There is a paper that has been causing lots of chatter recently. It is a paper released by authors from the Massachusetts Institute of Technology and it suggests that using generative AI (in this case ChatGPT) has important effects humans write essays. Folks who I associate with are educators and technology experts, so they have much to say about the paper, and with good reason.
One thing that seems to have confused some folks who are talking about the paper is the fact it is a preprint release. Usually when we hear about “new research” is has been published, and it if is scientific research it has been peer reviewed. This finds experts in the field reviewing the paper to assess its significance, the appropriateness of the methods to answer the research questions, the analysis, and the conclusions. Once reviewed the paper is reviewed it is usually sent back to the authors who make the improvements suggested by the reviewers before the editors of the journal publish it. Peer review is typically accepted as the “gold standard” for research.
The preprint paper of interest is entitled Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task and it was released on arXiv.org on June 10, 2025. It has been gathering attention in the way the published research usually does, but it must be remembered it has not been reviewed. This does raise some important questions.
Does this mean it is not good information? No. The authors of the study are known and their work can be reviewed. One of the reasons why authors release papers rather than publish papers is to encourage wider evaluation of and feedback on the work.
Are preprint papers published later? Sometimes. Of course, peer review is supposed to be a double-bind process, to the authors are not known to the reviewers and the reviewers are not known to the authors. Preprints, especially on papers like this, are likely to be well-known to reviewers later. Of course, any preprint is likely to be known by potential authors in popular fields.
Will the peer reviewed paper be the same on the was in the preprint version? Probably not. One of the great advantages of preprint papers is they can be long. This one was over 150 pages not including references and appendices which give man more details about the study. Few peer reviewed articles will be this long; preprint audiences get way more detail than many need (or want).
Should we be cautious about taking action based on a preprint? Yes. But this is true of all research, including that which has been per reviewed. The replication crisis has been a recognized reality in science for some time. There are several factors that affect findings authors choose to submit for publication and this leads to the problem that many important new announcements in the science headlines are not replicated. This does not mean the scientists lied in their original findings, it just means the finding were probably the result of variation and if they did the experiment again, they would not find the same thing.
Especially when the topic is generative AI and its effects on teaching and learning, folks like me and my colleagues are going to pay particular attention; close attention, and unfortunately, they are likely to take only those parts of the research that align with their thinking and accept them as fact. As with all research, these finding need to be replicated and expanded. As with all research these finding need to be explored to see if they are accurate for your students in you place and in your course. But that is the same as with all research.