Algorithmic sabotage looks different depending on the industry, the type of software used, and the goals of the workers. 1. The Gig Economy: Manipulating Supply and Demand
Creating "adversarial examples" that allow individuals to remain undetected by automated recognition systems [2]. Disrupting Decision-Making:
Flooding algorithms with garbage or false data to make the resulting model useless or biased. "Cloaking" and "Poisoning" Tools: Tools like Knee et al.'s work on Fawkes Nightshade
Employees may deliberately avoid using sanctioned AI tools, opting for unapproved tools or manual processes to maintain control over their work quality. algorithmic sabotage work
To combat this, warehouse workers practice collective pacing. They intentionally maintain a steady, moderate speed—just high enough to avoid triggering automated disciplinary warnings, but low enough to prevent the algorithm from inflating future quotas. 3. Corporate "Mouse Jiggling" and Engagement Spooking
But there is a darker side. Malicious actors can weaponize algorithmic sabotage:
Grocery cashiers, evaluated on scanning speed, may hold an item over the scanner or manipulate the checkout screen to artificially pause the timer between customers. 2. Data Poisoning y = make_classification(n_samples=1000
alter images in imperceptible ways to prevent AI models from training on them correctly, or to "poison" the model's understanding of a concept [1, 2]. Bot-Powered Noise:
Not all algorithmic sabotage is loud or collective. Some of the most effective acts of resistance are almost invisible, occurring at the individual level. In a foundational theoretical analysis, sabotage is defined as an act of resistance that consists of . In the age of digital capitalism, three distinct types of sabotage have been identified: classic sabotage (the deliberate destruction of machinery), subtle sabotage (the tactical reappropriation of digital tools for alternative purposes), and resistance to techno-science (actively rejecting the cybernetic ideal of always-on, always-connected digital life).
When workers organized against factory owners in the 19th century, they formed unions and went on strike. When platform workers fight back today, they often do so by manipulating the very algorithms that govern them. Researchers at Warwick Business School have extensively documented how Uber drivers have developed sophisticated practices to game the ride-hailing app's algorithmic management. random_state=42) core_model = Sequential([Dense(10
The Invisible Spanner: Understanding Algorithmic Sabotage at Work
X, y = make_classification(n_samples=1000, n_features=20, n_classes=2, random_state=42) core_model = Sequential([Dense(10, activation='relu'), Dense(1, activation='sigmoid')]) core_model.compile(optimizer='adam', loss='binary_crossentropy') core_model.fit(X, y, epochs=5, verbose=0)
