Objective To explore the performance of artificial intelligence (AI) auxiliary diagnosis system based on machine learning in the diagnosis of chronic non-specific back pain. Methods FlexComp Infiniti 10 Electromyography (Thought Technology Ltd) was used to collect EMG signals of multifidus muscles, longissimus muscles and lumbar iliac rib muscles on the left and right backs of 15 subjects. The 15 subjects were all diagnosed patients with chronic non-specific back pain. A round of test included five flexion relaxation exercises, one bipedal bridge exercise, one left foot bridge exercise, one right foot bridge exercise, and one Biering Sorensen isometric movement. Three senior physicians diagnosed the subjects by electromyography using conventional methods and this was taken as standard. AI aided diagnostic results were compared with the doctor’s results. Results The detection accuracy of AI auxiliary diagnosis and treatment system for 15 patients with non-specific back pain selected in this paper reached 100%. The results of muscle recruitment, fatigue speed and resting speed tests were in good agreement with the doctor’s diagnosis. The average diagnosis time required was reduced by 26.3 min, which was statistically significant (P<0.05). Conclusion Preliminary verification results show that the AI auxiliary diagnosis system can provide accurate and efficient diagnosis and quantitative evaluation for chronic non-specific back pain, and provide reliable help for clinical detection.