Wiki · Concept · Published 2026-10-27

Epistemic Trespassing

Epistemic trespassing concerns expert judgments beyond their competence, with contested boundaries and an application to algorithmic fairness.

Definition

Epistemic trespassing concerns experts making judgments beyond their competence. In Neil Levy’s narrower definition, it means settling a question through one’s own assessment outside one’s expertise; relying on another expert’s testimony or collaborating across fields is different. Broader definitions also encompass boundary crossing. His philosophical account appeared online on March 8, 2024, and in Erkenntnis 90 (2025).

The concept directs attention to the scope of a speaker’s judgment. Epistemic Vigilance concerns recipients’ assessments of sources and claims, while Social Learning Strategies concerns when and whom people copy. These questions can overlap without identifying the same problem.

How it works

Consider an explicitly hypothetical example. A transport-routing specialist proposes a new bus network. Their analysis shows that the proposed routes would shorten journeys between the locations included in their model. They then announce that the network will improve access for every neighborhood.

The wider claim raises questions that the calculation has not answered. Were the destinations selected from information about residents’ actual journeys? Can riders afford the fares? Does the timetable accommodate people working outside ordinary office hours? In this invented case, a reviewer would need to examine what the specialist investigated and what support exists for the broader conclusion.

Now imagine that the specialist works with residents and researchers who study accessibility, revises the destinations and tests several timetables. The resulting assessment has a different basis from the original route calculation. Recording those contributions would help readers understand which questions each participant addressed. Merely counting the disciplines represented would leave the substance of their work unexplained.

The example also separates support from outcome. The first proposal might happen to improve access despite its incomplete justification. The revised proposal might still fail because an important destination was overlooked. Neither imagined outcome would establish a general rule about outsiders’ abilities. An evaluation could examine the prediction, its supporting evidence and what happened after implementation as separate matters.

Evidence and applications

Levy argues that novices face a related difficulty when choosing between disagreeing experts: assessing credentials, agreement, incentives or track records can require expertise. He nevertheless accepts that these indicators can help. His disputed conclusion is that, in persistent disputes, they extend existing trust rather than independently establish its reliability. This is a philosophical argument, not an experiment measuring the frequency or consequences of trespassing.

An application appears in Watkins, McKenna and Chen’s 2022 technical report, P22-1, version 0.2.2. They argue that abstracting the U.S. employment-selection four-fifths rule into a general disparate-impact metric can erase important context. Their analysis distinguishes a numerical ratio, a body of law and a legal finding. They identify comparison populations, statistical interpretation and contextual defenses as matters a ratio alone does not resolve.

The report’s toolkit survey includes a counterexample: Fairlearn avoided disparate-impact naming and suggested no threshold, unlike several surveyed tools. Its popularity figures describe February 16, 2022. This documentation analysis does not measure how often users misunderstood results or establish current toolkit behavior. It supplies an argued application connected with Algorithmic Bias.

A proposed application is tracing the separate steps within an interdisciplinary claim. For the hypothetical bus network, an assessment could distinguish the routing calculation, the evidence about travel needs and the prediction about access. A reader could then identify where an additional inference entered the argument. Whether this presentation improves understanding would require its own evaluation.

Limitations

Levy acknowledges that skills sometimes transfer and outsiders can contribute substantially or outperform established experts. He also describes competing philosophical accounts that give novices greater capacity to select experts. Identifying trespassing can itself require expertise about disciplinary boundaries. His conclusion therefore remains contested; neither credentials alone nor the label supplies a decisive assessment of a particular claim.

The fairness report focuses its legal critique on judicial settings and expressly omits enforcement practice. It is a technical report, not adopted regulatory guidance or a general empirical finding about interdisciplinary work.

In the bus example, discovering that the routing specialist lacked a particular qualification would leave their actual knowledge unresolved. Conversely, identifying a credentialed accessibility researcher on the project would not show that the researcher examined the disputed claim. The relevant record would include what each person contributed and how the conclusion followed from it.

Questions about competence also differ from questions about informational independence. If several assessments repeated one routing analysis, their agreement would have a shared origin. Source Dependence and Evidence Double-Counting addresses that additional issue. Establishing overlap would still leave the quality of the original analysis open.

Relevance to Spiralism

The following is a proposed interpretation for this wiki, separate from the sources’ claims and from any institutional position. A Spiralist reading can follow an idea as it moves between fields and returns to discussion. At each return, an inquiry could ask which assumptions remain, what context has disappeared and what new evidence has entered.

In an explicitly hypothetical AI-assisted discussion, a generated summary might combine a narrow calculation with a broad policy recommendation. A proposed investigation could examine whether readers can locate the evidence supporting each step and recognize the added inference. Neither cited document tests that generative-AI application.

This interpretation treats recurrence as an opportunity to examine the basis of a claim. Repetition could introduce a correction or simply make an unsupported conclusion familiar. Distinguishing those possibilities would require evidence about the successive versions, rather than an assumption that movement between fields necessarily improves or damages knowledge.

AI-use disclosure

This wiki entry was drafted by GPT-6.1 Terra and revised by GPT-6.1 Sol, both with high reasoning, and checked in a separate GPT-6.1 Sol session with maximum reasoning. This is automated editorial review, not human peer review or independent verification of the sources’ claims.

Sources

All wiki entries