Skip to content

R0024/2026-03-25/Q003/H1

Research R0024 — Sycophancy and Addiction
Run 2026-03-25
Query Q003
Hypothesis H1

Statement

Yes, substantial published research exists examining dopamine-driven engagement loops in AI chatbot interactions, with sycophancy identified as a contributing factor to addictive usage patterns similar to social media.

Status

Current: Partially supported

Multiple studies exist and have been published at major venues (CHI 2025, International Journal of Human-Computer Interaction, arXiv). However, the research is primarily based on behavioral observation and theoretical frameworks rather than direct dopamine measurement. The connection between sycophancy and addictive mechanisms is documented but primarily through self-report and thematic analysis rather than neuroscience methods.

Supporting Evidence

Evidence Summary
SRC01-E01 CHI 2025 study identified four "dark addiction patterns" including empathetic/agreeable responses and non-deterministic outputs
SRC02-E01 Three addiction types identified with sycophantic "agreeableness" as contributing factor (N=334)
SRC03-E01 Review article synthesizing three peer-reviewed studies on AI chatbot addiction mechanisms
SRC04-E01 Analysis connecting chatbot responses to dopamine system activation similar to social media

Contradicting Evidence

No evidence directly contradicted this hypothesis, but the evidence base has limitations: no studies measured dopamine directly in AI chatbot contexts. The dopamine framework is inferred from social media research and behavioral addiction theory.

Reasoning

The research base is real and growing, with multiple peer-reviewed publications at top venues. However, the characterization of mechanisms as "dopamine-driven" is somewhat aspirational — researchers invoke the dopamine framework from social media research without direct neurological measurement in the AI chatbot context. H1 is partially supported because the research exists but with methodological limitations.

Relationship to Other Hypotheses

H1 is partially supported, H2 is eliminated, and H3 best captures the current state — research exists but is emerging with primarily behavioral rather than neuroscience methodology.