R0044/2026-04-01/Q003 — Assessment¶
BLUF¶
No publication was found that explicitly and deliberately bridges the human-factors vocabulary (automation bias, overtrust, complacency) with the AI safety vocabulary (sycophancy, RLHF-induced agreement). However, Ibrahim et al. (2025) come closest — their overreliance framework uses both vocabulary sets in a single analysis, discussing automation bias, sycophancy, trust calibration, and cognitive offloading as interconnected concepts. The CSET brief also bridges the traditions structurally. The vocabulary gap remains real: most researchers operate in one tradition or the other, as exemplified by Malmqvist's sycophancy survey which makes no reference to human factors research at all.
Probability¶
Rating: N/A (open-ended query)
Confidence in assessment: Medium
Confidence rationale: The finding that no explicit bridge exists is supported by comprehensive searching across both traditions. However, the inability to access some PDFs (CSET brief, NIST AI 600-1) and the possibility that conference presentations or working papers may contain bridging work not indexed in web search reduce confidence slightly.
Reasoning Chain¶
-
The query asks specifically whether anyone has explicitly connected "automation bias/overtrust" to "sycophancy." [JUDGMENT]
-
Ibrahim et al. (2025) discuss both automation bias and sycophancy in a unified overreliance framework, alongside trust calibration and cognitive offloading concepts. [SRC01-E01, Medium-High reliability, High relevance]
-
However, Ibrahim et al.'s bridging is functional (they use both vocabularies) rather than deliberate (they do not declare vocabulary bridging as a goal). The paper's purpose is overreliance measurement, not vocabulary mapping. [SRC01-E01, Medium-High reliability, High relevance]
-
The CSET brief places "AI Safety" and "Automation Bias" together in its title and bridges user-level and technical-level factors in its framework. The same author (Lauren Kahn) co-authored military AI research, demonstrating individual-level bridging. [SRC02-E01, High reliability, Medium-High relevance]
-
Malmqvist's sycophancy survey exemplifies the absence of bridging: it treats sycophancy as a purely technical LLM problem with no reference to human factors, aviation, healthcare, or regulated-industry research. [SRC03-E01, Medium reliability, Medium relevance]
-
JUDGMENT: The bridge is emerging but not yet formally constructed. Individual researchers are working across traditions (Kahn, Ibrahim et al.), but no publication provides a deliberate, formal vocabulary mapping. This gap represents an opportunity for original contribution. [JUDGMENT]
Evidence Base Summary¶
| Source | Description | Reliability | Relevance | Key Finding |
|---|---|---|---|---|
| SRC01 | Ibrahim et al. overreliance | Medium-High | High | Closest to vocabulary bridging; uses both traditions functionally |
| SRC02 | CSET automation bias | High | Medium-High | Title bridges traditions; shared authorship demonstrates individual bridging |
| SRC03 | Malmqvist sycophancy survey | Medium | Medium | Counterexample: no human factors connection |
Collection Synthesis¶
| Dimension | Assessment |
|---|---|
| Evidence quality | Medium — limited number of directly relevant sources |
| Source agreement | Medium — sources show a spectrum from bridging (Ibrahim) to siloed (Malmqvist) |
| Source independence | High — independent research teams |
| Outliers | Malmqvist is the outlier (complete silo) vs. Ibrahim (partial bridge) |
Detail¶
The evidence reveals a clear pattern: the vocabulary bridge is being constructed from the human-factors/policy side (CSET, Ibrahim) rather than from the AI safety/technical side (Malmqvist). This makes sense — human factors researchers studying AI naturally encounter AI safety concepts, while AI safety researchers studying model behavior may not consult decades of human factors literature.
Gaps¶
| Missing Evidence | Impact on Assessment |
|---|---|
| Conference presentations or working groups | May contain bridging work not indexed in web search |
| CSET brief full text | May contain more explicit vocabulary bridging than title suggests |
| Regulated-industry internal working papers | May contain bridging in classified/restricted contexts |
Researcher Bias Check¶
Declared biases: None declared.
Influence assessment: The query itself frames the absence of bridging as a gap, which may have biased toward confirming that absence. However, extensive searching for bridging papers was conducted, and Ibrahim et al. was surfaced as a partial counterexample.
Cross-References¶
| Entity | ID | File |
|---|---|---|
| Hypotheses | H1, H2, H3 | hypotheses/ |
| Sources | SRC01, SRC02, SRC03 | sources/ |
| ACH Matrix | — | ach-matrix.md |
| Self-Audit | — | self-audit.md |