I was the (co-)grader today for presentations about MSc students' literature studies. A student presenter shared that they had read through 100 papers to find the results for their literature study. Following the presentation, a student in the audience asked something like: "One hundred papers sounds like a lot, did you really read that many papers? Why didn't you use AI to go through them?".

Firstly, it was interesting that the audience member thought that reading 100 papers was unfathomable. The number may seem large, but assessing 100 papers for a literature study is not an extreme, more the norm. Whether you read 100 word-for-word for a 20-40 page literature study may not be the norm, but manually reviewing that many is. I can't say that I've explicitly heard the concern before, although students do often ask how many papers are sufficient for a literature study. [1]

Secondly, I am not sure the audience realized this but the speaker had a strong command of the material in their study and all other questions were easily and fully answered. The ability to do this is the product of reading 100 papers on a topic. The relevant information in the papers was inserted into the speakers brain through the action of reading and writing themselves.

Each day, more things point towards the negative effect generative AI is having on learning. I guess we will have to persuade new students to actually read the papers from now on, when in the past that was a given.

[1]	I do not give a numerical answer to this question, btw.