Artificial intelligence-assisted surgical skill acquisition: a systematic review and meta-analysis of randomized controlled trials

Artificial intelligence-assisted systems are increasingly evaluated for automated assessment and individualized feedback in surgical education. However, randomized evidence is limited, and their benefit over conventional training, particularly expert instruction, remains uncertain. We searched PubMed, Embase.com, the Web of Science Core Collection, the Cochrane Central Register of Controlled Trials, and IEEE Xplore from inception to 6 August 2026, followed by backward and forward citation searching on 7 August 2026. Randomized studies of artificial intelligence-assisted training for operative technical skills were eligible. The primary outcome was overall technical skill performance. Effect estimates were expressed as Hedges’ g and pooled with a random-effects model. Risk of bias was assessed using the Cochrane Risk of Bias 2 tool, and certainty of evidence using the Grading of Recommendations Assessment, Development and Evaluation approach. Nine studies enrolled 588 participants, of whom 582 were analyzed; four studies contributing 154 participants entered the primary meta-analysis. The pooled standardized mean difference was 0.40 (95% confidence interval − 0.30 to 1.10; P = 0.17), with I² = 36%. Compared with active human instruction, the pooled standardized mean difference was 0.13 (95% confidence interval − 0.24 to 0.51; P = 0.48). For passive conditions or training without artificial intelligence, the pooled standardized mean difference was 0.94 (95% confidence interval 0.30 to 1.59; P = 0.004). The subgroup difference was statistically significant ( P = 0.03), although each subgroup contained only two studies. Sensitivity analyses did not materially change the interpretation. One study was at low overall risk of bias and eight had some concerns. Certainty of evidence was low because of risk of bias and imprecision. Current evidence does not establish a clear overall benefit of artificial intelligence-assisted training for short-term surgical technical skill performance. Superiority over active expert instruction was not demonstrated, whereas the larger point estimate against passive conditions or training without artificial intelligence should be regarded as exploratory. These findings are compatible with a complementary rather than replacement role for artificial intelligence in surgical education. Larger trials with standardized outcomes and longer follow-up are needed.