College prof hides prompt to catch AI cheaters, finds human nature is pretty much as we thought

Busted! A college professor who was tired of reading AI responses to essay questions found a unique way to determine if his students were cheating. Sadly but not surprisingly, 32 of the 35 students who took the test fell right into his trap. Jason Gibson, who teaches history and African American studies at Alcorn State …

[LionelB]

"Madgacascar floats sideways / through the afternoon"

Very beautiful. Could be the last two lines of a haiku. Any takers for the first line (5 syllables)?

[Manolo]

Re: "Madgacascar floats sideways / through the afternoon"

A coconut falls.

Madagascar floats sideways

through the afternoon.

[LionelB]

Re: "Madgacascar floats sideways / through the afternoon"

I like to move it*

Madagascar floats sideways

Through the afternoon.

_{*Would work better as a last line, but hey :-)}

[Bebu sa Ware]

Re: "Madgacascar floats sideways / through the afternoon"

> A coconut falls / Madagascar floats sideways / through the afternoon.

No Lemurs… you've got to have Lemurs.

[Manolo]

Re: "Madgacascar floats sideways / through the afternoon"

Lemur tilts island.

Madagascar floats sideways

through the afternoon..

[I ain't Spartacus]

Re: "Madgacascar floats sideways / through the afternoon"

Whisky at sunrise

Madagascar floats sideways

Through the afternoon

[LionelB]

Re: "Madgacascar floats sideways / through the afternoon"

Are you sure it was whisky?

[Stanley Toolset]

Re: "Madgacascar floats sideways / through the afternoon"

A volcanic spliff!

Madagascar floats sideways

through the afternoon.

[Pete 2]

Continuing the tradition

I have never held history in high regard.

Possibly the best description of it comes from Professor Suzannah Lipscomb whose TED talk tells us that "it is the study of something that doesn't exist". However, it is also true that most writers of history textbooks merely copy content from other history texts. With little or no substantiation of the pirated material, nor reference to any (if they ever existed) primary sources. Just occasionally changing bits to reflect contemporary biases or opinions.

As such, it seems to me that punishing students for doing essentially the same thing for their course work is a little harsh. It also doesn't reflect well on the teacher, that their students felt no greater need to put effort into their assignments.

One also wonders if the (two?) students who didn't fall foul of this ruse merely typed their requests into an AI longhand, thus avoiding the hidden embedded trap?

[tychos nose]

Re: Continuing the tradition

> One also wonders if the (two?) students who didn't fall foul of this ruse merely typed their requests into an AI longhand, thus avoiding the hidden embedded trap?

Or indeed proof-read the AI response and manually removed the Madagascar reference?

[LionelB]

Re: Continuing the tradition

"[history] is the study of something that doesn't exist"

Of course. It's the study of something which existed.

[Pete 2]

Re: Continuing the tradition

> something which existed

Quite possibly. But it is by no means certain that historians have access to full, truthful and accurate records of any historical event. Most of "history" was never recorded. It exists only in the modern world as a collection of opinions that are more or less coloured by interpretation, assumption and conventional agreement.

[LionelB]

Re: Continuing the tradition

Sure. Nonetheless it is the study of the past, as best we can parse it through whatever records do exist, and (with best practice) taking into account the perspectives, agendas and biases inherent in those records.

A thoroughly honourable and worthwhile enterprise to be sure, but not, of course, to be confused with science.

[doublelayer]

Re: Continuing the tradition

Which is a thing that good historians spend a lot of time dealing with. A lot of their research is repeatedly asking the question "How much can we trust this source when it talks about this topic", finding and defending reasons to answer it, and using that to enhance the knowledge. But also, your implication applies better to things before the modern day where records are much more plentiful, even when historians still have to extract truth from biased or mendacious records.

[Grey Bird]

Re: Continuing the tradition

You seem to think that historians only exist to write historical text books. This is not true, it is also not true that historians just crib previous textbooks to do their work. There is this thing called research, which requires going back to original sources, some of which were actually created at the time of the historical events! While admittedly many basic history textbooks seem to exist only to rehash basic events which have been taught for many decades there is much more to history, and if you think that is all it is then o wonder you don't hold it in high regard. Those courses, however, are far from the best that studying history holds and using them to form your complete view of what history is shows a lack of critical thinking. (...and no, I didn't major in history or any other humanities subjects.)

[Bebu sa Ware]

"a chunk of white text that students wouldn’t spot"

Given the location and the academic's subject, I might have phrased that "a chunk of text displayed in the background colour that the students wouldn't spot."

With the understandable sensitivity of these matters in Loonyland you could be certain some loonier·lander would start ranting "white lies matter."

[Bebu sa Ware]

"Should you use AI for a task?"

Bruce Schneier^† in his academic teaching roles at Harvard and the University of Toronto considers this question in a recent grauniad opinion piece: Should you use AI for a task? Here’s a simple way to decide.

In the first paragraph "it will come as no surprise to you that my students regularly use AI to complete their writing assignments. Doing so is a waste of their tuition money" but his following arguments differentiates situations in which he contends that AI might be applied from those where it oughtn't with which the reader may or may not agree.

_{† a writer I suspect not entirely unknown to the Vulturati. :)}

[Pete 2]

Re: "Should you use AI for a task?"

> Doing so is a waste of their tuition money

One might venture that the waste comes about by setting tasks that require so little academic effort, that a machine can do them.

[Paul Kinsler]

Re: setting tasks that require so little academic effort, that a machine can do them.

Perhaps the machine might give the impression of doing the set tasks, but it's not the same as /actually/ doing them.

[LionelB]

Re: setting tasks that require so little academic effort, that a machine can do them.

It's obviously not the same as the student actually doing them.

However, the end product may well be virtually indistinguishable from that produced by an averagely-competent student. That's a problem for teachers (but see my later posts).

[Roland6]

Re: setting tasks that require so little academic effort, that a machine can do them.

>However, the end product may well be virtually indistinguishable from that produced by an averagely-competent student.

In an education setting "the product" is a student ie. human being, educated in a particular field to a particular standard.

[LionelB]

Re: setting tasks that require so little academic effort, that a machine can do them.

Ha, ha, point taken. But you knew what I meant.

[doublelayer]

Re: "Should you use AI for a task?"

If you care about teaching people who haven't already self-taught themselves most of the topic, you'll have to assign things that train the basic skills and knowledge so they gain the ability to do something more advanced. That means that the solutions to introductory courses are easy to find online, and therefore easy for LLMs to reconstitute without checking. Your argument is a similarly bad one as if I argued that arithmetic problems given to young students are trivially computed by a pocket calculator, so we might as well not teach them that at all.

That's how a lot of courses work. They train students on problems whose answers are already known, and usually known publicly, meaning you can find or create solutions with the right software. That software isn't always LLMs, for example if you want a proof generated you're more likely to get one using software specifically designed for doing that, but it exists. The only alternative to that is not bothering to teach students, just give them research papers and see if they can independently do valid original research in the field. Many of them won't be able to without some instruction, and LLMs will cheerfully produce fake original research too which would be much more annoying to detect.

[LionelB]

Re: "Should you use AI for a task?"

As a Research Fellow in academia, I supervise MSc and PhD students. I can assure you that fake research ("fake" as in not valid, that is) is not at all hard to detect if you're familiar with the field. "Fake" as in plagiarised research may be harder—well, at least more work—to detect. At that level of study, however, it really isn't in the student's interest to produce fake work in either sense; and if they're stupid enough not to realise that, chances are they wouldn't have made it to the MSc/PhD level in the first place.

[doublelayer]

Re: "Should you use AI for a task?"

Using LLMs on earlier assignments it's more likely to succeed at isn't in the students' interest either. Some of them are stupid enough to do it, either getting caught because they aren't good enough at using it, or succeeding and getting passed when they don't understand the topic, which they will likely regret later if they ever need what they were supposed to have learned.

But you're ignoring the rest of my comment. In responding to Pete's comment, I've explained why students are first assigned things with known answers, but if they were not as he suggests, then you would be testing many more students on their ability to produce research, some of them being the same ones who cheated with LLMs this time and others being ones who would not have cheated but, since they weren't taught, are incapable of doing the research. That does mean you'd have more fake research to detect. I expect you would catch quite a bit of it, but would you catch it quickly or might you have to do rather annoying verification to be conclusive?

[LionelB]

Re: "Should you use AI for a task?"

Sure, I was speaking about my personal experience.

The way it works at my (UK) higher-education institution, and probably quite generally in the UK (although not necessarily in Europe or other parts of the world) is that an undergraduate degree course teaches familiarity with a field of study up to a certain (sub-research) level. I don't personally teach at undergraduate level.

The idea of a Masters degree is to take you up to research level in the subject, and in particular to teach how to research. There will usually be some kind of final-term dissertation or project where you get to demonstrate your level of familiarity with the field, and that you have taken on board what research involves; this will not necessarily (or even generally) be expected to involve original research.

For a PhD you are expected to delve deeply into your chosen field and perform original research, under the guidance of your supervisor.

Getting accepted for a place on a Masters course will require an undergraduate degree with a high enough grade (e.g., a First). There may well also be an interview process; places will be limited, and competition strong. And Masters courses ain't cheap! Now if you cheated with AI to gain the requisite grade for a Masters, you will likely be found wanting and rejected in the interview process, or, if you make it onto the course you will simply flounder and fail. You will not get away with cheating with AI on your Masters; the degree of oversight is higher. Apart from hand-in coursework, you will at various stages be required to actually demonstrate your knowledge, be it in supervisions, study groups, or whatever. As a supervisor, I will spot it immediately (this has happened on some very rare occasions).

To be accepted for a PhD, you will require a very high standard Masters degree. You will be interviewed, possibly several times. Competition will be fierce – very fierce (you'll be competing against the best of the best internationally). You will need funding, which itself may well require demonstrating your suitability for research. You will not get away with cheating. No chance.

[LionelB]

Re: "Should you use AI for a task?"

So one incident I had with a (good) PhD student, is that I was reviewing some of her simulation code, at her request (which was fine). I came across a section of code which seemed to be doing… nothing – in a very complicated way. It didn't really impact the simulations, which in a way was almost worse. She rather sheepishly admitted that she'd been in a rush and used some LLM to generate the code. I had to explain that I didn't actually mind her using LLMs – either for coding or to help with surveying the existing literature in some area – but it was absolutely essential that she reviewed (and understood) the generated code, and always chased up references in the literature. And, of course, so long as using LLMs (in either case) was effective and efficient; i.e., useful. Which, after all the requisite reviewing, may or may not have been the case.

[Evil Auditor]

I'm going to use the same method with my trainees. I do not mind if they use GAI to get to the results. But I do mind if they are too damn lazy to properly verify the AI's output before I get to review their work.

[T. F. M. Reader]

Checking LLM assistants

Formulate a good question, "engineer the harness", etc., etc. - probably an iterative process. Assume it is done properly, a lot better than copy/pasting an essay question into a chat.

Examine the LLM response:

1. Does it make sense? [As a special case: does it mention Madagascar nonsensically?]

2. Does it answer the question?

3. Is the answer correct?

4. Is the answer complete?

5. Are there holes? Arguments against? What did the LLM miss in its argumentation?

6. What sources/references were used? Do they exist? Do they actually say what LLM says they do?

7. Are there other sources that the LLM does not list? Do they say anything substantially different or otherwise important?

What did I miss? I probably could think of more things to check or I could be more precise. The point is that whether you are writing an essay on history or plasma physics it is what the LLM does not tell you that you are supposed to learn how to do. That's what a university is supposed to teach you, and that's the only reason to pay tuition in the first place. Professors' wasting time on ingenious ways to catch students cheating is not a good way to spend taxes, donors' grants, or (other) students' tuition.

Without that there is no way you will eventually learn how to do something original and useful, e.g. how to make Madagascar float sideways through an afternoon by tilting your whisky glass at a lemur in the morning. Uh, does it make sense?

[doublelayer]

Re: Checking LLM assistants

No, it doesn't make much sense. Your complaint, "Professors' wasting time on ingenious ways to catch students cheating is not a good way to spend taxes, donors' grants, or (other) students' tuition." is a bad summary of what happened. The ingenious way they used was simple prompt salting, done for years and very publicly known, and testing the submissions could have been done with grep, or since this was probably not a grep-fluent professor, the find box and a quick human review to check that Madagascar didn't come up in the actual topic. In fact, it likely saved them grading time since the ones with Madagascar drifting could be skipped the full human grading process.

You seem to back this complaint up with a list of things that students are supposed to learn. Whether they learn it by learning to use the LLM properly or learning to write themselves, it still involves them deliberately doing work, and this episode demonstrates that many of them were unwilling to do that. Using an LLM also means they are not learning one more skill, learning to write your own opinions in your own words. Some LLM fans have argued that this is now obsolete if you do the rest of the items on your list. I don't agree, but even if someone does, that needs to be approved by the person assigning work, and otherwise it is considered cheating.

[LionelB]

Telltale signs

An acquaintance lectures in law at a well-known UK university. She says the easiest way to spot AI-generated essays tends to be that the grammar, spelling and language construction are simply too good, given her knowledge of the linguistic skills, or rather lack thereof, of the student. (Many of her students are not native English-speakers, although, depressingly, this applies too to many who are.) That, plus the bland, generic style, as well as the odd hallucinated reference (although this seems to be becoming less prevalent).

If there are sufficient grounds for suspicion, the student may be called in for a live examination; most own up at this point.

[cpzz]

Ironically, once these students graduate and enter the workforce, their manager will insist they use AI as much as possible for every task.