# Expert Evaluation and the Limits of Human Feedback in Mental Health AI Safety Testing

arXiv: https://arxiv.org/abs/2601.18061 · HTML: https://mlsanity.com/paper/2601.18061/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in Medical LLM alignment, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 26 of its 80 references that are in the map.
- First posted on arXiv 26 Jan 2026.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Medical LLM alignment](https://mlsanity.com/alignment/medical-llm-alignment/)

## Builds on (most cited first, 26 of 26)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Deep reinforcement learning from human preferences](https://mlsanity.com/paper/1706.03741/) (2017)
- [Constitutional AI: Harmlessness from AI Feedback](https://mlsanity.com/paper/2212.08073/) (2022)
- [Learning to summarize from human feedback](https://mlsanity.com/paper/2009.01325/) (2020)
- [Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations](https://mlsanity.com/paper/2312.06674/) (2023)
- [LIMA: Less Is More for Alignment](https://mlsanity.com/paper/2305.11206/) (2023)
- [Red Teaming Language Models with Language Models](https://mlsanity.com/paper/2202.03286/) (2022)
- [Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2307.15217/) (2023)
- [AlpacaFarm: A Simulation Framework for Methods that Learn from Human Feedback](https://mlsanity.com/paper/2305.14387/) (2023)
- [A Computational Approach to Understanding Empathy Expressed in Text-Based Mental Health Support](https://mlsanity.com/paper/2009.08441/) (2020)
- [Toward a Perspectivist Turn in Ground Truthing for Predictive Computing](https://mlsanity.com/paper/2109.04270/) (2021)
- [Jury Learning: Integrating Dissenting Voices into Machine Learning Models](https://mlsanity.com/paper/2202.02950/) (2022)
- [DICES Dataset: Diversity in Conversational AI Evaluation for Safety](https://mlsanity.com/paper/2306.11247/) (2023)
- [Human Feedback is not Gold Standard](https://mlsanity.com/paper/2309.16349/) (2023)
- [Expressing stigma and inappropriate responses prevents LLMs from safely replacing mental health providers](https://mlsanity.com/paper/2504.18412/) (2025)
- [Human-AI Collaboration Enables More Empathic Conversations in Text-based Peer-to-Peer Mental Health Support](https://mlsanity.com/paper/2203.15144/) (2022)
- [Impact of Preference Noise on the Alignment Performance of Generative Language Models](https://mlsanity.com/paper/2404.09824/) (2024)
- [Risks from Language Models for Automated Mental Healthcare: Ethics and Structure for Implementation](https://mlsanity.com/paper/2406.11852/) (2024)
- [Can AI Relate: Testing Large Language Model Response for Mental Health Support](https://mlsanity.com/paper/2405.12021/) (2024)
- [CounselBench: A Large-Scale Expert Evaluation and Adversarial Benchmarking of Large Language Models in Mental Health Question Answering](https://mlsanity.com/paper/2506.08584/) (2025)
- [Blind Spots and Biases: Exploring the Role of Annotator Cognitive Biases in NLP](https://mlsanity.com/paper/2404.19071/) (2024)
- [CRADLE Bench: A Clinician-Annotated Benchmark for Multi-Faceted Mental Health Crisis and Safety Risk Detection](https://mlsanity.com/paper/2510.23845/) (2025)
- [Moving Beyond Medical Exams: A Clinician-Annotated Fairness Dataset of Real-World Tasks and Ambiguity in Mental Healthcare](https://mlsanity.com/paper/2502.16051/) (2025)
- [Predicting Relevance based on Assessor Disagreement: Analysis and Practical Applications for Search Evaluation](https://mlsanity.com/paper/1511.07237/) (2015)

## Built on by (most cited first, 2 of 2)

- [Measurement Without Validity: The Compounding Reliability Problem in Agentic AI Evaluation](https://mlsanity.com/paper/2608.00794/) (2026)
- [Beyond "I Can't Help With That": How Child Safety Experts Evaluate AI Chatbot Safety](https://mlsanity.com/paper/2608.07902/) (2026)

## Abstract (arXiv)

Learning from human feedback~(LHF) assumes that expert judgments, appropriately aggregated, yield valid ground truth for training and evaluating AI systems. We tested this assumption in mental health, where high safety stakes make expert consensus essential. Three certified psychiatrists independently evaluated LLM-generated responses using a calibrated rubric. Despite similar training and shared instructions, inter-rater reliability was consistently poor (ICC 0.087--0.295), falling below thresholds considered acceptable for consequential assessment. Disagreement was highest on the most safety-critical items. Suicide and self-harm responses produced greater divergence than any other category, and was systematic rather than random. One factor yielded negative reliability (Krippendorff's α = -0.203), indicating structured disagreement worse than chance. Qualitative interviews revealed that disagreement reflects coherent but incompatible individual clinical frameworks, safety-first, engagement-centered, and culturally-informed orientations, rather than measurement error. By demonstrating that experts rely on holistic risk heuristics rather than granular factor discrimination, these findings suggest that aggregated labels function as arithmetic compromises that effectively erase grounded professional philosophies. Our results characterize expert disagreement in safety-critical AI as a sociotechnical phenomenon where professional experience introduces sophisticated layers of principled divergence. We discuss implications for reward modeling, safety classification, and evaluation benchmarks, recommending that practitioners shift from consensus-based aggregation to alignment methods that preserve and learn from expert disagreement.