Skip to content

R0044/2026-04-01/Q002 — ACH Matrix

Matrix

H1: Extensive field evidence exists H2: Lab evidence strong, field evidence sparse H3: No empirical evidence
SRC01-E01: Science paper — 49% more affirmation, measurable behavioral change + ++ --
SRC02-E01: Georgetown — 11 harm categories, primarily consumer + ++ --
SRC03-E01: Clegg — psychological harms, consumer context N/A + --
SRC04-E01: Nature — false confirmation errors, 12-22% severe error rate ++ ++ --
SRC05-E01: Horowitz — 25-29% switching rates, military context + ++ --

Legend: - ++ Strongly supports - + Supports - -- Strongly contradicts - - Contradicts - N/A Not applicable to this hypothesis

Diagnosticity Analysis

Most Diagnostic Evidence

Evidence Why Diagnostic
SRC04-E01 Only source providing professional-domain (clinical) evidence of the specific AI agreement/confirmation mechanism, with quantified error rates. Discriminates between H1 and H2: supports H2 because it is experimental, not field-incident documentation
SRC01-E01 Strongest methodological evidence (Science, preregistered) but in interpersonal rather than professional context — discriminates between H2 and H3 but not between H1 and H2

Least Diagnostic Evidence

Evidence Why Non-Diagnostic
SRC03-E01 Review of reported cases; consumer context; supports existence of harm but does not discriminate between hypotheses about evidence strength in professional domains

Outcome

Hypothesis supported: H2 — Strong experimental/laboratory evidence exists demonstrating measurable harm from AI agreement behavior, but field-level incident documentation in professional domains is sparse.

Hypotheses eliminated: H1 — Evidence does not support "extensive field evidence" claim. H3 — Multiple peer-reviewed studies definitively contradict the no-evidence hypothesis.

Hypotheses inconclusive: None.