神経心理学における大規模言語モデル：新たな応用と倫理的考察

Objective: There is immense excitement and fear regarding the potential of artificial intelligence, particularly large language models (LLMs), to transform healthcare services. In this paper, we aim to provide a future-oriented commentary on how neuropsychologists might consider integrating LLMs into clinical practice ethically, safely, and effectively. Methods: In Part 1, we provide a narrative review of the emerging applications of generative transformer-based LLMs in neuropsychological assessment, including how these technologies may support clinicians with data collection, clinical decision making, and documentation. In Part 2, we analyze the key ethical considerations using the framework of the American Psychological Association (APA) Ethical Principles of Psychologists and Code of Conduct (2017) and Ethical Guidance for AI in the Professional Practice of Health Service Psychology (2025) to formulate recommendations for future research, policy, and clinical integration. Conclusions: LLMs display promise for enhancing neuropsychological practice along a number of lines, such as extracting data from medical records and natural communication, automating test scoring, supporting differential diagnosis and treatment planning, developing test items, and generating text summaries of interviews and empirical literature. However, the rapid progress in LLMs over recent years has left little time for regulation, ethical discourse, and sufficient validation. Key barriers include privacy concerns, risk of bias, limited model transparency, threats to test security, and insufficient research supporting the validity and reliability across specialized tasks. We advise neuropsychologists to help shape regulation, lead empirical validation efforts, and augment educational resources to effectively integrate LLMs into clinical practice while maintaining high ethical standards.