Can ChatGPT give feedback?

In our previous post, we looked at how ChatGPT can provide relatively plausible marks for pieces of writing.

In our previous post, we looked at how ChatGPT can provide relatively plausible marks for pieces of writing.

It can also provide written feedback, which has got everyone excited about the potential time savings.

If you give ChatGPT the mark scheme and ask it to give the essay a grade and explain why, it can produce a comment that uses the language of the mark scheme.

Clearly, this will save lots of teacher time. But will it help students learn? I am not so sure. I don’t think this kind of written comment — whether it is written by a human or by a bot — is that helpful.

Why? Suppose you are learning to drive with an instructor, and you practise parallel parking. What kind of feedback does the instructor give? Well, as you are carrying out the manoeuvre, she might say remember your reference point! Or, OK, full lock now! Or, check your mirrors! Or, if you are trying to park in test conditions with no real time feedback, she might let you carry out a turn on your own and then at the end say “OK, that wasn’t too bad, but the reason you clipped the kerb at the end is that you were just a bit too late with the full lock, so let’s try again and bring your reference point forward a little bit”. Either approach is useful, and all this makes total sense in context.

Let’s imagine that she asks you to parallel park. She stays silent while you do the manoeuvre and after. Doesn’t say anything. Then a week later — maybe even after you have had another lesson with her — you get a letter in the post that says: Well done, your skills at parallel parking are improving and you showed good independent awareness of the steps required to effectively parallel park. You identified a safe and legal place to carry out the parallel park which was not too close to a junction or on a double yellow line. You were mostly in control of the manoeuvre but to improve you could make sure that you slow down and remember to check your mirrors. Well done, you are making good progress.

Which kind of feedback is more useful? The first, obviously. Which kind of feedback is more time consuming for the instructor? The second, obviously.

And this is why one of the most interesting innovations in English education in recent years has been the development of whole-class feedback. The principle behind whole-class feedback is that instead of a teacher spending hours writing out comments like the above, they read their students’ writing and replan their next lesson based on the strengths and weaknesses they have seen.

We have blogged about it extensively (eg see here) and provided lots of examples of teachers using this approach. There are also interesting variations on the theme — for example, ‘live marking’, where a teacher circulates in the class and attempts to provide driving instructor style verbal feedback for as many students as possible. This is obviously challenging because, unlike a driving instructor, teachers have 30 students, not one! But there are ways around the challenges, and Ashley Booth has a fascinating tweet thread here about how you might cope with this.

Whole-class feedback saves time, but that is not the only rationale for it. It is more effective than traditional written comments.

However, sometimes we have an aversion to the idea that something can save time and be more effective. Surely if it is saving time it must somehow be less effective, or a cheat? I don’t think this is the case, but I do think that many schools and students still see the written comment as a kind of gold standard. Perhaps written comments feel more ‘personalised’ than whole-class feedback, even though teachers often just have a bank of comments that they rotate round.

So is there any point in using ChatGPT’s mark-scheme based written comments? Perhaps they could fulfil the performative aspect of marking. A detailed comment that is linked to the mark scheme will tick a lot of bureaucratic boxes, and the teacher can get on with actually teaching and feeding back to students in a way that is useful.

I hope that whole-class feedback will continue. It is a far more effective innovation than the comment above, and I would hate to see the genuine pedagogical improvements brought about by whole-class feedback set back because people get starry-eyed about automatically generated text comments.

However, ChatGPT’s language abilities are undeniably impressive. And in fact, by trying to get it to replicate the traditional mark-scheme based written comment, perhaps we are missing a trick. We’re doing that thing everyone does when a new technology comes along, which is to use it in exactly the way we used the old one — eg, bosses who got their secretaries to print out all their emails and then dictated responses to them.

What if the secret of using ChatGPT for written feedback is not to use it to replicate the typical mark-scheme based comment — but to get it to do something completely different? Is there any way we could get it to provide feedback that is more specific and useful — in fact, a bit more like the type of feedabck teachers provide with whole-class feedback, but personalised for each individual? If it could do that, it would be really impressive. We will consider that in the next post.