Scientists propose strict guardrails for using generative AI in assessing pain in older adults with cognitive impairment

Pain assessment in older adults with cognitive impairment remains one of the most persistent challenges in geriatric care, particularly when changes in communication make self-report difficult. A new Perspective article argues that generative artificial intelligence, including large language models, could help clinicians organize and review the scattered information on which pain judgments depend, but only within tightly bounded, clinician-governed limits. The authors stress that AI should never independently detect, diagnose, or determine treatment for pain.

The article, published in Frontiers in Pain Research, proposes a person-centered framework for the bounded use of generative AI in pain assessment. According to the authors, validated observational tools, caregiver and staff knowledge, clinical examination, and repeated within-person assessment remain the foundation of care. The problem they identify is not a shortage of assessment methods but the nature of the information those methods produce: it may be incomplete, conflicting, or documented across different settings, making synthesis burdensome for clinicians.

The framework organizes seven testable use cases under three core functions: capacity-sensitive elicitation, source-attributed synthesis, and longitudinal support and communication. Under these functions, the authors outline potential applications including adaptive pain interviews, identification of possible pain-related information in everyday language, structuring of behavioral observations, multi-informant synthesis, synthesis of validated nonverbal indicators, longitudinal pain-related diaries, and caregiver communication support.

Each application, the authors write, requires clear limits, source identification, communication of uncertainty, consideration of alternative causes of distress, and mandatory clinician verification before any clinical action or documentation. This means an AI system might help structure a behavioral observation or assemble input from family members and care staff, but a clinician would need to verify the output before it could inform care decisions or enter the medical record.

The article also examines a set of risks that the authors argue must be addressed before any deployment. These include suggestibility, in which the way a question is framed shapes the AI’s response; bias in model outputs; medication effects; privacy concerns; alert fatigue among clinicians; added workload; and overreliance on AI-generated outputs. The authors frame these as reasons for caution rather than reasons for outright rejection, positioning the technology as a potential support layer rather than a replacement for existing practice.

To move from proposal to evidence, the authors propose a five-step research pathway. It begins with co-design, followed by technical testing, clinical validation, evaluation across populations and care settings, and risk-based governance.

The authors set a demanding condition for adoption. Generative AI should be adopted, they argue, only if prospective studies demonstrate meaningful value beyond validated assessment methods and usual clinical workflows, and only if that value comes without reducing patient choice, replacing clinical judgment, or increasing workload. This framing places the burden of proof on future research rather than on current enthusiasm for the technology.

As a Perspective article, the work reports no new clinical data or trial results; its contribution is a proposed structure for future research and governance. The authors are affiliated with Northern Illinois University, Illinois State University, Emory University, and the University of Florida.

Subject of Research: A proposed framework for the bounded, clinician-governed use of generative AI in pain assessment for older adults with cognitive impairment.

Article Title: Generative AI as a bounded elicitation and evidence-synthesis layer for pain assessment in older adults with cognitive impairment

Article References: Saravanan, A., Lewis, W., Bai, J., & Booker, S. (2026). Generative AI as a bounded elicitation and evidence-synthesis layer for pain assessment in older adults with cognitive impairment. Frontiers in Pain Research, 7, Article 1922289. https://doi.org/10.3389/fpain.2026.1922289

Image Credits: AI Generated

DOI: 10.3389/fpain.2026.1922289

Keywords: generative AI, pain assessment, cognitive impairment, dementia, older adults, large language models, clinician governance, evidence synthesis, geriatric pain, research framework, caregiver communication, Frontiers in Pain Research