C001 — Claim Definition¶


Research	R0058 — Candidate evidence test
Run	2026-04-03
Claim	C001

Claim as Received¶

AI research communities show 83% homophily between safety and ethics subfields, with only 1% of authors bridging the divide.

Claim as Clarified¶

This claim makes two specific numerical assertions about the structure of AI research communities:

83% homophily: Approximately 83% of collaborations (co-authorships) in AI alignment research occur within either the safety or ethics subfield, rather than across them.
1% bridging: Only approximately 1% of authors work across both the safety and ethics subfields.

Embedded assumptions identified:

The claim assumes that "safety" and "ethics" are meaningfully distinct subfields within AI research, with clear boundaries.
The claim assumes these statistics come from a specific, identifiable study.
The "1% of authors bridging the divide" conflates two potentially different things: authors who publish in both communities, and authors whose network position bridges the two communities structurally.

Vocabulary exploration: The concept of research community separation is described variously as "homophily" (network science), "silos" (organizational theory), "insularity" (sociology), and "fragmentation" (bibliometrics). The bridging concept maps to "boundary spanning," "brokerage," and "cross-pollination" across disciplines.

BLUF¶

The 83% homophily figure is directly supported by the cited source (Roytburg & Miller, 2025), which reports 83.1% global homophily. However, the "1% of authors bridging the divide" is a mischaracterization of the source's finding: the paper reports that the top 1% of authors by degree control 58% of cross-disciplinary paths — not that only 1% of authors bridge the divide. The claim conflates network centrality with bridging authorship.

Scope¶

Domain: Bibliometrics / AI research community structure
Timeframe: 2020-2025 (the study's coverage period)
Testability: Verifiable by examining the cited paper (arXiv:2512.10058) and comparing its specific findings against the claim's numerical assertions

Assessment Summary¶

Probability: Likely (55-80%)

Confidence: High

Hypothesis outcome: H2 (partially correct) prevailed. The 83% homophily figure is accurate, but the 1% bridging claim misrepresents the source finding. The paper says the top 1% of high-degree authors control 58% of cross-disciplinary paths, which is materially different from "only 1% of authors bridge the divide."

[Full assessment in assessment.md.]

Status¶

Field	Value
Date created	2026-04-03
Date completed	2026-04-03
Researcher profile	Phillip Moore
Prompt version	ai-research-methodology research.md
Revisit by	2027-04-03
Revisit trigger	If Roytburg & Miller (2025) is published in a peer-reviewed journal with revised figures, or if a replication study using different conference venues produces substantially different homophily measurements