AI detectors used by educators to detect use of ChatGPT don't work, says OpenAI.
There's no way for teachers to figure out if students are using ChatGPT to cheat, OpenAI says in new back-to-school guide::AI detectors used by educators to detect use of ChatGPT don't work, says OpenAI.
Education has a fundamental incentive problem. I want to embrace AI in my classroom. I've been studying ways of using AI for personalized education since I was in grade school. I wanted personalized education, the ability to learn off of any tangent I wanted, to have tools to help me discover what I don't know so I could go learn it.
The problem is, I'm the minority. Many of my students don't want to be there. They want a job in the field, but don't want to do the work. Your required course isn't important to them, because they aren't instructional designers who recognize that this mandatory tangent is scaffolding the next four years of their degree. They have a scholarship, and can't afford to fail your assignment to get feedback. They have too many courses, and have to budget which courses to ignore. The university holds a duty to validate that those passing the courses met a level of standards and can reproduce their knowledge outside of a classroom environment. They have a strict timeline - every year they don't certify their knowledge to satisfaction is a year of tuition and random other fees to pay.
If students were going to university to learn, or going to highschool to learn, instead of being forced there by societal pressures - if they were allowed to learn at their own pace without fear of financial ruin - if they were allowed to explore the topics they love instead of the topics that are financially sound - then there would be no issue with any of these tools. But the truth is much bleaker.
Great students are using these tools in astounding ways to learn, to grow, to explore. Other students - not bad necessarily, but ones with pressures that make education motivated purely by extrinsic factors than intrinsic - have a perfect crutch available to accidentally bypass the necessary steps of learning. Because learning can be hard, and tedious, and expensive, and if you don't love it, you'll take the path of least resistance.
In game design, we talk about not giving the player the tools to optimize their fun away. I love the new wave of AI, I've been waiting for this level of natural language processing and generation capability for a very long time, but these are the tools for students to optimize the learning away. We need to reframe learning and education. We need to bring learning front and center instead of certification. Employers need to recognize this, universities need to recognize this, highschools and students and parents need to recognize this.
This is fine and all and you have a point, but in the current system many times the subject isn't about the subject it's about the auxiliary skills you pick up along the way. My history classes in high schopl weren't really about history. I mean if I retained those facts, fantastic, they were more about analyzing given evidence and multiple references to make a point. I'm an engineer and I use that skill all the time. Facts about the Civil War not so much.
Even in college I had classes like that. It's why just programming the answer wasn't always allowed although literally everyone in the university took a programming class freshmen year. That wasn't always the point.
To always allow AI is like never taking the time to teach kids how to do arithmetic by hand. I mean, sure, we could do that, but learning arithmetic is not really about memorizing times tables and more about understanding the concept of a number and internalizing counting and so much stuff people don't realize they use all the time the existence of a calculator or not.
I think there is some value in not allowing AI usage sometimes. Before you use a calculator you should learn how to do it by hand so you can have a sense of when you've keyed something in wrong. AI has entered my workplace and it's so annoying. People who never knew how to write the things they ask AI to do can't vet the AI output and the result is somehow worse to me than if they'd bumbled something by hand. That's kind of what I'm afraid of in the future. I don't think that AI is ever going to be perfect and kids have to know what output they're looking for before they're taking this shortcut.
100%, and this is really my main point. Because it should be hard and tedious, a student who doesn't really want to learn - or doesn't have trust in their education - will bypass those tedious bits with the AI rather than going through those tedious, auxiliary skills that you're expected to pick up, and use the AI was a personal tutor - not a replacement for those skills.
So often students are concerned about getting a final grade, a final result, and think that was the point, thus, "If ChatGPT can just give me the answer what was the point", but no, there were a bunch of skills along the way that are part of the scaffolding and you've bypassed them through improper use of available tools. For example, in some of our programming classes we intentionally make you use worse tools early to provide a fundamental understanding of the evolution of the language ergonomics or to understand the underlying processes that power the more advanced, but easier to use, concepts. It helps you generalize later, so that you don't just learn how to solve this problem in this programming language, but you learn how to solve the problem in a messy way that translates to many languages before you learn the powerful tools of this language. As a student, you may get upset you're using something tedious or out of date, but as a mentor I know it's a beneficial step in your learning career.
Maybe it would help to teach students about learning early, and how learning works.
I agree mostly with your comment. Math class used to be called logic, because that's really what the general population is there for. Math is a vehicle to learn logical thinking. Those who become engineers or physicists will then learn how to apply those logic skills in their chosen field.
I disagree that history classes were mainly there to analyze evidence to make a point. We learn history, so that we can better participate in discussions, know where we came from, and learn from past mistakes. It's vital that voters have an understanding of history to prevent bad things from happening. Don't get me wrong, the main point about history classes wasn't learning exact dates, but to have a good understanding of the timeline and have a good grasp of major events in the country/world. It's that part, learning details about major events that I'm concerned will be glossed over with AI coming into play. How can you recognize a destructive political trend if you never learned why it was destructive in the past?
If someone can use the tool to do the job successfully, I don't see if that learning was actually necessary. Like I learned to use a phone rather than a telegraph. I learned how to drive a car rather than ride a horse. I learned a calculator rather than a sliderule.
Of course we're still at the stage where you need to double check the tool,but that skill is maybe more like supervising someone rather than directly doing the task.
I can imagine prompt engineering will actually be a thing, and asking the AI to fix parts that don't work is the short term. We already can ask the AI to look over it's own work for mistakes, I have to imagine that's going to be built in soon...
The worse thing is if the student can actually ootimize the learning away with the AI, so too can employers optimize away the potential employees.
This is a very output-driven perspective. Another comment put it well, but essentially when we set up our curriculum we aren't just trying to get you to produce the one or two assignments that the AI could generate - we want you to go through the motions and internalize secondary skills. We've set up a four year curriculum for you, and the kinds of skills you need to practice evolve over that curriculum.
This is exactly the perspective I'm trying to get at work my comment - if you go to school to get a certification to get a job and don't care at all about the learning, of course it's nonsense to "waste your time" on an assignment that ChatGPT can generate for you. But if you're there to learn and develop a mastery, the additional skills you would have picked up by doing the hard thing - and maybe having a Chat AI support you in a productive way - is really where the learning is.
If 5 year olds can generate a university level essay on the implications of thermodynamics on quantum processing using AI, that's fun, but does the 5 year old even know if that's a coherent thesis? Does it imply anything about their understanding of these fields? Are they able to connect this information to other places?
Learning is an intrinsic task that's been turned into a commodity. Get a degree to show you can generate that thing your future boss wants you to generate. Knowing and understanding is secondary. This is the fear of generative AI - further losing sight that we learn though friction and the final output isn't everything. Note that this is coming from a professor that wants to mostly do away with grades, but recognizes larger systemic changes need to happen.
My wife teaches at a university. The title is partly bullshit:
For most teachers it couldn't be more obvious who used ChatGPT in an assignment and who didn't.
The problem, in most instances, isn't the "figuring out" part, but the "reasonably proving" part.
And that's the most frustrating part: you know an assignment was AI-written, there are no tools to prove it and the university gives its staff virtually no guidance or assistance on the subject matter, so you're almost powerless.
I agree with you for sure. However if I'm playing devil's advocate ... I think some people will fall under the pressure and perform poorly just because it's oral rather than written.
I generally think that even if that's the case that it's an important skill to teach too, but I'm just thinking of contradictions.
My coven-mate was called in by her college dean, accusing her of faking or plagiarizing her mid-term thesis. (I totally forget what the subject was. This was late 1980s. She wanted to work in national intelligence.)
But the thing is, she could expain every part of her rabbit-hole deep dive (which was a trip to several libraries and locating books themselves rather than tracking leads through the internet.) It was all fresh in her head, and to the shock and awe of her dean and instructor (delight? horror?) it was clear she was just a determined genius doing post-grad quality work because she pushed herself that hard. And yes, she was out of their league and could probably write the thesis again if that was necessary.
In our fucked up society, the US has little respect for teachers or even education so I don't expect anything real to happen, but this would be grounds to reduce classroom size by increasing faculty size so that each teacher is familiar with their fifteen students, their capabilities and ambitions and challenges at home. That way when a kid turns in an AI essay but then can't expain what the essay says, the teacher can use it as a teachable moment: point out that AI is a springboard, a place to start as a foundation for a report, but it's still important for the student to make it their own, and make sure it comes to conclusions they agree with.
Detecting whether a student used ChatGPT to write an assignment can be challenging, but there are some signs and strategies you can consider:
Unusual Language or Style: ChatGPT may produce content that is unusually advanced or complex for a student's typical writing style or ability. Look for inconsistencies in language usage, vocabulary, and sentence structure.
Inconsistent Knowledge: ChatGPT's knowledge is based on information up to its last training cut-off in September 2021. If the assignment contains information or references to events or developments that occurred after that date, it might indicate that they used an AI model.
Generic Information: If the content of the assignment seems to consist of general or widely available information without specific personal insights or original thought, it could be a sign that ChatGPT was used.
Inappropriate Sources: Check the sources cited in the assignment. If they cite sources that are unusual or not relevant to the topic, it may indicate that they generated the content using an AI model.
Plagiarism Detection Tools: Use plagiarism detection software, such as Turnitin or Copyscape, to check for similarities between the assignment and online sources. While these tools may not specifically detect AI-generated content, they can identify similarities between the assignment and publicly available text.
Interview or Discussion: Consider discussing the assignment topic with the student during a one-on-one interview or discussion. If they struggle to explain or elaborate on the content, it may indicate they didn't personally generate it.
It's important to approach these situations with caution and avoid making accusations without concrete evidence. If you suspect that a student used an AI model to complete an assignment, consider discussing your concerns with the student and offering them the opportunity to explain or rewrite the assignment in their own words.
You can tell because it's grammatically correct but logically incongruous. For example:
ChatGPT's knowledge is based on information up to its last training cut-off in September 2021. If the assignment contains information or references to events or developments that occurred after that date, it might indicate that they used an Al model.
That is the exact opposite of the conclusion you could draw.
It makes some sense. If a tool could reliably discern it, the tool would used to train the model to be more indistinguishable from regular text, putting us back to where we are now.
At the core of learning is for students to understand the content being taught. Using tools and shortcuts doesn't necessarily negate that understanding.
Using chatGPT is no different, from an acidemic evaluation standpoint, than having somebody else do an assignment.
Teachers should already be incorporating some sort of verbal q&a sessions with students to see if their demonstrated in-person comprehension matches their written comprehension. Though from my personal experience, this very rarely happens.
That's going on the supposition that a person just prompts for an essay and leaves it at that, which to be fair is likely the issue. The thing is, the genie is out of the bottle and it's not going to go back in. I think at this point it'll be better to adjust the way we teach children things, and also get to know the tools they'll be using.
I've been using GPT and LLAMA to assist me in writing emails and reports. I provide a foundation, and working with the LLMs I get a good cohesive output. It saves me time, allowing me to work on other things, and whoever needs to read the report or email gets a well-written document/letter that doesn't meander in the same way I normally do.
I essentially write a draft, have the LLMs write the whole thing, and then there's usually some back-and-forth to get the proper tone and verbiage right, as well as trim away whatever nonsense the models make up that wasn't in my original text. Essentially I act as an editor. Writing is a skill I don't really possess, but now there are tools to make up for this.
Using an LLM in that way, you're actively working with the text, and you're still learning the source material. You're just leaving the writing to someone else.
Calling it cheating is the wrong way to think about it. If you had a TI 80 whatever in the early 90s, it was practically cheating when everyone else had crap for graphing calculators.
Cat GPT used effectively isn't any different than a calculator or an electronic typewriter. It's a tool. Use it well and you'll do much better work
These hand wringing articles tell us more about the paucity of our approach to teaching and learning than they do about technology.
The policies, and more importantly, the pedagogy are out of date and basically irrelevant in an age where machines can and do create better work than the majority of university students. Teachers used to ban certain levels of calculator from their classrooms because it was considered 'cheating' (they still might). Those teacher represent a backwards approach towards preparing students for a changing world.
The future isn't writing essays independent of machine assistance just like the future of calculus isn't slide rulers.
I don't fully agree with OP but I think we could probably do with adjusting some of them. Personally I think with current AI, if somebody composes something by making multiple AI prompts and selects the best result, they should get some kind of authorship because they used a tool to create something.
Meh. You’ll do better if you actually know some math as well. No engineer is going to pull up the calculator to calculate 127+9. I hang around math-wizards all day, and it’s me who need to use the calculator, not them. I’ll tell you that much.
Same goes for writing. Sure, ChatGPT can do amazing things. But if you can’t do them yourself, you’ll struggle to spot the not so amazing things it does.
It’s always easy when you know basic math, writing and reading to say schools are doing it all wrong. But you’re already mostly fluent in what they’re teaching. With that knowledge, you can use ChatGPT as a great tool. Without that knowledge, you couldn’t.
Not a classroom setting, but I recently needed to investigate a software engineer in my team that has allegedly been using ChatGPT to do their work. My company works with critical customer data, so we're banned from using any generative AI tools.
It's really easy to tell. The accused engineer cannot explain their own code, they've been seen using ChatGPT at work, and they're stupid enough to submit code with wildly different styling when we dictate the use of a formatter to ensure our code style is consistent. It's pretty cut and dry, IMO.
I imagine that teachers will also do the same thing. My wife is a teacher, and has asked me about AI tools in the past. Her school hasn't had any issues, because it's really obvious when ChatGPT has been used - similarly to how it's obvious when someone ripped some shit off the internet and paraphrased some parts to get around web searches.
Havent read the article yet but guess teacher's best option is to go back to paper and do all of their work during class period. Not sure how they'll handle homework though or outside projects
I dont mean to be rude or anything but if I understand correctly what youre saying is dont teach them anything but how to input questions into a text field, right? Its the same old statement as "Why learn basic math if calculators exist" do you not see the problem there? Sure you get the answer but you wont understand what any of it means which will hinder the knowlege that humanity gained through the thousands of years they been on this earth. Thats a horrible take and I pray you understand why. I do agree that technology is a great tool and should be taught how to be used in school, it shouldn't be the only method of getting answers, thats just dumb and hurts everyone in the future.
Don't know why the downvote(s). Like many great technology advancements it can be used for good or for malice. AI definitely can be a great boon to society, but one of the unique aspects of this vs something like the computer or vaccines is that the tech is quite new, organizations and governments are scrambling to regulate it, and almost any fool can get their hands on it.
OpenAI is preparing teachers for the back-to-school season, releasing a guide on how to use ChatGPT in the classroom, months after educators raised the alarm on students turning to AI for cheating.
Bad news for teachers and professors though: OpenAI says that sites and apps promising to uncover AI-generated copy in students' work are unreliable.
Such content detectors also have a tendency to suggest that work by students who don't speak English as a first language is AI-generated, OpenAI stated, confirming a problem reported earlier by The Markup.
Teachers are concerned however that students are cheating by presenting ideas and phrases from the chatbot as their own, and that they are becoming over-dependent on a tool which remains prone to errors and hallucinations.
Professors began to detect students using ChatGPT to cheat on college essays a little over a month after the chatbot was released in November 2022.
OpenAI also acknowledged that ChatGPT is not free from biases and stereotypes, for instance, so "users and educators should carefully review its content."
The original article contains 360 words, the summary contains 171 words. Saved 52%. I'm a bot and I'm open source!