According to the OECD educational report, it looks like students who were using AI did worse with their science and mathematics tests, while children who did not use AI did much better with their tests. It also becomes obvious that children are giving more complex tasks to AI, and this results in an effect called cognitive offloading. So children tend to give more complex tasks and assignments to AI, and this doesn't allow them to learn as efficiently.
At the same time, what happens is that children who use AI, but use it to assist them with their studies and research, and when students apply critical thinking towards what AI is offering, and they check the sources, and they read them, and then they work through this information, then in that case AI really helps them to learn, and they are doing way better both at exams and on a daily basis. It is saving them time, and it reduces the amount of effort students used to put into study when they had to find information.
Also, many students found AI useful when it actually helps them to learn, and it acts as an assistant, providing them with various sources of information. But it becomes less efficient when students are asking AI to actually write an essay or solve the problem for them, especially for primary‑age students, where they actually have to train their brains and build the needed connections in their brains, and develop various skills like critical thinking and logic. So this is where AI actually doesn't help much and, on the opposite, is preventing students from learning and doing better.
So, to conclude, parents and teachers should find a proper balance of how, when, and how much AI tools should be used. And by AI, we mean generative AI, of course. There should be some balance where these tools can actually help students find information and assist them with learning, but not substitute the process of learning and working on the task, especially when the tasks are complex.