R0058/2026-04-03/C001/H3¶
Statement¶
The claim is materially wrong: both the 83% homophily figure and the 1% bridging figure are inaccurate or misleading, and the underlying premise of a sharp safety-ethics divide is overstated.
Status¶
Current: Eliminated
Supporting Evidence¶
| Evidence | Summary |
|---|---|
| SRC02-E01 | A separate bibliometric study finds 94% of AI ethics institutions in the largest connected component, suggesting a more integrated research ecosystem than the claim implies |
| SRC04-E01 | Active workshops bridging fairness, ethics, and safety at major venues suggest meaningful cross-pollination exists |
Contradicting Evidence¶
| Evidence | Summary |
|---|---|
| SRC01-E01 | The 83.1% homophily figure is directly reported by the primary source with robust methodology (6,442 papers, null model comparisons, p<0.001) |
| SRC01-E02 | Bridge concentration statistics are well-documented with specific measurements |
| SRC03-E01 | Independent scoping review finds safety underrepresented in ethics-focused venues, consistent with community separation |
Reasoning¶
While some evidence suggests the broader AI research ecosystem is more connected than the claim implies (Qiu et al.'s 94% connected component), the specific safety-ethics divide measured by Roytburg & Miller is well-evidenced with rigorous methodology. The 83.1% homophily figure is supported by null model comparisons and reachability analysis. H3 is eliminated because the core phenomenon is real, even though the "1% bridging" characterization is inaccurate.
Relationship to Other Hypotheses¶
H3 would require both components of the claim to be wrong. Since the 83.1% homophily figure is confirmed by the primary source with strong methodology, H3 cannot be sustained. The evidence instead supports H2 (partially correct).