REINCANTAMENTO Drop #0
Discover more from REINCANTAMENTO Drops
Entroducing…
Hey everyone (▰˘◡˘▰)
In a caffeine-induced rush, I finally decided to activate a Substack-offshoot for REINCANTAMENTO this morning. The decision was simple: I’m tired of posting almost exclusively on Instagram or publishing articles in magazines. I want to find a midpoint where more loosely-organized formats can exist, where I can share interesting projects and considerations without the 2’000 characters’ limits or the algorithimic-induced anxiety of disappearing. Moreover, I’ve found myself over-worrying about cosmetic aspects of the project and quietly losing the focus on writing. This is an attempt in reverting the process. This doesn’t mean that @re.incantamento on Instagram will not be active anymore, but probably will lose this unexpected centrality that reached in the past year or so, at least in my mind. I will not state my intentions extensively or anticipate the content of this newsletter because, honestly, I have no clue about what I will do with it. And I enjoy that. I can just say that it will be written in English. Before jumping into today’s topic, a brief introduction to the project, in case somebody found it just through this post (if you already follow us, you can skip it).
What is REINCANTAMENTO?
REINCANTAMENTO is an independent research platform curated by Alessandro Y. Longo. REINCANTAMENTO is born as a series of articles on Medium in April 2020, concerning philosophical approaches to digital technologies, the problem of surveillance, the matter of design, and possible solutions to re-enchant the digital tools of our time. From this crucial philosophical frame, an online meditation was born with a daily activity on Instagram and the release of articles, videos, and more. REINCANTAMENTO is not exclusively a personal project. First, the constant contamination and influence of other minds have been crucial for my research ideas. Secondly, many collaborators helped during the years, and a collective entity, Speculum!, was born from this project.
REINCANTAMENTO wants to contribute to the construction of a common and collective horizon for a new techno-social evolution. To help dissolve the disciplinary boundaries that separate concepts from matter, code from artworks. To connect the dots across the jagged map of the present so that new paths can emerge. Trying to occupy the Insta-space without falling victim to influencers-anxiety and hype cycles. To nurture the hope that new networks and new machines can be born for the people. Alternatively, to fail spectacularly while trying.
Our website / live archive, inspired by the principles of Digital Gardens, is visitable here and here you can find the Instagram page.
Drop #0
In today’s Drop, I want to share an artsy documentary I stumbled upon yesterday night. It’s called, quite poetically I believe, ‘Flowers Blooming Backward Into Noise’ and it’s realized by Eryk Salvaggio, “an interdisciplinary design researcher and new media artist”, according to its website. You can watch it at this link.
The documentary is centered around a critique of the production of AI-generated images, or syntographies, to adopt this up-and-coming term (I personally like it). While you may have heard of these arguments before, the sharp and fast exposition (it’s only 20 minutes long) and the aesthetic vibe, vaguely Adam Curtis-core, make it a must-watch. Most of all, it matches a solid critique with an openness to use that makes it rare in this landscape.
Check it out and then come back to my observations.
Syntographies are the product of Diffusion Models. Diffusion Models are siloed entities, they don’t communicate with reality. They elaborate images through statistical calculations starting from noise: they try to find patterns in pixels based on the categories that they have been thought. Previously, I wrote:
What is the fundamental idea behind diffusion models? A kind of dual motion operates on the image: at first, the image is filled with noise in the first moment of diffusion (forward diffusion); the second moment (reverse diffusion) "cleans" the image of noise and reconstructs it. This process is iterative and occurs through a series of steps so that the model can learn better.
To work, these models need all the images to be precisely categorized: they don’t see and, consequently, they need statistics to navigate all the data they have been fed. They follow the trajectory of the Gaussian Noise, a probability density function, that tells them where are the images they are looking for. A flower generated by a diffusion model is not a flower, Salvaggio argues, but it is the stereotype of a flower, the central distribution of the “flower” category, the platonic idea of a flower.
Salvaggio connects the origins of this “categorization of the world” with the Lombrosian idea of eugenics, of categorizing people according to their external traits. The stereotype of a flower is not harmful but when we move to people we can see how models like DALL-E are extremely biased. I think this makes them very boring and standardized; they have the imagination of a middle-class white man when talking about cultures. This is not new, we know for years how the whole AI endeavor is biased, probably even built-on-bias.
One last point. Towards the end, Salvaggio opens up the possibility for misuse of this model, a truly artistic hacking of these boring, racist categorizing machines. Once we know the weak points and the ways of “reasoning” these tools, we can find their black spots and play with them, make them hallucinate, and lose their petit bourgeois rational facade. He plays with the weaknesses of the models in generating precise images of human hands and looping them within the same idea of Gaussian noise, realizing impossible images. It reminded me of an old Deleuze quote from Difference and Repetition:
Such a disposition is diabolical rather than divine, since the peculiarity of demons is to operate in the intervals between the fields of action of the gods, as if to leap over barriers and fences, bringing confusion to the property.
I like this idea of bringing confusion to these immense properties of data. Artists have to be diabolical with these tools, to know them better than what we are supposed to, and to slide beyond their conventions. A slogan could be: don’t generate beautiful flowers, bring the machine to the delirium.
I hope you liked this Drop #0. I hope I can be consistent and use more Substack. Meanwhile, as the netiquette imposes me, I leave the subscribe button here. Share this with your friend. See you soon ^____^