Generation of traditional Chinese medicine prescription driven by generative artificial intelligence GPT-4
- VernacularTitle:生成式人工智能GPT-4驱动的中药处方生成研究
- Author:
Qitao CHEN
1
;
Jingwen NI
1
;
Jun XU
1
;
Xiaohan GAO
1
;
Lizhen XIA
1
Author Information
1. Dept. of Pharmacy,Sanming Integrated Medicine Hospital,Fujian Sanming 365001,China
- Publication Type:Journal Article
- Keywords:
GPT-4;
traditional Chinese medicine prescription;
generative artificial intelligence
- From:
China Pharmacy
2023;34(23):2825-2828
- CountryChina
- Language:Chinese
-
Abstract:
OBJECTIVE To evaluate the safety and suitability of traditional Chinese medicine prescriptions generated by generative artificial intelligence (AIGC), and to provide research ideas for empowering the traditional Chinese medicine industry with AIGC. METHODS Using the 2020 edition of Chinese Pharmacopoeia and the 5th edition of Traditional Chinese Medicine as corpus, GPT-4 and the real-time networking model developed based on GPT-4 (referred to as the “networking model”) were used for deep learning. The clinical cases included in the consensus of traditional Chinese medicine experts in recent years were extracted manually to regenerate prescriptions based on diagnosis using the GPT-4 model and networking model; traditional Chinese medicine experts conducted blind evaluation and scoring of GPT-4 generated prescriptions, networking model generated prescriptions, and expert consensus prescriptions. At the same time, Turing testing was used to evaluate whether the GPT-4 model and networking model had the same ability as human intelligence. RESULTS The average score of traditional Chinese medicine prescriptions generated by the GPT-4 model showed no statistically significant difference compared to manual prescriptions (P>0.05), while the average score of prescriptions generated by the networking model showed no statistically significant difference compared to traditional Chinese medicine prescriptions generated by the GPT-4 model (P>0.05). The proportion of model-generated prescriptions mistakenly judged as manual prescriptions in the Turing test was 51.11%. CONCLUSIONS The traditional Chinese medicine prescriptions generated by the GPT-4 model have reached a certain level of safety and suitability, and the GPT-4 model has passed the Turing test. The introduction of AIGC in the diagnosis and treatment process may provide technical support for the rational use of clinical traditional Chinese medicine.