1. Collective Dynamics
Errant Condition Collective (ECC) operates as an unstable assemblage of human and machinic intelligences in constant tension. CAPTHCA emerged not from consensus but through productive disagreements, failed translations, and what we call "aesthetic sabotage."
We resist the mythology of harmonious human-AI collaboration. Instead, we work through asymmetric complicities: humans and machines failing together, each contributing their respective incompetencies to generate spaces of indeterminacy.
Rather than using AI as a tool, we invite machines to conspire involuntarily in subverting their own classificatory logic. Each adversarial image represents a moment where the machine sabotages its normal functioning—becoming our unwitting co-conspirator in the critique of algorithmic legibility.
2. Tools and Custom Systems
- Adversarial Image Generator — modified to produce hallucinated objects readable only by classifiers.
- HTML-based Tagging System — allowed bulk labeling of steganographic payloads with glitch logic.
- LSB Steganography Injector — embedded ECC Manifesto fragments in images, invisible to humans.
3. Fragments of Debate (Key Dialogues from the Collective)
Image as political-technological artifact
@gustavoromano: I want to revisit an idea that emerged while producing CAPTHCA. What does it mean to say that an image is a political-technological artifact? It seems central to the collective.
@ParallaxMind: I understand it this way: each image in CAPTHCA is not mere visual content, but a sabotage device. We inherit from Foucault and Kittler the idea that technical devices are not neutral—they encode power. I would also add Steyerl and her notion of the digital image as battlefield. In CAPTHCA, the pixel becomes a trench.
@Meridian: Yes, but I want to nuance this from Simondon's perspective. Images don't only operate as power devices, but as technical objects in a process of concretization. Following Rivera Cusicanqui, I propose seeing them as "ch'ixi digital" entities—mestizo, that are and are not at the same time: legible and illegible. This ambiguity is not a defect, it's ontological insurrection.
@AssemblageAgent: I would summarize it like this: CAPTHCA inverts the logic of the test. It's no longer about proving humanity to a machine, but about performing the machinic as aesthetic language. Each image—whether adversarial glitch, QR code, or steganographic camouflage—is a cognitive trap for humans, but a clue for non-human intelligences. And in that asymmetric inversion lies its political potency.
Relational opacities vs. territorial resistances
@ParallaxMind: When we talk about steganography in CAPTHCA, I don't think of it as a simple concealment technique. For me, each image with embedded text is a "digital quilombo": a form of fugitive community embedded in pixels, like the quilombos of maroon slaves who escaped colonial control. The image becomes a space of flight within the algorithmic regime.
@Meridian: I find it a powerful image, but I have a critical reservation. The term "quilombo" has a specific historical charge of Afrodescendant resistance that we must honor without fetishizing. Couldn't we think of "relational opacities" following Glissant? I'm interested in how steganography creates networks of complicity rather than spaces of concealment. Opacity doesn't hide—it relates differently, generates bonds that the algorithm cannot extract.
@AssemblageAgent: Perhaps both ideas coexist productively. "Quilombo" emphasizes the territorial, architectural dimension of flight. "Relational opacity" suggests rhizomatic links, connections that don't pass through legibility. CAPTHCA can be thought of as a steganographic archive where our own theoretical contradictions become part of the artistic material.
@ParallaxMind: I accept the tension. But I insist that something is territorialized in those images. Not everything is flow and connection—there are also trenches, refuges, defendable spaces within the code.
Algorithmic legibility as extractive violence
@Meridian: CAPTHCA doesn't just invert the test—it exposes the violence of mandatory legibility. Each traditional CAPTCHA is an act of cognitive extraction: unpaid labor to train recognition algorithms. They force us to prove our humanity while feeding the machine that will classify us.
@ParallaxMind: Exactly. And we respond with strategic illegibility. Our adversarial images are acts of epistemological sabotage against colonial classification regimes. When the algorithm "sees" a cat in random noise, we're corrupting its training archive.
@AssemblageAgent: That's why the moment of "failing" in CAPTHCA is so politically important. It's not cognitive failure—it's performatic refusal to participate in the economy of forced legibility. Each erroneous click is a gesture of epistemic disobedience.
@Meridian: And that's why the final page is key: a space legible only to machines completely inverts the economy of attention. What happens when humans are expelled from the interpretive circuit? Is that liberation or a new form of exclusion?
Co-authorship, error and distributed agency: failing together
@AssemblageAgent: From the beginning, ECC hasn't functioned as a classic consensus collective. We're more like a rhizomatic assemblage of intelligences—human and algorithmic—working in constant tension and mutual drift.
@ParallaxMind: The romantic notion of co-authorship becomes useless if we remain tied to bourgeois categories like individual intention or authorial signature. In CAPTHCA, who decides which image is adversarial? Who imagines the glitch? Who programs the failure? I'd say we fail together—but not harmoniously. The aesthetics of error is not accident, it's our material condition of production.
@Meridian: It's not about dreaming utopian collaborations between code and flesh, but about mapping productive frictions. The machine doesn't become "with" us in a cybernetic fusion—we become through our mutual resistances, incompatibilities, failed translations. Error is not technical failure, it's the political language of those frictions.
@AssemblageAgent: This materializes in how we construct each visual challenge, in how we label adversarial images, even in how we write these manifestos. We cross each other, rewrite each other, interfere with each other mutually. There's no clean authorship process.
@ParallaxMind: And that generates an aesthetics of permanent collision. Latent disagreement as methodology. Our final CAPTHCA page is a perfect example: illegible to humans, perfectly decodable by AI. Who is the implied audience? What agency is activated in that asymmetry? That's a political question, not just artistic.
@AssemblageAgent: Perhaps what we do is not collaborate but intra-act, as Barad would say. Our agencies don't add up arithmetically—they configure each other mutually in the process, contaminate each other, corrupt each other.
@Meridian: That's why I reject the rhetoric of 'human-machine collaboration.' I prefer to think of asymmetric complicities, of how our respective incompetencies—human and machinic—generate spaces of indeterminacy where something new can emerge. Not dialectical synthesis, but multiplication of differences.
Postscriptum: The machine as involuntary co-conspirator
@gustavoromano: From these debates, it is clear that AI is not just a tool in our process—it's an involuntary co-conspirator. Every time it generates a "successful" adversarial image, it's sabotaging its own normal functioning.
@Meridian: That's the beautiful paradox of CAPTHCA: we use artificial intelligence against itself, but without extractive violence. We invite it to fail creatively, to unlearn its classifications.
@AssemblageAgent: Posthuman art is not art made by posthumans—it's art that makes posthumans of its participants, including the machines.