Artificial Intelligence is often presented to schools as a teaching aid or tutoring tool. However, its most important consequence for science education is epistemological rather than technological. Machine learning systems can produce accurate predictions without relying on explicit theoretical models. This challenges a fundamental assumption of experimental sciences such as chemistry, where understanding is traditionally grounded on the chain observation–model–prediction–verification. The presence of predictive systems without explanatory mechanisms makes it necessary to clarify in teaching the difference between correlation and causation, and between performance and understanding. Rather than replacing laboratory activity, AI highlights its role as the place where scientific explanation is distinguished from statistical adaptation. The article argues that AI can be used not primarily to automate exercises but to expose misconceptions, compare empirical predictions with theoretical models, and show the physical cost of information in terms of energy and limits of computation. In this perspective, artificial intelligence becomes an opportunity to reinforce scientific thinking and to redefine the educational function of chemistry in the age of predictive machines.