R0044/2026-03-29/Q003/SRC02/E01¶
"Bending the Automation Bias Curve" mentions sycophancy alongside automation bias in national security contexts, finding that human factors (experience, knowledge, trust) are the key drivers of over-reliance.
URL: https://academic.oup.com/isq/article/68/2/sqae020/7638566
Extract¶
The paper tested automation bias theories across 9,000 adults in 9 countries. Key finding: "Research on automation bias suggests that humans can often be overconfident in AI, whereas research on algorithm aversion shows that, as the stakes of a decision rise, humans become more cautious about trusting algorithms."
The paper identifies "experiential and attitudinal" factors as key drivers: task difficulty, background knowledge and experience with AI, trust and confidence in AI, and self-confidence. It also identifies a Dunning-Kruger effect where those with the lowest AI experience are slightly more algorithm-averse, then automation bias increases before leveling off at higher knowledge levels.
JUDGMENT: The paper mentions both automation bias and sycophancy but does not systematically map the relationship between them. Sycophancy appears as a passing reference rather than a structured conceptual bridge.
Relevance to Hypotheses¶
| Hypothesis | Relationship | Strength |
|---|---|---|
| H1 | Supports | Both terms appear in the same paper, indicating awareness of both vocabularies |
| H2 | Contradicts | The co-occurrence eliminates complete siloing |
| H3 | Supports | The connection is incidental rather than the focus of the work — exactly what H3 describes |
Context¶
This paper represents the national security research community's awareness of both phenomena, but it frames them primarily through the human factors lens (what drives human over-reliance) rather than the AI safety lens (what makes AI systems sycophantic).