Cognitive barricades

For the past two weeks, we have discussed how the synthetic proletariat is created and stratified in a digital media order led by artificial intelligence (AI) and algorithms. We talked about a stratified synthetic proletariat ranging from ghost workers in Kenya to Reddit moderators, from the “customer-worker” who pays a subscription ransom to access the data they produce, to the academic who silently loses their professional expertise. The situation points to a dispossession operation driven by data-intensive monopoly capital that goes deeper than classic surveillance capitalism. The story did not end there. Because concrete counter-moves against algorithmic exploitation have emerged on a line stretching from the US to Europe, and from creative industries to software communities.

TECHNICAL SABOTAGE: POISONING DATA AND MODEL COLLAPSE

AI systems are statistical and algorithmic structures; they feed on our data and learn from it. Data is the most fragile and fundamental point of the system. Based on this, US artists have actively used tools like Nightshade and Glaze, developed at the University of Chicago, to poison models. While the content remains normal for humans, it turns into corrupting data for the model. In some cases, the model even learned to identify a “cat” as a “car.” Artists, in the words of McQuillan (2022), disseminated corrupt data as a practice of technical self-defense. In the long run, these disruptive actions could cause the model to collapse by drowning in its own synthetic outputs (AI slop) (Jacobsen, 2023).

TEMPORAL SABOTAGE: BREAKING THE SPEED OF THE ALGORITHM

Algorithmic capitalism feeds on real-time speed and interaction. Yet, one of the oldest forms of human social resistance is to slow down work. Content creators rejecting the pressure for daily likes and comments, experts not responding instantly on forums, and behaviors that consciously lower “active user” metrics on platforms reduce speed and profit margins. The mass content strikes, slowdowns, and blackouts on Reddit and Stack Overflow following Reddit's 2023 announcement that it would sell its data to AI companies are aimed at this. These actions have targeted AI's dependence on organic expertise, data, speed, and meaning.

WITHDRAWING EMOTIONAL LABOR, CUTTING OFF THE ORGANIC DATA FLOW

Human emotions and reactions—likes, anger, empathy, and irony—are the “humanization” raw materials the model uses for learning. Consciously not participating in discussions or not reacting can be seen not as passivity, but as a strike of meaning. For example, in the 2023 Hollywood Writers Guild (WGA) strike, writers refusing to provide qualified data by saying “our data cannot be trained without consent” is a collective use of opt-out options against platforms.

DATA UNIONISM: THE COLLECTIVE ARMOR OF EXPERTISE

Individual users are powerless; organized data, however, generates bargaining power. Data cooperatives like MIDATA in Switzerland and the EU-backed DECODE Project are trying to take collective ownership away from tech capital and return it to society. For doctors, lawyers, and academics, data unionism is the unionized protection of cognitive property. While MIDATA is a model where members decide how and to whom data is sold by gathering health data under a cooperative umbrella, the DECODE Project offers an approach where citizen data is managed through collective licenses.

DISTINGUISHING THE HUMAN: EXPOSING SYNTHETIC CONTENT

As the flood of synthetic data makes human production invisible, the discernibility of human labor has become a political demand. Content provenance disclosures and signatures resistant to reverse engineering are tools of resistance. While artists and publishers in the US are pushing for the “human-made content” label to become standard, requirements for labeling synthetic content and transparency in training data have been opened for discussion in the European Union's AI Act drafts.

COGNITIVE SANCTUARIES: NON-ALGORITHMIC SPACES

Non-algorithmic spaces have begun to be created through blogs closed to search engines, archives, private Discord communities, and publishing venues that reject algorithmic indexing.

LEGAL EROSION: SLOWING DOWN THE SYSTEM

Global tech capital loves the narrative of “inevitable necessity” when marketing technology. Law, even if it does not stop the system right now, can make it more difficult. Lawsuits regarding copyright, personal data, professional secrets, and consent violations may not shut down models, but they disrupt training cycles, shape public opinion, and cause financial losses for companies. Union-led AI initiatives in Canada and Europe have shown that law is a constant pressure mechanism by slowing down training cycles.

The items listed above are actions and forms that have “already happened.” And human intellect and creativity will find many more ways to act. All of these are efforts and calls to change the property regime of data. Digital resistance must target not technology and its intelligence, but the information hierarchy and property regime upon which it is built.