Can we talk about AI and concerns with college writing?

[mlktwins]

So, at one of my boys' colleges, these 2 anonymous posts (see below) have been on the school's parent page in the last week. There are all kinds of comments about this being an issue with children attending other schools as well. I'm sure it is going to be a growing issue in the years to come.

How accurate are AI detectors? Is there any recourse for students? How do they prove they didn't use it? I've shown my boys these posts and told them to keep all their notes, outlines, and all drafts for any of their papers they submit. I don't think either of my guys would ever do this, they have been told the potential consequences of getting caught, but who knows? Quoting inaccurately can be a problem? Some parents responded that even previous papers your child wrote (their own words) can be considered plagiarism on a different paper?  

For the first post, the TA refuses to go to the professor to even discuss it. For the second post, that professor is going to run the paper by several other professors to get their opinion.

Like life and college aren't hard enough already in this crazy world 😞. And I am already so stressed out trying to figure all of this out with 2 kids at once -- LOL.

Here are the posts:

I need support. My child wrote a paper. The TA sent it through the AI device and it came back as identified that my child used AI, but my child didn't. My child doesn't even know how or what to use to do that. The TA said some inappropriate things to my child and now my child has to go before the Honor Code!! Help!

AND

My DD just called me.

She met with one of her professors a short while ago after receiving an email request from the professor yesterday asking to meet.

The professor informed her that her assignment/essay was “detected as being 80% AI written”.

AI = not written by my daughter.

I am not one to believe that my daughter is always right.

However based on our phone call, and calls with her this week, I believe her.

What is the use of “AI detection” and what retribution does an “accused student” have.

She said her professor was very nice.

And that the professor is going to share the essay with other professors and decide what happens next.

My daughter is beside herself - and the research that I am doing - some this really bad phone call - does show the inaccuracies of the use of “AI detectors”.

Thoughts? Anyone else with similar situation?

What is the policy and is this a normal practice and is it common to “find and suggest/accuse a student of this?” And what happens next?

Again - she is an adult - and will be handling this if there is an accusation - but this is not favorable and a horrible experience as a student and a parent.

[mlktwins]

Just to add, my son said, "Why do you read these FB posts? They just stress you out!" 🤪I told him knowledge is power and I would rather us not be blindsided by something like this down the road. It is the same for all the posts about how hard the pre-reqs are for the College of Business at his school. They really try to weed the kids out before they apply to the College of Business (it isn't competitive, but you need a certain GPA in 9 specific classes). I said I would rather know how to proceed, keep lists of tutors that are mentioned, etc.  I would rather be proactive than reactive when we can. Ugh!!! 

I do know that maybe the people with the problems are the ones posting and we aren't hearing from all the others, but I would still rather know.

[Kassia]

I've seen this happen (infrequently) on our university parent facebook page too.  It is really awful if the student is innocent! 

[cintinative]

I found this: 

https://www.vanderbilt.edu/brightspace/2023/04/04/turnitins-new-ai-writing-detection-and-what-it-means-for-your-teaching/

"This tool’s intended purpose is to identify when student writing has been produced by an AI tool like ChatGPT. If you intend to use this new functionality, it is important that you understand that we have no insight into how it works. We cannot determine whether it is using reliable or accurate metrics, and some studies have already been conducted that suggest it may be heavily unreliable. The nature of tools which use artificial intelligence is such that they are developing quickly. False positives are not just a possibility, but an inevitability. As such, this tool cannot be fairly used as a way to assess whether a student’s work may have been written by an AI tool. We strongly recommend that any faculty member intending to use this functionality uses it as a guideline, not a grading metric. Treat any flags the tool may raise as a reason to review the student’s work further, but recognize that the technology is fallible. "

 

Cited article:  https://www.washingtonpost.com/technology/2023/04/01/chatgpt-cheating-detection-turnitin/?pwapi_token=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJzdWJpZCI6IjI0NjA4MTIzIiwicmVhc29uIjoiZ2lmdCIsIm5iZiI6MTY4MDQ5NDQwMCwiaXNzIjoic3Vic2NyaXB0aW9ucyIsImV4cCI6MTY4MTc5MDM5OSwiaWF0IjoxNjgwNDk0NDAwLCJqdGkiOiJiYzk3ZmI1Yy0zNzgxLTRhOTctYWQ0NS1mNTZlNmM0ODA3YjEiLCJ1cmwiOiJodHRwczovL3d3dy53YXNoaW5ndG9ucG9zdC5jb20vdGVjaG5vbG9neS8yMDIzLzA0LzAxL2NoYXRncHQtY2hlYXRpbmctZGV0ZWN0aW9uLXR1cm5pdGluLyJ9.xmXQMc7C2ECvD8G9-YCsYCNBlcdlN2WO_ZrSpnDMD_Y

Edited April 24, 2023 by cintinative

[JennyD]

2 hours ago, mlktwins said:

Some parents responded that even previous papers your child wrote (their own words) can be considered plagiarism on a different paper?  

This has always been the case, not only for students but also for faculty.  Students cannot reuse work they wrote for one class in a paper for another class without permission from the professor (which is sometimes granted and sometimes not).  And faculty cannot reuse work they wrote for one book or article in a different piece without (1) citing yourself, and if necessary, (2) getting permission from both publishers (which is frequently not granted).  

As for ChatGPT, it is obviously brand-new and nobody quite knows what to do with it at the moment.  There is no "normal practice" yet.  For now, I would encourage your college kids first, not to use AI to generate their papers, even first drafts.  Second, if a situation arises with a TA like the one described above, they should go straight to the professor themselves, immediately.  And finally, if a professor asks them to come and discuss their paper, stay calm and just explain how they wrote the paper, as the student did above.  

This will get sorted out but it will take a little time and adjustment. 

Edited April 24, 2023 by JennyD

[regentrude]

4 hours ago, mlktwins said:

 Some parents responded that even previous papers your child wrote (their own words) can be considered plagiarism on a different paper?  

That is standard and not technically plagiarism but academic dishonesty.  A student may not turn in work from one class as an assignment in a different class. Nothing new, and nothing to do with ai.

[Clemsondana]

It's really hard, because false accusations of cheating are horrible.  At the same time, cheating is rampant.  When I first started teaching homeschooled kids, I might catch 1-2 students out of 20 cheating, usually mild, and there were panicked, tearful apologies.  These days I probably catch at least 1/3 of the students every semester, sometimes on multiple assignments, and often have others I'm suspicious about.  At one point I contemplated switching to oral tests, with individual chats with each student, so that I didn't have to deal with any more cheating.  It takes forever to copy and paste their answers into search engines and scroll until you find the match.  This semester, I think that i have one student who is paying a college student to take the class for them, based on getting articulate, upper-level college answers with impressive vocabulary and perfect grammar on some questions but then when I ask a 2 point gimme question like 'How do you think that you did on the test?' I get something like 'i did ok i think idk'.  Maybe they are using AI, I don't know.  It's usually easier for me since I catch matches to websites and I can point to the specific link.  I don't know what will happen as more students start using AI.  I think that it will be a challenge to figure out how to catch it faster than people make workarounds available.  I know that some push to work with it, but I don't see how any person can ever become fluent in a subject if they don't actually know anything.  

But, as for how I would go about determining whether a student used AI, I'd start by having a conversation with the student.  If they can chat intelligently about the topic of their paper, then there is at least a chance that they wrote it.  They should be able to describe the subject matter and also talk a bit about organization - like, why did you put this point first?  Over time, I think more emphasis will be placed on seeing earlier drafts, or having an outline early on, or some other way of showing that they've worked on the paper over time.  

[Bootsie]

The use of AI is a hot topic on my campus right now.  I used ChatGPT to generate several papers on a prompt that I had assigned my students.  I found the results to be "accurate" but general.  It was good at integrating material from sources that were several years old.   Because I asked students to write about the major issue facing the financial markets over the next 12 months, they were required to consider much more recent information than ChatGPT used.  I have graded less than 20% of my students' papers at this point, but I haven't come across any paper that appears to me to be written by AI.  I don't run the papers through a detector; I have found plagiarism detectors to miss some of the most flagrant plagiarism issues and report many things as potential plagiarism issues which are simply commonly used combinations of words in my discipline.  One issue that I have seen with increasing frequency is that a student will essentially copy from Source A (often a low-quality blog), and then will go in and add "citations" to make it look as if they have used information from a variety of high-quality sources thinking that their writing will not match the "citation" word-for-word, so plagiarism won't be caught.  However, they don't realize when they are using something in a foreign language or a technical piece that doesn't address the issue of their paper as their "fake citation"--which is a big red flag to me to look more closely at their paper.

Professors generally don't want the hassle of bringing academic dishonesty charge forward; I think most professors would want more than some AI detector results to give them reason to believe a paper was AI generated (the paper was missing specific components asked for but was otherwise well-written, the paper is inconsistent with the other work by the student, etc.). 

[Sebastian (a lady)]

AI detectors will have a strong whack a mole aspect in that they will be aligned to one generation of AI, and then the AI will be refined (I hesitate to say get better).

One early detector looked for liveliness in the righting. AI responses often are pretty flat and predictable (because they are predictive language models). But students who aren't mature writers also often produce flat and generic papers. 

I put several essays from students I coached last year through one detector. The struggling writers were flagged as possible AI or as needing more input to make a determination. 

If I were a student, 

I would stay clear of AI, even for idea generation. 

Keep rough drafts of work and notes, for example by saving interim drafts as separate documents rather than editing over them.

Read any statements about AI or plagiarism from a syllabus and ask about anything that seems unclear or not specific. For good measure, this should include understanding what help students can get from other students on work that isn't explicitly group work. (I dislike using the term plagiarism for reusing one's own work, but students should assume it violates assignment expectations unless they have pre-approval. )

Be ready to self advocate. It's great to work with the TA or Prof of a course. But students can also reach out to the next higher level for assistance and clarity. 

If there are campus discussions of this subject, be involved. Maybe the college explanation of plagiarism needs to be refined and expanded. Maybe the process for what happens if paper is flagged needs to be clarified. Be part of the discussion and solution. (I notice for example that Bowdoin lists four types of plagiarism, including self plagiarism. They also have extensive examples of direct and mosaic plagiarism. Though I wish it included an example of acceptable paraphrasing too.)

There is a lot of cheating happening. And it's not new. I had copying when I taught coop classes over decade ago. It's not just other people's kids who are tempted or choose to use something found on the internet. 

 

[maize]

I just ran several things through this AI detector https://writer.com/ai-content-detector/

1) part of a story I wrote

2) part of a story my 10 year old son dictated

3) part of a story my 5 year old son dictated

4) a report I asked chatGPT to write

 

1 and 2 came back as 100% human generated

3 came back as 1% human generated

4 came back as 98% human generated

based on this experiment, the AI detector does a lousy job both of differentiating between not-great human generated content and AI content, and of detecting entirely AI generated content.

This, by the way, is my 5 year old's dictation (as transcribed by one of his older siblings) that the content detector thought was 99% AI:
 

"Gassin started in a house, but he went outside and saw his pet nibbling on an apple tree. But his apple tree was too tall for the pet to reach any apples, so the pet was just nibbling on the wood, but his pet was a cat because, you know, cats are playful, and if it wasn’t a cat, why would it nibble wood? An adventure was long if he got on one, his dad said, but he set off to find some wood probably. But he didn’t know what to do because the other person was gone, but this was chapter one. Right? So Gassin just went off to the wood closest to the town. So he set off to find some meat from a Toothy Rabbit in the middle of the forest. You know what you mean, are you gongna go on an adventure or two? Now you're gonna know who wrote this. Mabey. Probably… Whatever. But anyways, he didn’t know what happened, so some vikings where looking at the beach, but he was already in trouble from a Toothy, Globy, Snowy owl. It was a very gigantic, ginormous, gigantic owl. The biggest owl he had seen in his life, but he got out of trouble from it when the owl got its smallest to get (Toothy, Globy, Snowy owls can get very small.) But, by the way, you know why they can do it, right? You don’t I bet. Well, you know that, or… maybe not? Well, you don't know about the forest, the forest can get very dark. But anyways, some of you know about some forests, just not this forest. You know those bumpy digtoads in Wingfeather? There’s something based off them in this book. But it’s gonna tell you what they are, you know, I know you see a picture of one."

Edited April 25, 2023 by maize

[maize]

Chat GPT's own text classifier did a better job https://platform.openai.com/ai-text-classifier marking all of the human generated content as very unlikely to be AI gereated and the ChatGPT-generated content as unclear if it is AI generated. I'm a bit skeptical though over wether ChatGPT is only recognizing content it generated or might have generated itself vs. actually being able to recognize content generated by any AI vs a human.

[maize]

I just ran a poem through that my kids had chatGPT generate and their text classifier classified it as unlikely to be AI generated, the same classification it used for a story by one of my younger kids and an email from my 19 year old nephew.

It does not differentiate well.

[Amoret]

When I was teaching college writing, which was before AI, but when buyapaper.com type sites were big, we had to have basically incontrovertible proof of cheating for the honor board to even consider hearing a case. It is a lot of work for a faculty member to bring forth a case, and I cannot imagine that an X% chance of a paper having been plagiarized would have held any water with the board that was, even at that time, overburdened.

My colleagues and I all felt like it was on us to develop assignments that made cheating very difficult. If I were teaching now, I might try to incorporate AI into an assignment -- such as generate two papers on a given topic and write an analysis that compares their approaches, critique an AI analysis of a poem, grade an AI generated paper and defend your assessment, or something like that. We would then do a lot of the work and discussion in-class and I would require notes, outlines, individual conferences, multiple drafts, and other process materials to validate the student's original work and my grade (I always placed a significant portion of the grade on process work in any case). I would absolutely advise students to keep all process work (especially early pre-writing activities such as outlines as well as multiple drafts showing changes) -- if they ever found themselves in this situation, they would have a strong case for having done original work.

I would also suggest that they take advantage of office hours to meet with a professor or TA to discuss any essays they are writing -- even if it's just to ask for feedback on a section or to pass an idea by them -- this might be especially useful for a class where only a final draft is collected and read by the professor. Apart form proving that they are doing original work, they will probably get some helpful feedback.

Writing labs, at least those I am familiar with, also keep records of students who come in, so that would be additional evidence of process work and effort, should the need arise to prove originality. Again, it might also provide some helpful guidance.

Ironically, in my first year as a TA at a university with a top-tier rhetoric and composition department, I had a student buy a paper (I found it online and printed out the free sample portion of it). He had no process work at all. I brought it to the department head just to make sure I would have support in failing the student, and the dept. head told me to give it back and have him create process work for the plagiarized essay so he could see that "cheating was still a lot of work." I got a second opinion.

Edited April 25, 2023 by Amoret

[Farrar]

I've now had a handful of writing turned in to me that I suspect was AI or that AI helped with significantly (as in, not to clean it up or generate ideas). But I've only given one zero. In that case, the writing had additional errors (it claimed things about the text that weren't true) and I pulled a paragraph that multiple detectors said was 100% AI. None of this maybe stuff. Plus it was turned in to me after I told the student that their writing seemed like it may have had some AI help and do not do that. Student hasn't disputed the grade. So if it looks like a duck, walks like a duck, acts like a duck... I just started to feel like a dummy if I didn't finally just come right out with it and give the kid a zero.

Edited April 28, 2023 by Farrar

[Eos]

Update from dd's grad student TA experience: two of the writers genuinely didn't know they were not supposed to use it (!) and one was expelled.  The one who was expelled wrote another paper after receiving the one more strike and you're out info.  I got into an argument with my kids using words someone here had written to the effect of plagiarism might be a mistake but you can't not know if you've asked ChapGPT to write your paper.  My kids say that some students today literally do not realize it's not a citation generator or other legal assistant.  I'm skeptical but I guess I can see it that way. The posters above submitting pieces to it that they know the provenance of is quite interesting.

[Sebastian (a lady)]

19 hours ago, Eos said:

Update from dd's grad student TA experience: two of the writers genuinely didn't know they were not supposed to use it (!) and one was expelled.  The one who was expelled wrote another paper after receiving the one more strike and you're out info.  I got into an argument with my kids using words someone here had written to the effect of plagiarism might be a mistake but you can't not know if you've asked ChapGPT to write your paper.  My kids say that some students today literally do not realize it's not a citation generator or other legal assistant.  I'm skeptical but I guess I can see it that way. The posters above submitting pieces to it that they know the provenance of is quite interesting.

I have used citation generators. The appearance and prompt requirement is quite distinct from ChapGPT. 

If their argument us that students think because citation generators are permitted, AI writing generation is also allowed, that's an interesting distinction. 

I think many syllabuses will have specific AI sections in the fall. And some colleges already have a requirement for students to complete a plagiarism training as part of orientation. If AI is discussed and the student took the training, it would be hard to claim ignorance later. 

[Clemsondana]

I wouldn't be surprised if individual classes had students sign something to deal with all of the 'I didn't know' replies.  I've started doing that with other classroom guidelines.  My first quiz of the year for my high schoolers is now an open book syllabus quiz - Work is due on _______ (Fridays), Homework is ___________ but tests are ___________. (open book, closed book).  Because students insist that they didn't know, even if the policy is written at the top of the assignment.  I could also imagine it being the first answer students write for certain assignments.  Sign your name that you understand that AI programs such as X, Y, and Z may not be used/may only be used for ________ purpose on this assignment. 

I've had students copy and paste quiz answers into their tests, saying that they didn't think it was cheating since it was their work, even though the assignment isn't supposed to be open-book.  It's made easier by the fact that students who do that are also likely to have missed the question the first time, so their copy and paste also includes MY CORRECTION to the homework/quiz.  I've caught a couple of students that way - my corrections are full of slightly idiosyncratic phrases and abbreviations since they should know what I'm talking about and are not typical of how a student would answer the question in their own words.  

[regentrude]

3 hours ago, Clemsondana said:

I've had students copy and paste quiz answers into their tests, saying that they didn't think it was cheating since it was their work, even though the assignment isn't supposed to be open-book.  It's made easier by the fact that students who do that are also likely to have missed the question the first time, so their copy and paste also includes MY CORRECTION to the homework/quiz.  I've caught a couple of students that way - my corrections are full of slightly idiosyncratic phrases and abbreviations since they should know what I'm talking about and are not typical of how a student would answer the question in their own words.  

The bolded OMG. They are really that dumb?

[teachermom2834]

My dh had a completely AI generated research paper turned in for his online community college course. He recognized it immediately. The works cited page included books he knew this student wouldn’t have accessed. Ones that were not available in English, for instance, or not available in the United States. Students that don’t complete simple assignments like commenting on the message board aren’t going above and beyond to translate graduate level texts for a two page research paper. He gave the kid a zero and the response was that the student didn’t know that wasn’t an acceptable way to complete the assignment. He actually called my dh on the phone to plead his ignorance. 
 

He had busted several students for cheating throughout the course. His course evaluations largely read to the effect of “this course was so difficult I had no choice but to cheat” and blamed dh for the cheating. No mention of the fact that the students were not reading the material or logging into the course or turning anything in on time. Just they didn’t know that cheating is cheating and anyways it is the teacher’s fault anyway that cheating is necessary. 
 

I was hoping dh would keep his side gig while we got our last kid through school because the extra income is nice but the cheating was already out of hand and the AI might just be making it unbearable. 

[Eos]

5 hours ago, Clemsondana said:

  My first quiz of the year for my high schoolers is now an open book syllabus quiz - Work is due on _______ (Fridays), Homework is ___________ but tests are ___________. (open book, closed book).  Because students insist that they didn't know, even if the policy is written at the top of the assignment.

 

Genius.

[Clemsondana]

3 hours ago, regentrude said:

The bolded OMG. They are really that dumb?

More than once.  I had one student say it wasn't plagiarism because they quoted their own work.  It was cheating, but if they wanted to go that route, they quoted MY work.  One easy way for me to tell is that I often don't type all possible answers.  If they are on the wrong track, I'll type enough that they can reference the right set of material - something like 'Amount of a cellular protein can be affected by transcription factors, DNA packaging, RNA stability, etc' since there are at least 10 different things they could list.  Any time the answer includes 'etc' I double check because it's not something that students usually do.  

[Miss Tick]

Side note here, but am I the only person who reads ChatGPT but hears it with a French pronunciation? As if there is some French cat named "GPT"? I don't know how this happened, I don't speak French. I've even caught myself saying it that way out loud. I just try to avoid verbal conversations about it.

Edited May 4, 2023 by Miss Tick

[Bootsie]

I am in the midst of grading papers and some signs that something "isn't right" in a paper

1)  the student quotes Scotish economist Adam Smith.  Quote is about hyperinflation in Germany after WWI. (Smith died in 1790)

2) student has a reference of what the inflation rate was in December 2022.  Date of source--2020

[maize]

On 5/4/2023 at 4:58 PM, Bootsie said:

I am in the midst of grading papers and some signs that something "isn't right" in a paper

1)  the student quotes Scotish economist Adam Smith.  Quote is about hyperinflation in Germany after WWI. (Smith died in 1790)

2) student has a reference of what the inflation rate was in December 2022.  Date of source--2020

Oh my. That's pretty egregious. 

[cintinative]

[LauraClark]

22 minutes ago, cintinative said:

What a great assignment. Dh and I were talking today and one of my concerns is that young people tend to not believe older/wiser people. So if AI says something to a young person that an older person with life experience knows to be wrong, who will the young person believe?  

[Sebastian (a lady)]

I think the insertion of incorrect and fictitious information is a huge problem. I've played around with ChatGPT on some topics I know well.

It produces writing that is well written and authoritative sounding, but is simply wrong.

At on point, I asked it to write a bio for me. It was organized like a bio for my industry, but none of the content was correct. There was no caveat that it couldn't find info about me, but wrote a sample biography for my industry instead.