Should AI Be Trusted With Grading Students’ Work?

[00:00:00] What will the world look like 10 or 20 years from now? The Wall Street Journal's Future of Everything podcast is here to give you a peek. And we can't wait to show you what's coming. Subscribe now. Welcome to Tech News Briefing. It's Wednesday, July 3rd.

[00:00:22] I'm Zoe Thomas for The Wall Street Journal. Doctors are using the Apple Watch as part of how they diagnose and help their patients manage disease. Even in cases where the device hasn't been specifically approved by regulators for the applications it's being recommended for.

[00:00:39] We'll explain why some doctors are recommending the watch. And then, generative artificial intelligence is spilling into classrooms and nachos from students looking for shortcuts. Teachers are turning to AI tools to help them mark students' work.

[00:00:55] But marks can be very different depending on which AI grading program a teacher uses. Our reporter Sarah Randazzo will join us to explain how the tools are being used. But first, there's a large and growing body of research on how the Apple Watch is being

[00:01:16] used informally in medical care. Doctors are recommending some patients use it to track health information even if there are other approved devices available that track the same metrics. Here to tell us more is our tech columnist Christopher Mims.

[00:01:31] So Christopher, what is it about Apple Watches that have doctors recommending them to be used by patients? The Apple Watch has two sensors on it which are really well validated as they say in research circles.

[00:01:43] So the heart rate sensor is very accurate and it can also do an EKG. On the newer models where it's looking at the electrical signal of your heart, it's obviously not the same level as one you would get in a hospital. But apparently it works really well.

[00:02:02] Tell us more about what kinds of conditions doctors are suggesting Apple Watches for. Doctors are basically off-label prescribing the Apple Watch, especially if patients have arrhythmia or other heart issues because the watch can alert you when you are having an arrhythmia.

[00:02:22] You can also use it if you have other problems. Maybe you feel some symptoms, you can do an EKG right then. And then of course it's easy to put that data on your phone and just email it to your doctor.

[00:02:33] So especially cardiologists are actually telling patients to go buy one. That's actually how I found this story is that my mother's cardiologist said, go get an Apple Watch SE, do these recordings, email me the data.

[00:02:48] What does the Food and Drug Administration said about off-label use of the Apple Watch? FDA approvals do limit what kind of sensors go on the watch in some ways and how they can be used.

[00:03:00] So the FDA has only approved, for example, a fibrillation alerts on the watch if you don't already have a history of it, funny enough. If you do have a history, it won't give you those alerts because of the sort of trickiness

[00:03:16] of what the FDA has approved it for and not. All right. So what has research shown about the usefulness of Apple Watches in tracking these health conditions? Because the watch can track so much data, researchers, it's kind of a bonanza for them

[00:03:30] because they can get the data more or less directly from the watch. And so I talked to researchers who were using it to track post-surgical recovery and try to make that better. Other researchers said that the heart rate and other data on the watch could become the

[00:03:48] new gold standard for determining how stressed people are. And they've basically discovered there's all this kind of almost like secret information. I mean, the characteristics of the heart rate data that no one had realized before was indicative of stress.

[00:04:01] So they're taking all that data and then they run it through a machine learning algorithm. So it's allowing them to use AI to examine how humans are behaving in the real world in a way that they just couldn't before.

[00:04:14] Because before we weren't all walking around with these high definition wearable sensors on our bodies. What are some of its limitations and the concerns about using the Apple Watch in these ways? Some doctors I talked to are concerned that if more alerts are added to the watch, like

[00:04:29] right now it'll give you like high resting heart rate or arrhythmia or whatever, you'll end up with too many false positives. And that could send people to the hospital where immediately doctors have to take them seriously.

[00:04:44] And so there can be a battery of expensive tests and it could be a big additional burden on the health care system. All right, that was our tech columnist, Christopher Mims. Coming up, AI bots are marking papers in some schools.

[00:05:00] But should something as high stakes as grades be left to AI? We'll find out what supporters and critics say after the break. What then will the future reveal? There's one thing we know about the future. It's being built now. We all have a stake in the future.

[00:05:24] The future. The future. The future. And the Wall Street Journal's Future of Everything podcast is here to give you a glimpse of what's on the way. I'm Danny Lewis. Join us as we dig into how science and technology are shaping the future.

[00:05:36] For that is where you and I are going to spend the rest of our lives. Subscribe wherever you get your podcasts. Researchers are embracing new artificial intelligence grading tools. Some say the programs let them give students faster feedback and more chances to practice,

[00:05:59] while others say the technology isn't reliable enough yet to be used for something as high stakes as grading. Here to tell us more is our reporter Sarah Randazzo. Sarah, what sorts of grading are these tools being used for? Right now, the majority are being used for writing.

[00:06:16] It's really the easiest use case for these generative AI tools. And so most of the tools will have ways for teachers to upload all their student essays, and then it scans through them within a few minutes, offers both a numerative grade, you

[00:06:28] know, like your 87 percent, your 90 percent, as well as a bunch of feedback. So it'll give feedback on whether you had a thesis statement correctly, whether you had correct narrative elements, and then things like that. Do we have a sense of how popular these grading tools are?

[00:06:43] How many teachers or schools are using them? Yeah, it's hard to say because a lot of these tools are new and they're using the model of giving it to teachers for free, and then they try to get districts and schools to pay

[00:06:53] for it on more of a larger basis. Definitely thousands of teachers have probably signed up for these tools. It's hard to say if it's tens of thousands. There's just not good numbers right now.

[00:07:02] But really this last school year is the first year when a lot of these tools were even available for teachers to start playing with. Why have teachers turned to them? The teachers I spoke to who are big believers in them said they basically allow them to give

[00:07:15] feedback in a way they never could by hand. So for instance, there was an AP US history teacher I spoke to who had his students at least once or twice a week use this one program to do a prompt and do a timed essay, and then

[00:07:27] they instantly got feedback from it on how they did and how they would have done on the AP exam had they written that response. And so if he was having his say 90, 120 students do essays once or twice a week, for him to

[00:07:41] hand grade those and give constructive feedback would have taken a long time. And then by the time they get it back, they've moved on to a different unit. So for him, it was a way to really give real-time feedback and get them just practicing writing

[00:07:51] a lot more often than they could have otherwise. What sort of criticism is being leveled against these tools? Some people say grading really should not be a use of AI, especially the numeric grading part. Most people think the feedback is okay.

[00:08:04] You know, sometimes it might spit out improper feedback, but the teacher could override that. But the real concerns around when it's spitting out that grade, that 80 percent, that 90 percent, these tools can just be pretty unpredictable still.

[00:08:16] So some people, especially, interestingly to me, a lot of people who are in the kind of AI evangelism world are pretty hesitant about using AI for the grading aspect right now. But with the teacher's input, some of these things everyone says could be useful, but no

[00:08:31] one wants it to be a system where a teacher is pushing a button. It's doing the final grades without any kind of input. So how do teachers' grades compare then to these AI systems? It really depends.

[00:08:43] We ran a test using a colleague's paper that they had from when they were senior in high school, and it was a perfect example because she still had the grades and rubric and feedback from the original essay.

[00:08:53] And so we compared that to three different sites that we tried. One of the sites gave the essay a 100 percent. Another one gave it in the 60s, and another one gave it 85 percent. Those are three different tools all aiming to do the same thing.

[00:09:06] You know, I'd say that kind of shows it really is a little bit all over the board right now. What have the companies that make these tools said about the results? Yeah, so a lot of them say they continue to refine the systems. They're constantly doing updates.

[00:09:18] They're adding tools based on teacher feedback. Like one of the companies added an option so that teachers could only use the feedback side of it, not the number grade part, so they can just hide the number grades altogether

[00:09:28] if they don't agree with them and only send the more constructive feedback to students. One company has a button where you can push it to make the feedback nicer, essentially. So they say that they're continuing to work on the systems and like everything with generative

[00:09:41] AI, you know, it's all a little bit in the test phase right now, but can do some powerful things but with some caveats. How are students reacting to the use of these tools?

[00:09:50] So I was able to speak with a couple students from one of the teachers who is a big believer in this, that AP US history teacher I had mentioned. And this one student was really into it and said that at the beginning of the year there

[00:10:01] was kind of a blanket skepticism among his classmates of what is this? Why do we have to use this? But as the year went on, they got more and more into it.

[00:10:08] And he felt like it really prepared him for the test, the AP exam, so that when he walked in, he knew what to expect. Other teachers have said that students almost turn it into a game and get emotionally connected

[00:10:18] in a way that they didn't before and will want to beat the robot as teachers explained it. So, you know, for better or worse, students are treating it a little bit in that gamified fashion of saying, okay, on my first draft, it gave me a 70%.

[00:10:31] Let me see if I can write better to get it higher. So there's some of that going on and I'm sure there's some students who just don't like it at all, but I wasn't able to speak with those students directly. That was our reporter, Sarah Randazzo.

[00:10:43] And that's it for Tech News Briefing. Today's show was produced by Julie Chang with supervising producer, Catherine Millsop. I'm Zoe Thomas for The Wall Street Journal. We'll be back this afternoon with TNB Tech Minute. Thanks for listening.