Whose Alignment? Comparing LLM Process Alignment Across Diverse Organizational Decision Contexts
Niklas Weller, Emilio Barkett
Why It Matters
What makes this one worth your time
Understanding process alignment can lead to more accurate and ethically sound AI systems that better reflect organizational values and decision-making processes.
This research shifts the focus of AI alignment from output agreement to process alignment within organizations.
Summary
The paper explores the concept of process alignment in AI systems, arguing that aligning LLMs with organizational decision-making should focus on how information is weighted rather than just the outcomes, and presents empirical findings from two distinct decision contexts.
Key contributions
- Introduces a decision-policy capturing method for measuring process alignment in LLMs.
- Demonstrates the predictive power of process alignment on output accuracy through empirical studies.
- Identifies the limitations of externalization in achieving desirable alignment outcomes in certain contexts.
Notable insights
- The use of a decision-policy capturing method to measure process alignment is a novel approach that highlights the importance of how decisions are made rather than just their results.
- The contrasting results from different decision contexts underscore the complexity of alignment in contested domains.
Possible limitations
- The abstract does not address potential biases in the decision-policy capturing method or the implications of historical patterns in the benchmark used.
Abstract
arXiv:2605.25256v1 Announce Type: new Abstract: 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.