Birthplace effect on AI models demonstrated by differences in understanding of Traditional Chinese Medicine and Western medicine for the Treatment of Chronic Kidney Disease (Preprint)

Abstract

<sec> <title>UNSTRUCTURED</title> Aim: Potential bias is regarded as one of the problems in the utilization of AI models. While ChatGPT and several other models are produced in US, DeepSeek is originated from China. The question is whether and how these AI models provide information and guide the application of drugs of Traditional Chinese Medicine (TCM) and Western medicine (WM) to treat chronic kidney disease (CKD) Material and Methods: ChatGPT version 4.o, Claude 3.7, Gemin 2.5, Grok 3, and DeepSeek were asked to provide opinions, names and functions of of top drugs of TCM and WM for the CKD. The answers from these models were compared for similarities and differences. Potential factors for the similarities and differences were evaluated. Results: While no two models showed the same opinions, there are significant differences in the opinion on TCM and WM between ChatGPT and DeepSeek (P =1.054E-07 for TCM and 6.741E-06 for WM). The three repeated tests indicated similar differences between the ChatGPT and DeepSeek. The major difference between DeepSeek and other four AI models are in treatment of CKD by TCM and WM are in 1) How it works, 2) how CKD affects daily life, and 3) the insurance coverage. The overall scores for the similarities among AI models for the names of WM is higher than that for TCM. The similarity scores between DeepSeek and other four models for drug names of WM are higher than that for TCM. The similarity between DeepSeek and other models was scored the lowest for the description of drug functions of WM and TCM. Conclusion: Birthplace effect may cause significant differences between DeepSeek and other models. While differences showed among all AI models, the difference between DeepSeek and other models was the largest. </sec>

Ähnliche Arbeiten