Have you ever asked AI to explain a tricky problem, read the answer, and thought, “Oh, I get it now” and only to get stuck on a similar problem the next day? Every step made sense while you were reading. But when you had to decide where to start, you realised you weren’t quite sure.
AI makes this easy to experience because it can produce a clear explanation in seconds. Following an explanation, though, takes different work from coming up with one yourself. The OECD’s Digital Education Outlook 2026 notes that in some studies, students performed better with AI assistance, but that advantage disappeared and sometimes reversed when the tool was taken away.[1] That raises a question: when we see a good assignment, how much does it tell us about what the student can do independently?
Still, it would be too quick to conclude that using AI makes us worse learners. If you’re studying alone and have been stuck for an hour, a timely explanation with a useful example can make a real difference. The OECD also finds that AI can support learning when it is used with a clear educational purpose and appropriate guidance.[1] What matters is what happens next: do we work through the idea ourselves, or stop because we have the answer?
A finished assignment has also become a less reliable way to see what someone understands. Reuters reported on October 28, 2026, that some teachers in the US were turning to more classroom writing, discussion, and oral assessments so students could explain their thinking.[2] These approaches give teachers more to go on, although one shaky answer doesn’t necessarily mean a student hasn’t understood.
So, if we’re learning on our own, how can we tell whether AI is helping? One small thing I’d suggest is closing the chat, explaining the idea in your own words, and trying a similar problem. If you get stuck, you’ve found something worth asking about. This is my practical suggestion, rather than a finding directly established by the two sources above.
I’m in favour of learning with AI, provided we leave room to work things out ourselves after getting help. To see whether that approach works, we need to watch how our independent problem-solving develops over time. Take something you learned today: without reopening the AI answer, how much of it could you explain and use on your own?
---
By Huỳnh Thanh Tân · thanhtanmit.com
Notes and sourcesLink to this heading
[1] OECD, *OECD Digital Education Outlook 2026*. See the sections on why completing a task with generative AI does not automatically lead to learning, and how using AI with a clear educational purpose can support learning.
[2] Reuters, *How teachers are remaking classrooms for the AI age*, October 28, 2026. Reporting on how some teachers are changing lessons and assessments to better understand what students can explain and do independently.