E01¶


Research	R0058 — Candidate evidence test
Run	2026-04-03
Claim	C001
Source	SRC03
Evidence	SRC03-E01
Type	Analytical

Safety underrepresented in AIES/FAccT venues; topic modeling shows spatial separation between technical and ethical themes

URL: https://arxiv.org/html/2510.21293v2

Extract¶

Key findings from the scoping review:

"System safety does not appear as a prominent measure for trustworthiness" in AIES and FAccT articles, despite being designated in EU AI Act and ISO/NIST guidelines
BERTopic analysis reveals distinct topic clusters where "fairness," "transparency," and "ethics" cluster together, while "explanations" maintains "considerable distance" from ethical themes
Current research emphasizes "technical attributes such as reliability...while overlooking the sociotechnical dimensions"
The authors note this separation suggests safety and ethics communities may pursue "parallel rather than integrated agendas"

Relevance to Hypotheses¶

Hypothesis	Relationship	Strength
H1	Supports	The underrepresentation of safety in ethics venues is consistent with the community divide described in the claim
H2	Supports	Consistent with the divide being real while not directly confirming the specific numbers
H3	Contradicts	The independent finding of topic separation in ethics venues makes it harder to argue the divide does not exist

Context¶

This study examines content themes rather than authorship networks. The finding that safety is underrepresented at AIES/FAccT is consistent with Roytburg & Miller's authorship-based homophily measurement, but provides a complementary (content-based) rather than duplicate (network-based) confirmation.

Notes¶

The venue selection (AIES and FAccT) inherently biases toward ethics content, so the absence of safety themes may partly reflect editorial scope rather than community separation. However, the fact that even these interdisciplinary venues do not substantially feature safety research supports the broader claim of community divide.