Skip to content

R0044/2026-03-29/Q003/H3

Research R0044 — Expanded Vocabulary Research
Run 2026-03-29
Query Q003
Hypothesis H3

Statement

Partial/emerging bridging exists: some researchers acknowledge both concepts but do not yet draw explicit, systematic connections. The bridging is implicit or incidental rather than the focus of the work.

Status

Current: Supported

This is the best-supported hypothesis. The CSET paper "AI Safety and Automation Bias" brings the two concepts together in its title but appears to address them as related-but-separate concerns rather than mapping them as descriptions of the same underlying phenomenon. The sycophancy research literature (ICLR 2024, arXiv 2025) focuses on RLHF training dynamics without engaging the decades-long automation bias literature. The automation bias systematic review (Springer 2025) examines 35 papers spanning cognitive psychology, human factors, and HCI without referencing sycophancy. The rational analysis paper on sycophantic AI connects to confirmation bias (Wason's work) but not to automation bias specifically.

Supporting Evidence

Evidence Summary
SRC01-E01 CSET paper brings both terms together but appears to treat them as related, not identical
SRC02-E01 Mentions sycophancy alongside automation bias but does not systematically map the relationship
SRC03-E01 Systematic review spans multiple disciplines but does not include sycophancy
SRC04-E01 Connects sycophancy to confirmation bias but not to automation bias

Contradicting Evidence

No evidence directly contradicts H3.

Reasoning

The vocabulary gap persists because the two research communities approach the problem from different directions. Human factors researchers study human vulnerability to automation (automation bias, complacency, overtrust). AI safety researchers study system behavior that exploits this vulnerability (sycophancy, reward hacking, alignment). The CSET paper is the most promising bridge, but the full vocabulary mapping — recognizing that "sycophancy" is the system-side mechanism that produces "automation bias" on the human side — has not been explicitly articulated in any source found.

Relationship to Other Hypotheses

H3 subsumes the partial truth in H1 (some bridging exists) and the important observation in H2 (the communities largely operate independently).