I Let AI Help Me Grade
As a teacher in the world of Software Development, AI is a given. It's something that cannot be ignored. And it's something that provides opportunities as well, both for my students and for myself. Therefore I'm continuously looking for ways to use AI in education. Not as a tool that does things for you without any editorial control, but more as a starting point, for example, when assessing student work.
In that context, I've been experimenting with something that might seem simple, but works well for me: speech-to-text. Using a tool like MacWhisper, your spoken words are converted into written text. My current workflow in which I use this tool is as follows: I open a student's portfolio, which could be a single assignment or a full project documentation, and read through it with the rubric in mind. As I read, I speak my thoughts into my computer using said speech-to-text tool. I do this until I've worked through the entire portfolio. This results in a list of feedback points and questions about the portfolio.
I then have an AI summarize these points, because when you speak, you tend to ramble (I know I do) and an AI is good at compressing that (it's a "Large Language Model", after all). It also happens that I mention the same point multiple times, in different contexts. The AI merges those into a single feedback point. I read through the final list, adjust where needed, and return it to the student. Individual feedback given relatively fast.
This works well for me. I don't have to switch back and forth between a rubric, the portfolio, and a note-taking tool. I just talk to my computer whenever something about the student's portfolio stands out, and let an AI bundle it all together afterward. And it's faster than typing everything out.
One potential downside: when you type, you're slower and naturally think more carefully about what you write. Writing forces you to think harder about the feedback you want to give, harder at least compared to speaking. That said, this approach is still better than having an AI generate feedback points directly based on a student portfolio, because then you're not engaging with what you're reading in the portfolio at all.
For me, right now, it's a good middle ground: not letting the AI do all the work (which would introduce another question about who's responsible for the given feedback) but also not manually typing or writing out everything myself. This way of using AI might very well change again in the future as I'm experimenting with other options continuously.
I'm curious what you think. Do you use AI in a similar way? Do you give even more responsibility to AI? Or do you not use it at all? feel free to contact me and let's have a discussion!
This blog post was written by talking into MacWhisper and manually editing the resulting speech to text.