The Myth of Neutral Tools

The neutrality claim is the oldest trick in the empire of systems. Every generation builds a tool it swears is objective, then watches that tool reproduce the same hierarchies it claimed to transcend. IQ tests, maps, zoning laws, each was marketed as a mirror of truth. Each quietly drew its borders around who counts. AI is only the latest version of that lie. The Politics Hiding in the Code Every dataset, every prompt, every default is a policy decision disguised as code. To build an algorithm is to decide what matters, what counts as “normal,” what outcomes are worth optimizing. None of these are neutral acts; they are moral and political ones. When a platform says it is optimizing for “engagement,” it is not describing a technical process. It is choosing dopamine over democracy. The Original Sin of Data AI systems are trained on human history, and human history is anything but impartial. The “original sin” of training data is that it reflects centuries of structural bias. When machines learn from the world as it is, they inevitably learn to preserve the world as it was. Hiring algorithms that penalize women because past high performers were men. Predictive policing tools that send more officers to neighborhoods they already over-police. Loan systems that punish poverty while calling it “risk assessment.” The bias is not a bug, it is the memory of injustice written in code. Optimization as Ideology Every optimization target carries an agenda. If you optimize for engagement, you get outrage. If you optimize for efficiency, you get precarity. If you optimize for growth, you get extraction. The parameters may look mathematical, but they are moral statements about what a good society looks like, and whose comfort is worth the collateral damage. The False Promise of Fairness Attempts to “debias” AI often reinforce the very systems they claim to fix. A fairness algorithm that equalizes outcomes across groups might do so by flattening real differences or masking…