Whose Alignment? Comparing LLM Process Alignment Across Diverse Organizational Decision Contexts
For AI alignment researchers and practitioners, this paper demonstrates that output agreement alone is insufficient for evaluating alignment; process-level measurement is necessary, especially in pluralistic or contested organizational contexts.
The authors argue that aligning LLMs with organizational decision-making is a pluralistic challenge, not a single-target problem. They measure process alignment (how models weight information) in two domains: for ECHR decisions, process alignment strongly predicts output accuracy (r=0.85), but for German credit decisions, the relationship collapses (r=0.15), revealing that in contested domains, high process alignment is neither achievable nor unconditionally desirable.
Aligning AI systems with organizational decision-making is typically framed as a single-target problem: make the model behave like the organization. We argue this framing obscures a deeper pluralistic challenge. We rely on a decision-policy capturing method to measure process alignment: whether an LLM weights information as the organization does, not merely whether it reaches the same conclusions. Applying this method to ECHR Article 6 decisions, process alignment strongly predicts output accuracy (r = 0.85, p < .001) and externalization substantially improves alignment for poorly-aligned models. Applying it to German consumer credit decisions, this relationship collapses (r = 0.15, p = .60): interventions produce inconsistent effects and the benchmark encodes potentially discriminatory historical patterns. This contrast is itself a pluralistic alignment finding: in contested domains, high process alignment is neither achievable via externalization nor unconditionally desirable. Output agreement alone cannot distinguish a model that has internalized an organizational policy from one that merely approximates its outcomes; process-level measurement is a necessary component of any pluralistic alignment evaluation.