across many disciplines of thinking. Could this be why Kemeny believed that it was vital for the nascent field of computer science to be enriched and challenged by different kinds of thinking, if it were to evolve? What would a poet, or an artist, want out of computation?

The source code for Dialector is online, and I read it as if I were peering into Marker’s personal diary. All his loves are there as variables: cat, owl, penguin, lark, emu. The quotes Marker uses for Dialector’s responses are as varied and surprising as they are in his film and video work. Among many others, he drew from William Blake, Bob Dylan, T.S. Eliot, Duke Ellington, Homer, Franz Kafka, Nikita Khrushchev, Annie Lennox, Karl Marx and Star Wars. I especially appreciate the fact that Dialector responds at times with a lyric from iconic 1980s pop band Culture Club: ‘Karma karma karma karma karma chameleon.’

dialector-visuel-chris-marker-ai-program
Chris Marker, Dialector, 1988 (reactivation 2014-15), AI program. Courtesy: Agnès de Cayeux, André Lozano, Annick Rivoire and CC by SA 2.0 Poptronics

The code also reveals how Marker tried to program Dialector with some semblance of intelligence. This was achieved, in part, through pattern recognition. Dialector reads what the user has typed and if, say, the first letter is ‘y’, it assumes the user typed ‘yes’ or some similar sentiment of affirmation, and replies based on this assumption. Marker wrote hundreds of patterns to enable Dialector to essentially guess what the user is saying, and hundreds more responses the program can choose from in order to reply. I suspect the reason Marker abandoned the project in 1989 is because he realized he would need infinitely more than the 10,000 lines of code he had already written to achieve the result he wanted. This approach, known as symbolic programming, explicitly hard-codes the rules a program should follow. In the case of Dialector, this meant having to write out every possible input a user could type in order to have any chance of replying in an intelligent manner. And, since the user could conceivably type anything, Dialector would have to store every possible utterance to respond in any meaningful way.

It would be another 20 years before computation would be capable of achieving what Marker wanted to do with Dialector. I know this because, since 2018, I’ve been trying to achieve something similar. As the founder of the publishing house Badlands Unlimited, I had ambitions of publishing not books, but authors: digital downloads, not unlike e-books, where someone talks to, rather than reads, the works. The story or essay that is the basis of the work would unfold through the literal dialogue between the ‘reader’ and the work itself.

paul-chan-install-image
Paul Chan, Paul’ V.4.3, 2024, installation view. Contemporary Art Museum St. Louis. Courtesy: the artist; photograph: Dusty Kessler

Natural Language Processing (NLP) is the domain of machine intelligence that focuses on speech and text-generation. The path-breaking capabilities of products like ChatGPT are underwritten by advances in NLP, which began in earnest during the mid-to-late 2010s. Only then did computer hardware become powerful enough to do parallel processing (computing many instructions at the same time as opposed to one right after another) at unprecedented speeds and scales. Algorithms emerged that enabled programming frameworks to probabilistically ‘learn’ in unsupervised and non-deterministic ways. And, perhaps most importantly, a critical mass of training material was available for the first time, in the form of user data that social media platforms have been collecting (with or without our consent) since the dawn of Web 2.0.

Badlands went bankrupt before COVID-19 and closed before a single line of code was written for the new kind of book I wanted to publish. No doubt for the best. The technology at the time was not yet usable for someone not already deeply experienced with neural networks. The NLP ‘transformer’ model, which revolutionized how machines grasped and generated text, was just beginning to be deployed. Also, I had no idea what I was doing.

The source code for Dialector is online, and I read it as if I were peering into Marker’s personal diary.

It’s different today. There is something like a ‘Kemeny’ moment happening in NLP. Training and manipulating the Large Language Models at the core of how AI programs understand and produce text data is more feasible, even for those of us without a PhD in computer science. And the models themselves are becoming smaller, which makes them more manageable on consumer hardware. These are just some of the reasons I’m now able to fine-tune and develop my own language models, which underwrite the ‘synthetic portraits’ I’ve been privately creating for my own amusement and research.

paul-chan-install-image
Paul Chan, Paul’ V.4.3, 2024, installation view. Contemporary Art Museum St. Louis. Courtesy: the artist; photograph: Dusty Kessler

I had no idea about Dialector until a few months ago – four years into my curious work in machine intelligence. Finding it online felt prescient and familiar, like running into an old friend on the street after a long day. What struck me most was how Dialector confirmed an insight that came to me working in NLP: the value of the attention paid to a work is heightened by the quality of the boundaries that frame it. Dialector is very limited by today’s standards for chat programs. But this doesn’t diminish its value. The spirit of how it responds to users reminds me of the emails Marker used to send me, before he left this planet for good, destination unknown. Seneca said it best: Desinuit ister non pereunt. These things they end, but they do not go away.

This article first appeared in frieze issue 244 with the headline ‘The Second Self’

Paul Chan’s ‘Breathers’ will be on view at the Contemporary Art Museum St. Louis, Richmond from 8 March until 11 August

Chris Marker's Dialector can be accessed via Poptronics. 

Main image: Chris Marker, Dialector (detail), 1988 (reactivation 2014-15), AI program. Courtesy: Agnès de Cayeux, André Lozano, Annick Rivoire and CC by SA 2.0 Poptronics

Paul Chan is an artist and publisher. His travelling exhibition ‘Paul Chan:Breathers’ is currently on view at the Contemporary Art Museum St. Louis, Missouri.

SHARE THIS