Does AI make us smarter, or just more dependent? 10,503 posts offer a clue
What happened
A study published in the journal Comunicar in 2026 set out to answer a question that schools, universities and parents keep arguing about: is using AI to learn making people cleverer or lazier? The research was led by Associate Professor Ye Weiming of Peking University HSBC Business School and Li Qian of Arizona State University's Walter Cronkite School of Journalism and Mass Communication. Rather than running an experiment in a classroom, they looked at what people said when nobody was testing them. The team analysed 10,503 valid Reddit posts from 2013 to 2025, using a method called the Biterm Topic Model alongside qualitative analysis. They sorted what they found into three stages of learning. In knowledge acquisition - finding things out - people valued AI's speed and its ability to organise information. But they also reported what the researchers call 'credibility anxiety': a persistent worry about factual errors. In knowledge comprehension - understanding things - people found AI useful for simplifying difficult concepts. But over-reliance risked weakening independent reasoning and critical thinking. In knowledge application - actually using what you know - AI reduced the cost of trial and error. It also raised concerns about plagiarism and intellectual property. Across all of it, the researchers found a repeating pattern they describe as 'a cycle of trust, doubt and recalibration'. People did not settle on a fixed view. They adjusted, and increasingly turned to collaborative verification and community checking rather than taking AI output at face value.
Why this matters
Most of the public argument about AI and learning is conducted between two loud positions. One says AI is a calculator for words and resisting it is pointless. The other says it destroys the mental effort that learning depends on. This study suggests that ordinary people, left to themselves, do neither. They use it, get burned, become sceptical, work out where it is reliable, and adjust. That is a more encouraging picture than either camp offers. The finding about verification is the most interesting part. People increasingly checked AI answers against each other and against communities of other people. That is a skill - and it is arguably the central skill for the next twenty years. The warning about comprehension is the one worth taking seriously. Understanding something is effortful by nature. The struggle is not a bug in learning; it is a large part of how learning happens. A tool that removes the struggle can leave you feeling you have understood something when you have only watched someone else understand it. That distinction - between following an explanation and being able to produce one - is where the real risk sits, and it is not obvious from the inside.
What the evidence actually says
This study has a clear strength and a clear limitation, and both should be stated plainly. The strength is that it captures what people say voluntarily, over twelve years, without a researcher watching. Survey answers about technology are notoriously unreliable, because people say what sounds sensible. Reddit posts are messier and more honest. The limitation is that Reddit users are not a representative sample of anyone. They skew younger, more male, more technically minded and more Western than the general population. A study of Reddit tells you about people who post on Reddit. It also does not measure learning. Nobody was tested. There are no marks, no retention scores, no comparison between people who used AI and people who did not. The study measures what people said about their experience, which is a different thing from what actually happened in their heads. So the honest reading is: this is a rich picture of how people talk about learning with AI, and a genuinely useful map of the concerns that recur. It is not evidence that AI does or does not improve learning outcomes. Anyone citing it as proof either way is overreaching. The researchers' own conclusion is measured. They describe AI as becoming 'an intellectual partner, behavioural facilitator and ethical interlocutor' rather than just a tool, and call for transparent governance that keeps pace with the technology.
Practical advice
If you are a student, or a parent of one, a few habits follow from this reasonably directly. Try it yourself first. Use AI after you have attempted something, not instead of attempting it. The struggle before the answer is where most of the learning sits. Asking first and reading the answer feels efficient and usually is not. Ask it to explain, then explain it back without looking. If you can say it in your own words to someone else, you have learned it. If you cannot, you watched an explanation happen. This is the single most useful test there is. Check anything that matters. The 'credibility anxiety' in the study is well founded. AI tools produce confident, fluent, wrong answers regularly, and they are most convincing when you know least about the subject. Dates, figures, citations and medical information are worth verifying against a real source every time. Never rely on it for health information. AI can be helpful for understanding a word you have been given, but it does not know your history, your medicines or your test results. For anything about your own health, ask a GP, a pharmacist, or use NHS 111. Be clear about the rules. Schools, colleges and universities have their own policies on AI use, and they differ. Ask rather than assume - the consequences for getting this wrong in coursework are real. And notice your own cycle. If you have moved from excitement to disillusion, that is the pattern the study found. The useful place to land is not 'always' or 'never' but knowing which tasks it is reliable for.
What to know
Researchers analysed 10,503 Reddit posts from 2013 to 2025 to see how people describe learning with AI. They found benefits in speed and in simplifying hard ideas, alongside persistent worry about factual errors, over-reliance weakening independent reasoning, and plagiarism concerns. The pattern running through it was a cycle of trust, doubt and recalibration, with people increasingly verifying AI output collaboratively rather than accepting it. The study is a picture of what people say, not a measure of what they learn, and Reddit users are not representative. It does not settle the question in its own headline. The practical takeaway stands on its own regardless: attempt things before asking, explain the answer back without looking, check anything that matters, and never use it for your own health questions. Sources: Comunicar (Grupo Comunicar), Ye Weiming and Li Qian, study of public discourse on AI-assisted learning, 2026, https://www.revistacomunicar.com/. Phys.org, 'Does AI make us smarter or more dependent?', 17 September 2026, https://phys.org/news/2026-09-ai-smarter.html. NHS, 'NHS 111', https://111.nhs.uk/. This article is for general information and does not replace advice from a doctor, pharmacist or other qualified healthcare professional.
Related articles
Free school meals: schools get six extra weeks, but families still have to apply
The Department for Education has given schools and councils six more weeks to complete free school meal eligibility checks. The deadline has moved from Thursda…
Ofsted checked 68 nurseries in one chain. Here is what inspectors found
Ofsted has published an update on its regulatory action against Bright Horizons, one of the UK's large nursery groups. Since September last year, Ofsted has be…