John MacDonald: I asked an AI bot how it felt about marking exams

With the Government wanting artificial intelligence to be used to mark high school exams, I asked a ChatGPT bot how it felt about that.   

And here’s what it said: "If the Government were to ask me to mark high school exams, I would feel (if I could feel) cautious optimism, with a big dose of responsible hesitation."  

The bot said, on the upside, it could process thousands of scripts quickly, never get tired, and not suffer from the end-of-day brain fog that human markers do. It also wouldn't mark one student generously and another harshly on a bad day. And it could apply marking schedules with perfect consistency.  

The bot also said there would be downsides. It said exam answers aren’t always clean-cut. A brilliant, unconventional insight might not fit the marking template, but a good teacher sees its value. That’s harder for AI to interpret correctly without "massive nuance training".  

It said AI can reflect biases in its training data. Even small disparities in language use, cultural references, or phrasing could disadvantage students if the system isn’t carefully designed and constantly audited.  

The ChatGPT bot also told me students, parents, and teachers would rightly ask how something was marked, and public confidence could be eroded.  

It said, on balance, it would be in favour of "hybrid marking”, with AI doing pre-marking and humans handling the grey areas and double-checking. So AI would be a support tool.  

In short, the AI bot said: "I’m capable. But I shouldn’t be trusted alone. Exam marking is too important to hand over fully to a machine - at least not yet."  

See omnystudio.com/listener for privacy information.

Episode Transcript

Speaker 1 (00:06):
You're listening to the Canterbury Mornings podcast with John McDonald
from Newstalk ZB.

Speaker 2 (00:13):
Here's one I prepared earlier. Here's one I prepared earlier
knowing that we were going to be talking today about
artificial intelligence after the government's announcement that AI is going
to be used to mark high school exams. Knowing that,
I thought I would go straight to the horse's mouth
and ask the robot itself what it thought of that.

(00:37):
So I put this question to chat GPT. I said,
the New Zealand government plans to ask you, the AI robot,
to mark high school exams. How do you feel about that?
And here's what the robot said. It said, if the
New Zealand government were to ask me to mark high
school exams, here's how I'd feel in brackets, if I

(01:02):
could feel, said, I would feel cautious optimism with a
big dose of responsible hesitation. And the robot said to
me said there would be three upsides. One speed and scalability.
It said, I can process thousands of scripts quickly and
never get tired, and I don't suffer from the end

(01:24):
of day brain fog that human markers do. Second upside, consistency,
said I won't mark one student generously and another harshly
on a bad day. I can apply marking schedules with
perfect consistency. And the third upside the robot saw in
marking high school exams was feedback potential. It said, if

(01:46):
integrated properly, I could give students more detailed feedback instantly,
not just a grade, but where they went wrong and
how to improve. Sounds quite good, the robots said, though
there would also be three downsides of it marking high
school exams. First one was judgment calls. It said, human
answers aren't always clean cut. A brilliant, unconventional insight might

(02:10):
not fit the marking template, the robot said, but a
good teacher i e. Human sees its value. That's harder
for AI to interpret correctly without massive nuanced training. The
second downside, equity concerns, said AI can reflect biases in
its training data. Even small disparities in language use, cultural references,

(02:33):
or phrasing could disadvantaged students of the system isn't carefully
designed and constantly audited. And the third downside of marking
high school exams, according to the robot itself, transparency and trust,
said students and parents and the teachers would rightly ask
how was this marked? It said, if the system as

(02:53):
a black box, public confidence could erode. And then the
robot the chat gpetty robots, suggested how this could be handled.
Given the pros and cons that it dished up, it said, quote,
I would support high marking. Let the AI do the
first test or the first past or pre marking, flagging
of his answers, and sorting levels of quality. Then let

(03:15):
human markers handle the gray areas and double check edge
cases or exemporary work. And it said, use AI as
a support tool, not a sole decider. The robot said,
in short, I'm capable, but I shouldn't be trusted alone.
Exam marking is too important to hand over fully to
a machine, at least not yet. So that's what the

(03:37):
chap GPT robot told me, or had to say to
me about marking exams. Then I said to the robot,
I said, oh, look, it's not just education where AI
is making an impact. And I told the robot that
I was going to ask you, as you who you
trusted most or more humans or the robots, And the

(03:59):
robot said, I'm quoting the robots said, most people see
the benefits like AI spotting cancer early, help with farming
or translating languages, but there's a nagging sense that things
are moving too fast. It's a very honest robot, it said.
There is fear of being replaced. Truck drivers, teachers, even
radio hosts might wonder if the robots are coming for

(04:20):
their jobs, so there's defensiveness. The robots summed it up
by saying AI is clever, but it will never have
what humans have, and it said people under thirty are
generally more open that they've grown up with tech, more
likely to ask how can I use it? As opposed
to should we stop it? The robot said. People over

(04:42):
the fifties me possibly you, we're a lot more cautious,
and then you might sell what the hell you've been
doing for the last few hours since you've been at work.
You've got the robot to do it all. Because the
robot then gave me some advice on what I should
be asking you to get the conversation about AI going,
and so I'm taking it up on its advice. The

(05:06):
robot said this. He said, ask your audience which situation
they would most or least trust aiin, and the robot
said to me, you got five scenarios. So here we go,
diagnosing cancer, driving a bus, writing news headlines, approving a mortgage,

(05:29):
and marking the high school English essay. That's what the
robot suggested, and I'm taking up as suggestion. So this
is the question for you, Which of those would you
be happy for an AI robot to handle? And which
of those wouldn't you be happy for a robot to handle?
And why? For me, I'm perfectly happy for the robot

(05:52):
to try and diagnose cancer, perfectly happy for the robot
to write news headlines, and I'm perfectly happy for the
robot to approve a mortgage. I'm not in favor, though,
of robots driving buses, and maybe the government doesn't want
to hear this. I'm not in favor of robots marking
school English essays.

Speaker 1 (06:11):
For more from Category Mornings with John McDonald, listen live
to news Talks at be Christchurch from nine am weekdays,
or follow the podcast on iHeartRadio.