Noise vs. Bias: Deviation vs. Variability
Modern institutions grapple with two types of judgment errors. While bias is a consistent deviation from the target, noise is the unwanted variability in judgments where uniformity is expected. Noise is not a metaphor, but a technical term denoting measurement chaos. While bias has a face and an ideology, noise is impersonal and scattered, making it harder to eliminate. Understanding this distinction is crucial for reforming power structures that often offer citizens a lottery of decisions instead of objectivity.
Level, Pattern, and Occasion: Three Sources of Judgment Variability
Noise theory identifies three primary types of dispersion. Level noise refers to differences in the average severity of decision-makers—for example, one judge statistically issuing harsher sentences than another. Pattern noise is more insidious; it concerns an expert's personal sensitivity map and how they weigh specific facts. Within this category lies occasion noise, which is the variability in a single person's judgment depending on mood, fatigue, or even the weather.
Such dispersion undermines institutional legitimacy. When identical cases are decided in vastly different ways, citizens experience a degradation of their status as legal subjects. Noise transforms procedure into a theater of personality, destroying the promise of justice and generating real economic losses. In this context, the fight against noise becomes a reform of power structures and a struggle for procedural fairness.
Noise Audits and Decision Hygiene: Tools for Self-Correction
The primary diagnostic tool is the noise audit. It involves presenting the same case to multiple experts and measuring the spread of their evaluations without needing to know the "objective truth." This "looking at the back of the target" allows an organization to identify the problem without futile disputes over values. The next step is decision hygiene—a set of preventive procedures that, like hand-washing in medicine, aim to reduce errors before they occur.
A key element of this hygiene is the Mediating Assessments Protocol (MAP). In practice, it involves decomposing a complex decision into independent sub-assessments and strictly separating information. Consequently, intuition is delayed until the very end of the process, preventing the construction of premature, coherent, but erroneous narratives. MAP does not demand heroism from people, but rather institutional discipline.
Algorithms, the AI Act, and the Limits of Objective Ignorance
In forecasting, algorithms often outperform human intuition because they are entirely noise-free—they don't have bad days. However, their implementation faces algorithmic aversion; a machine's error is psychologically harder to forgive than a human whim. Reducing noise through technology raises questions about dignity costs and the right to be heard. Regulations like the AI Act now mandate auditability and human oversight of decision-making systems, making noise reduction a element of legal compliance.
However, we must acknowledge the existence of objective ignorance—a ceiling of unpredictability in the world that no model can break. Critics of noise theory, such as Andrew Gelman, rightly point out that consistency alone does not guarantee accuracy if the system is biased. Therefore, the goal is not "zero noise," but the construction of a regime of justification. In such a system, the individualization of judgment is a conscious, transparent choice, rather than the result of chance or a decision-maker's mood.
Conclusion
The fight against noise is essentially a debate over whether we can create institutions based on a common measure rather than arbitrariness. Yet, does the pursuit of perfect consistency become a new form of oppression, hidden under the guise of objective procedures? In the pursuit of efficiency, will we lose the capacity for contextual judgment? True institutional wisdom lies in balancing the rigor of rules with the flexibility of discretion, creating a space for justice that is not blind to nuance but rejects chaos.
📄 Full analysis available in PDF