College professors are going back to paper exams and handwritten essays to fight students using ChatGPT::The growing number of students using the AI program ChatGPT as a shortcut in their coursework has led some college professors to reconsider their lesson plans for the upcoming fall semester.
Removed by mod
I think that’s actually a good idea? Sucks for e-learning as a whole, but I always found online exams (and also online interviews) to be very easy to game.
Prof here - take a look at it from our side.
Our job is to evaluate YOUR ability; and AI is a great way to mask poor ability. We have no way to determine if you did the work, or if an AI did, and if called into a court to certify your expertise we could not do so beyond a reasonable doubt.
I am not arguing exams are perfect mind, but I’d rather doubt a few student’s inability (maybe it was just a bad exam for them) than always doubt their ability (is any of this their own work).
Case in point, ALL students on my course with low (<60%) attendance this year scored 70s and 80s on the coursework and 10s and 20s in the OPEN BOOK exam. I doubt those 70s and 80s are real reflections of the ability of the students, but do suggest they can obfuscate AI work well.
deleted by creator
So, we’re working on a study for online Vs in person.
We’ve noticed that stiict time limits alone tend to shift grades. A locked down browser sounds great, but anyone can search using their phone, so proctoring is a must (but also time consuming to check) if you want to get the intended effect.
As for online grading, it’s a mixed bag. With a very strict rubric, gradescope can save a lot of time, but otherwise it takes a lot longer. MCQs and single number answers can be auto-matked, but they’re awful at assessing ability and should be avoided. Overall, grading online costs more than it saves, and tends to give much more rigid feedback to students.
deleted by creator
You could indeed go for such a setup, however, in a room with 50+ students it becomes very hard to angle a camera with a clear view on all of them, their computer screens, and under their desks. It’s easier just to walk around the room to invigilate. However, I might have a read up on this as it might be an option for students with exam anxiety (I realise we look scary walking around the exam room!).
deleted by creator
Here’s a somewhat tangential counter, which I think some of the other replies are trying to touch on … why, exactly, continue valuing our ability to do something a computer can so easily do for us (to some extent obviously)?
In a world where something like AI can come up and change the landscape in a matter of a year or two … how much value is left in the idea of assessing people’s value through exams (and to be clear, I’m saying this as someone who’s done very well in exams in the past)?
This isn’t to say that knowing things is bad or making sure people meet standards is bad etc. But rather, to question whether exams are fit for purpose as means of measuring what matters in a world where what’s relevant, valuable or even accurate can change pretty quickly compared to the timelines of ones life or education. Not long ago we were told that we won’t have calculators with us everywhere, and now we could have calculators embedded in our ears if wanted to. Analogously, learning and examination is probably being premised on the notion that we won’t be able to look things up all the time … when, as current AI, amongst other things, suggests, that won’t be true either.
An exam assessment structure naturally leans toward memorisation and being drilled in a relatively narrow band of problem solving techniques,1 which are, IME, often crammed prior to the exam and often forgotten quite severely pretty soon afterward. So even presuming that things that students know during the exam are valuable, it is questionable whether the measurement of value provided by the exam is actually valuable. And once the value of that information is brought into question … you have to ask … what are we doing here?
Which isn’t to say that there’s no value created in doing coursework and cramming for exams. Instead, given that a computer can now so easily augment our ability to do this assessment, you have to ask what education is for and whether it can become something better than what it is given what are supposed to be the generally lofty goals of education.
In reality, I suspect (as many others do) that the core value of the assessment system is to simply provide a filter. It’s not so much what you’re being assessed on as much as your ability to pass the assessment that matters, in order to filter for a base level of ability for whatever professional activity the degree will lead to. Maybe there are better ways of doing this that aren’t so masked by other somewhat disingenuous goals?
Beyond that there’s a raft of things the education system could emphasise more than exam based assessment. Long form problem solving and learning. Understanding things or concepts as deeply as possible and creatively exploring the problem space and its applications. Actually learn the actual scientific method in practice. Core and deep concepts, both in theory and application, rather than specific facts. Breadth over depth, in general. Actual civics and knowledge required to be a functioning member of the electorate.
All of which are hard to assess, of course, which is really the main point of pushing back against your comment … maybe we’re approaching the point where the cost-benefit equation for practicable assessment is being tipped.
- In my experience, the best means of preparing for exams, as is universally advised, is to take previous or practice exams … which I think tells you pretty clearly what kind of task an exam actually is … a practiced routine in something that narrowly ranges between regurgitation and pretty short-form, practiced and shallow problem solving.
Ah the calculator fallacy; hello my old friend.
So, a calculator is a great shortcut, but it’s useless for most mathematics (i.e. proof!). A lot of people assume that having a calculator means they do not need to learn mathematics - a lot of people are dead wrong!
In terms of exams being about memory, I run mine open book (i.e. students can take pre-prepped notes in). Did you know, some students still cram and forget right after the exams? Do you know, they forget even faster for courseworks?
Your argument is a good one, but let’s take it further - let’s rebuild education towards an employer centric training system, focusing on the use of digital tools alone. It works well, productivity skyrockets, for a few years, but the humanities die out, pure mathematics (which helped create AI) dies off, so does theoretical physics/chemistry/biology. Suddenly, innovation slows down, and you end up with stagnation.
Rather than moving us forward, such a system would lock us into place and likely create out of date workers.
At the end of the day, AI is a great tool, but so is a hammer and (like AI today), it was a good tool for solving many of the problems of its time. However, I wouldn’t want to only learn how to use a hammer, otherwise how would I be replying to you right now?!?
let’s rebuild education towards an employer centric training system, focusing on the use of digital tools alone. It works well, productivity skyrockets, for a few years, but the humanities die out, pure mathematics (which helped create AI) dies off, so does theoretical physics/chemistry/biology. Suddenly, innovation slows down, and you end up with stagnation.
Rather than moving us forward, such a system would lock us into place and likely create out of date workers.
I found this too generalizing. Yes, most people only ever need and use productivity skills in their worklife. They do no fundamental research. Wether their education was this or that way has no effect on the advancement of science in general, because these people don’t do science in their career.
Different people with different goals will do science, and for them an appropriate education makes sense. It also makes sense to have everything in between.
I don’t see how it helps the humanities and other sciences to teach skills which are never used. Or how it helps to teach a practice which no one applies in practice. How is it a threat to education when someone uses a new tool intelligently, so they can pass academic education exams? How does that make them any less valuable for working in that field? Assuming the exam reflects what working in that field actually requires.
I think we can also spin an argument in the opposite direction: More automation in education frees the students to explore side ideas, to actually study the field.
“I don’t see how it helps the humanities and other sciences to teach skills which are never used.” - I can offer an unusual counter here, you’re assuming the knowledge will never be used, or that we should avoid teaching things that are unlikely to be used. Were this the case, the field of graph theory would have ceased to exist long before it became useful in mapping - indeed Bool’s algebra would never have led to the foundations of computer science and the machines we are using today.
“How is it a threat to education when someone uses a new tool intelligently, so they can pass academic education exams?” - Allow me to offer you the choice of two doctors, one of whome passed using AI, and the other passed a more traditional assessment. Which doctor would you choose and why? Surely the latter, since they would have also passed with AI, but the one without AI might not have passed the more traditional route due to a lack of knowledge. It isn’t a threat to education, it’s adding further uncertainty as to the outcome of such an education (both doctors might have the same skill levels, but there is more room for doubt in the first case).
“Wether their education was this or that way has no effect on the advancement of science in general, because these people don’t do science in their career.” - I strongly disagree! In an environment where knowledge for the sake of knowledge is not prised, a lie is more easy plant and nurture (take for example the antivax movement). Such people can be an active hinderence to the progress of knowledge - their misconceptions creating false leads and creating an environment that distrusts such sciences (we’re predisposed to distrust what we do not understand).
you’re assuming the knowledge will never be used, or that we should avoid teaching things that are unlikely to be used.
Not exactly. What I meant to say is: Some students will never use some of the knowledge they were taught. In the age of information explosion, there is practically unlimited knowledge ‘available’. What part of this knowledge should be taught to students? For each bit of knowledge, we can make your hypothetic argument: It might become useful in the future, an entire important branch of science might be built on top of it.
So this on it’s own is not an argument. We need to argue why this particular skill or knowledge deserves the attention and focus to be studied. There is not enough time to teach everything. Which in turn can be used as an argument to more computer assisted learning and teaching. For example, I found ChatGPT useful to explore topics. I would not have used it to cheat in exams, but probably to prepare for them.
the choice of two doctors, one of whome passed using AI, and the other passed a more traditional assessment. Which doctor would you choose and why? Surely the latter, since they would have also passed with AI, but the one without AI might not have passed the more traditional route due to a lack of knowledge.
Good point, but it depends on context. You assume the traditional doc would have passed with AI, but that is questionable. These are complex tools with often counterintuitive behaviour. They need to be studied and approached critically to be used well. For example, the traditional doc might not have spotted the AI hallucinating, because she wasn’t aware of that possibility.
Further, it depends on their work environment. Do they treat patients with, or without AI? If the doc is integrated in a team of both human and artificial colleagues, I certainly would prefer the doc who practiced these working conditions, who proved in exams they can deliver the expected results this way.
In an environment where knowledge for the sake of knowledge is not prised
I feel we left these lands in Europe when diplomas were abandoned for the bachelor/master system, 20 years ago. Academic education is streamlined, tailored to the needs of the industry. You can take a scientific route, but most students don’t. The academica which you describe as if it was threatened by something new might exist, but it lives along a more functional academia where people learn things to apply them in our current reality.
It’s quite a hot take to paint things like the antivax movement on academic education. For example, I question wether the people proposing and falling for these ‘ideas’ are academics in the first place.
Personally, I like learning knowledge for the sake of knowledge. But I need time and freedom to do so. When I was studying computer science with an overloaded schedule, my interest in toying with ideas and diving into backgrounds was extremely limited. I also was expected to finish in an unreasonably short amount of time. If I could have sped up some of the more tedious parts of the studies with the help of AI, this could have freed up resources and interest for the sake of knowledge.
You use literally everything you learn; it shapes your worldview and influences everything you do, especially how you vote. Don’t tell us that useless knowledge exists. It all has inherent worth.
Yes, within limits. Due to the information explosion, it became impossible to learn “everything”. We need to make choices, prioritize.
How does your voting behaviour suffer because you lack understanding about how exactly potentiometers work, or how to express historic events in modern dance?
Both have inherent worth, but not the same for each person and context. We luckily live in a society of labor division. Not everyone has to know or like everything. While I absolutely admire science, not everyone has to be a scientist.
Because there is more knowledge available than we can ever teach a single person, it is entirely possible to spend a lifetime learning things with no use informing your ballot decision. I would much rather have students optimize some parts of their education with AI, to free up capacity for other important subjects which may seem less related to their discipline. For example, many of my fellow computer science students were completely unaware how it could be ethically questionable to develop pathfinding algorithms for military helicopters.
So … I honestly think this is a problematic reply … I think you’re being defensive (and consequently maybe illogical), and, honestly, that would be the red flag I’d look for to indicate that there’s something rotten in academia. Otherwise, there might be a bit of a disconnect here … thoughts:
- The
calculatorwas in reference to arithmetic and other basic operations and calculations using them … not higher level (or actual) mathematics. I think that was pretty clear and I don’t think there’s any “fallacy” here, like at all. - The
value of learning (actual) mathematicsis pretty obvious I’d say … and was pretty much stated in my post about alternatives to emphasise. On which, getting back to my essential point … how would one best learn and be assessed on their ability to construct proofs in mathematics? Are timed open book exams (and studying in preparation for them) really the best we’ve got!? Still forgetting with open book exams… seems like an obvious outcome as the in-exam materials de-emphasise memory … they probably never knew the things you claim they forget in the first place. Why, because the exam only requires the students to be able to regurgitate in the exam, which is the essential problem, and for which in-exam materials are a perfect assistant. Really not sure what the relevance of this point is.Forgetting after coursework… how do you know this (genuinely curious)? Even so, course work isn’t the great opposite to exams. Under the time crunch of university, they are also often crammed, just not in an examination hall. The alternative forms of education/assessment I’m talking about are much more long-form and exploration and depth focused. The most I’ve ever remembered from a single semester subject came from when I was allowed to pursue a single project for the whole subject. Also, I didn’t mention ordinary general coursework in my post, as, again, it’s pretty much the same paradigm of education as exams, just done at home for the most part.Rebuilding education toward employer centric training system… I … ummm … never suggested this … I suggested the opposite … only things that were far more “academic” than this and were never geared toward “productivity”. This is a pretty bad staw man argument for a professor to be making, especially given that it seems constructed to conclude that the academy and higher learning are essential for the future success of the economy (which I don’t disagree with or even question in my post).- You speak about AI a lot. Maybe your whole reply was solely to the whole calculator point I made. This, I think, misses the broader point, which most of my post was dedicated to. That is, this isn’t about us now needing to use AI in education (I personally don’t buy that at all for probably much of the same reason you’d push back on it). Instead, it’s about what it means about our education system that AI can kinda do the thing we’re using to assess ourselves … on which I say, it tells us that the value of assessment system we take pretty seriously ought to be questioned, especially, as I think we both agree on, given the many incredibly valuable things education has to offer the individual and society at large. In my case, I go further and say that the assessment system is and has already detracted from these potential offerings, and that it does not bode well for modern western society that it seems to be leaning into the assessment system rather than expanding its scope.
- The
I think a central point you’re overlooking is that we have to be able to assess people along the way. Once you get to a certain point in your education you should be able to solve problems that an AI can’t. However, before you get there, we need some way to assess you in solving problems that an AI currently can. That doesn’t mean that what you are assessed on is obsolete. We are testing to see if you have acquired the prerequisites for learning to do the things an AI can’t do.
- AI isn’t as important to this conversation as I seem to have implied. The issue is us, ie humans, and what value we can and should seek from our education. What AI can or can’t do, IMO, only affects vocational aspects in terms of what sorts of things people are going to do “on the job”, and, the broad point I was making in the previous post, which is that AI being able to do well at something we use for assessment is an opportunity or prompt to reassess the value of that form of assessment.
- Whether AI can do something or not, I call into question the value of exams as a form of assessment, not assessment itself. There are plenty of other things that could be used for assessment or grading someone’s understanding and achievement.
- The real bottom line on this is cost and that we’re a metric driven society. Exams are cheap to run and provide clean numbers. Any more substantial form of assessment, however much they better target more valuable skills or understanding, would be harder to run. But again, I call into question how valuable all of what we’re doing actually is compared to what we could be doing, however more expensive and whether we should really try to focus more on what we humans are good at (and even enjoy).
AI can’t do jack shit with any meaningful accuracy anyway so it’s stupid to compare human education to AI blatantly making shit up like it always does
As they are talking about writing essays, I would argue the importance of being able to do it lies in being able to analyze a book/article/whatever, make an argument, and defend it. Being able to read and think critically about the subject would also be very important.
Sure, rote memorization isn’t great, but neither is having to look something up every single time you ever need it because you forgot. There are also many industries in which people do need a large information base as close recall. Learning to do that much later in life sounds very difficult. I’m not saying people should memorize everything, but not having very many facts about that world around you at basic recall doesn’t sound good either.
Learning to do that much later in life sounds very difficult
That’s an interesting point I probably take for granted.
Nonetheless, exercising memory is probably something that could be done in a more direct fashion, and therefore probably better, without that concern affecting the way we approach all other forms of education.
Here’s a somewhat tangential counter, which I think some of the other replies are trying to touch on … why, exactly, continue valuing our ability to do something a computer can so easily do for us (to some extent obviously)?
My theory prof said there would be paper exams next year. Because it’s theory. You need to be able to read an academic paper and know what theoretical basis the authors had for their hypothesis. I’m in liberal arts/humanities. Yes we still exist, and we are the ones that AI can’t replace. If the whole idea is that it pulls from information that’s already available, and a researcher’s job is to develop new theories and ideas and do survey or interview research, then we need humans for that. If I’m trying to become a professor/researcher, using AI to write my theory papers is not doing me or my future students any favors. Ststistical research on the other hand, they already use programs for that and use existing data, so idk. But even then, any AI statistical analysis should be testing a new hypothesis that humans came up with, or a new angle on an existing one.
So idk how this would affect engineering or tech majors. But for students trying to be psychologists, anthropologists, social workers, professors, then using it for written exams just isn’t going to do them any favors.
I also used to be a humanities person. The exam based assessments were IMO the worst. All the subjects assessed without any exams were by far the best. This was before AI BTW.
If you’re studying theoretical humanities type stuff, why can’t your subjects be assessed without exams? That is, by longer form research projects or essays?
In my experience, the best means of preparing for exams, as is universally advised, is to take previous or practice exams … which I think tells you pretty clearly what kind of task an exam actually is … a practiced routine in something that narrowly ranges between regurgitation and pretty short-form, practiced and shallow problem solving.
You are getting some flak, but imho you are right. The only thing an exam really tests is how well you do in exams. Of course, educators dont want to hear that. But if you take a deep dive into (scientific) literature on the topic, the question “What are we actually measuring here?” is raised rightfully so.
Getting flak on social media, through downvotes, can often (though not always!) be a good thing … means you’re touching a nerve or something.
On this point, I don’t think I’ve got any particularly valuable or novel insights, or even any good solutions … I’m mostly looking for a decent conversation around this issue. Unfortunately, I suspect, when you get everyone to work hard on something and give them prestigious certifications for succeeding at that something, and then do this for generations, it can be pretty hard to convince people to not assign some of their self-worth to the quality/value/meaning of that something and to then dismiss it as less valuable than previously thought. Possibly a factor in this conversation, which I say with empathy.
Any links to some literature?
Used only papers in german so far, sadly.
Here is something I found interesting in english:
Testing the test: Are exams measuring understanding? Brian K. Sato, Cynthia F. C. Hill, S. Lo Biochemistry and Molecular Biology Education
in general: elicit.org
really good site.
Hadn’t heard of that
elicitcite … thanks! How have you found it? It makes sense that it exists already, but I hadn’t really thought about it (haven’t looked up papers recently but may soon).Also thanks for the paper!!
Have found it relatively early after it was created, using it for getting a quick overview over papers when writing my own. It is sooo good for that.
Yea cool. Thanks for this!
It’s an interesting point… I do agree memorisation is (and always has been) used as more of a substitute for actual skills. It’s always been a bugbear of mine that people aren’t taught to problem solve, just regurgitate facts, when facts are literally at our fingertips 24/7.
Yea, it isn’t even a new problem. The exam was questionable before AI.
Removed by mod
Removed by mod
Removed by mod
Removed by mod
Removed by mod
In my experience, they love to give exams where it doesn’t matter what notes you bring, you’re on the same level whether you write down only the essential equations, or you copy down the whole textbook.
Graduated a year ago, just before this AI craze was a thing.
I feel there’s a social shift when it comes to education these days. It’s mostly: “do 500 - 1,000 word essay to get 1.5% of your grade”. The education doesn’t matter anymore, the grades do; if you pick something up along the way, great! But it isn’t that much of a priority.
I think it partially comes from colleges squeezing students of their funds, and indifferent professors who just assign busywork for the sake of it. There are a lot of uncaring professors that just throw tons of work at students, turning them back to the textbook whenever they ask questions.
However, I don’t doubt a good chunk of students use AI on their work to just get it out of the way. That really sucks and I feel bad for the professors that actually care and put effort into their classes. But, I also feel the majority does it in response to the monotonous grind that a lot of other professors give them.
I recently finished my degree, and exam-heavy courses were the bane of my existence. I could sit down with the homework, work out every problem completely with everything documented, and then sit to an exam and suddenly it’s “what’s a fluid? What’s energy? Is this a pencil?”
The worst example was a course with three exams worth 30% of the grade, attendance 5% and homework 5%. I had to take the course twice; 100% on HW each time, but barely scraped by with a 70.4% after exams on the second attempt. Courses like that took years off my life in stress. :(
If you don’t mind me asking - what kind of degree was it, and what format were the exams?
Case in point, ALL students on my course with low (<60%) attendance this year scored 70s and 80s on the coursework and 10s and 20s in the OPEN BOOK exam. I doubt those 70s and 80s are real reflections of the ability of the students
I get that this is a quick post on social media and only an antidote, but that is interesting. What do you think the connection is? AI, anxiety, or something else?
It’s a tough one because I cannot say with 100% certainty that AI is the issue. Anxiety is definitely a possibility in some cases, but not all; perhaps thinking time might be a factor, or even just good old copying and then running the work through a paraphraser. The large amount of absenses also means it was hard to benchmark those students based on class assessment (yes, we are always tracking how you are doing in class, not tp judge you, but just in case you need some extra help!).
However, AI is a strong contender since the “open book” part didn’t include the textbook, it allowed the students to take a booklet into the exams with their own notes (including fully worked examples). They scored low because they didn’t understand their own notes, and after reviewing the notes they brought in (all word perfect), it was clear they did not understand the subject.
That sounds like AI. If you do your homework then even sitting in a regular exam you should score better than 20%. This exam being open book, it sounds like they were unfamiliar with the textbook and could not find answers fast enough.
Not the previous poster. I taught an introduction to programming unit for a few semesters. The unit was almost entirely portfolio based ie all done in class or at home.
The unit had two litmus tests under exam like conditions, on paper in class. We’re talking the week 10 test had complexity equal to week 5 or 6. Approximately 15-20% of the cohort failed this test, which if they were up to date with class work effectively proved they cheated. They’d be submitting course work of little 2d games then on paper be unable to “with a loop, print all the odd numbers from 1 to 20”
We have no way to determine if you did the work, or if an AI did, and if called into a court to certify your expertise we could not do so beyond a reasonable doubt.
Could you ever though, when giving them work they had to do not in your physical presence? People had their friends, parents or ghostwriters do the work for them all the time. You should know that.
This is not an AI problem, AI “just” made it far more widespread and easier to access.
“Sometimes” would be my answer. I caught students who colluded during online exams, and even managed to spot students pasting directly from an online search. Those were painful conversations, but I offered them resits and they were all honest and passed with some extra classes.
With AI, detection is impossible at the moment.
Student here - How does that cursive longhand thing go again?
“Avoid at all costs because we hate marking it even more than you hate writing it”?
An in person exam can be done in a locked down IT lab, and this leads to a better marking experience, and I suspect a better exam experience!
Is AI going to go away?
In the real world, will those students be working from a textbook, or from a browser with some form of AI accessible in a few years?
What exactly is being measured and evaluated? Or has the world changed, and existing infrastructure is struggling to cling to the status quo?
Were those years of students being forced to learn cursive in the age of the computer a useful application of their time? Or math classes where a calculator wasn’t allowed?
I can hardly think just how useful a programming class where you need to write it on a blank page of paper with a pen and no linters might be, then.
Maybe the focus on where and how knowledge is applied needs to be revisited in light of a changing landscape.
For example, how much more practically useful might test questions be that provide a hallucinated wrong answer from ChatGPT and then task the students to identify what was wrong? Or provide them a cross discipline question that expects ChatGPT usage yet would remain challenging because of the scope or nuance?
I get that it’s difficult to adjust to something that’s changed everything in the field within months.
But it’s quite likely a fair bit of how education has been done for the past 20 years in the digital age (itself a gradual transition to the Internet existing) needs major reworking to adapt to changes rather than simply oppose them, putting academia in a bubble further and further detached from real world feasibility.
If you’re going to take a class to learn how to do X, but never actually learn how to do X because you’re letting a machine do all the work, why even take the class?
In the real world, even if you’re using all the newest, cutting edge stuff, you still need to understand the concepts behind what you’re doing. You still have to know what to put into the tool and that what you get out is something that works.
If the tool, AI, whatever, is smart enough to accomplish the task without you actually knowing anything, what the hell are you useful for?
But that’s actually most of the works we have nowadays. IA is replacing repetitive works such as magazine writers or script writers
Writers are repetitive work???
Well, it seems they will be replaced, at least certain writers. https://www.npr.org/2023/05/20/1177366800/striking-movie-and-tv-writers-worry-that-they-will-be-replaced-by-ai Also, callcenters https://www.bbc.com/news/business-65906521 And junior programmers. The problem here it’s not my opinion, those already happened so its not debatable.
Sorry but it was never about OUR abilility in the firts place.
In my country exams are old, outdated and often way to hard. In my country all classes are outdated and way to hard. It often feels that we are stucked in the middle of the 20th century.
You have no change when you have a disability. When you have kids, parents to take care of. Or hell: you have to work, cause you can’t effort university otherwise.
So i can totaly understand why students feel the need to use AI to survive that torture. I don’t feel sorry for an outdated university system.
When it is about OUR abilility, then create a System that is for students and their needs.
If you can use AI to do the homework for you, we can use AI to do the job instead of you
For the love of god, think beyond yourself for one damn minute
deleted by creator
AI was never about making our lives easier. It’s been a corporate tool out of the gate.
But this kind of thing really should be handled by Washington, and maybe it would be if we had a functional Congress. Make it illegal for AI companies to sell their services to people for the purpose of cheating or impersonation.
Make it illegal for AI companies to sell their services to people for the purpose of cheating or impersonation.
How would that work? How do you know someone is cheating? AI can be a great studying tool and those same functions could be considered cheating based on the context that the user is using them in. There’s no way to tell what the user’s intent is.
They’re about to find out that gen Z has horrible penmanship.
You’d be surprised. My daughter (13) has better penmanship than I do (46). Although I’m sure my left-handedness doesn’t help there.
There are places where analog exams went away? I’d say Sweden has always been at the forefront of technology, but our exams were always pen-and-paper.
Same in Germany
Norway has been pushing digital exams for quite a few years, to the point where high school exams went to shit for lots of people this year because the system went down and they had no backup (who woulda thought?). In at least some universities most of or all exams have been digital for a couple years.
I think this is largely a bad idea, especially on engineering exams, or any exam where you need to draw/sketch or write equations. For purely textual exams, it’s fine. This has also lead to much more multiple-choice or otherwise automatically corrected questions, which the universities explicitly state is a way of cutting costs. I think that’s terrible, nothing at university level should be reduced to a multiple choice question. They should be forbidden.
The university I went to explicitly did in person written exams for pretty much all exams specifically for anti-cheating (even before the age of ChatGPT). Assignments would use computers and whatnot, but the only way to reliably prevent cheating is to force people to write the exams in carefully controlled settings.
Honestly, probably could have still used computers in controlled settings, but pencil and paper is just simpler and easier to secure.
One annoying thing is that this meant they also usually viewed assignments as untrusted and thus not worth much of the grade. You’d end up with assignments taking dozens of hours but only worth, say, 15% of your final grade. So practically all your grade is on a couple of big, stressful exams. A common breakdown I had was like 15% assignments, 15% midterm, and 70% final exam.
Covid forced the transition to electronic exams in many areas.
Here we went back as quickly as possible because electronic exams are hell
This isn’t exactly novel. Some professors allow a cheat sheet. But that just means that the exam will be harder.
Physics exam that allows a cheat sheet asks you to derive the law of gravity. Well, OK, you write the answer at the bottom pulled from you cheat sheet. Now what? If you recall how it was originally created you probably write Newtons three laws at the top of your paper… And then start doing some math.
Calculus exam that let’s you use wolfram alpha? Just a really hard exam where you must show all of your work.
Now, with ChatGPT, it’s no longer enough to have a take home essay to force students to engage with the material, so you find news ways to do so. Written, in person essays are certainly a way to do that.
Or you have the classic, “you can write anything down that you’d like, but it won’t tell you how to answer the questions”.
And, in fact, it doesn’t help at all beyond a few formulas. I was ChemE, our cheat sheets never saved us.
You can always write down what gpt shows on the screen onto a paper
Not in class though. The idea is teaching has always had to adjust to new technology, realities and goals.
Hate to break it to you, but you picked probably the one law in physics that is empirically derived. There is no mathematical equation to derive newton’s law of gravity.
Yes but you can still start with Kepler and newton’s three laws and with basic math skills recreate the equation. I know, because it was on a physics exam I took ten years ago.
Am I wrong in thinking student can still generate an essay and then copy it by hand?
Not during class. Most likely a proctored exam. No laptops, no phones, teacher or proctor watching.
…then why can’t you do that with a school laptop that can’t access the web…?
Because laptops cost money lol
And so do colleges. If they don’t want to invest $2000 every 5/6 years for a hundred dumpster windows 95 PCs it shouldn’t be the paying student to suffer.
Only if you memorize it. In which case, why not just memorize the test material?
My first thought was exactly this.
Sounds like effort, I’m making a font out of my handwriting and getting a 3d printer to write it
Obviously that is the next step for the technically inclined, but even the less inclined may be capable of generating them copying to save time and brain effort.
You can still have AI write the paper and you copy it from text to paper. If anything, this will make AI harder to detect because it’s now AI + human error during the transferring process rather than straight copying and pasting for students.
Noooo. That’s a genious countermeasure without any obvious drawbacks!!1! /s
When I was in College for Computer Programming (about 6 years ago) I had to write all my exams on paper, including code. This isn’t exactly a new development.
So what you’re telling me is that written tests have, in fact, existed before?
What are you some kind of education historian?
He’s not pointing out that handwritten tests are not something new, but that using handwritten tests over typing them to reflect the student’s actual abilities is not new.
I had some teachers ask for handwritten programming exams too (that was more like 20 years ago for me) and it was just as dumb then as it is today. What exactly are they preparing students for? No job will ever require the skill of writing code on paper.
deleted by creator
I personally never had a problem performing well in those tests, I happen to have the skill to compile code in my head, and it is a helpful skill in my job (I’ve been a software engineer for 19 years now), but it’s definitely not a required skill and should not be considered as such.
Education is not just for getting a job, you dink.
Same. All my algorithms and data structures courses in undergrad and grad school had paper exams. I have a mixed view on these but the bottom line is that I’m not convinced they’re any better.
Sure they might reflect some of the student’s abilities better, but if you’re an evaluator interested in assessing student’s knowledge a more effective way is to make directed questions.
What ends up happening a lot of times are implementation questions that ask from the student too much at once: interpretation of the problem; knowledge of helpful data structures and algorithms; abstract reasoning; edge case analysis; syntax; time and space complexities; and a good sense of planning since you’re supposed to answer it in a few minutes without the luxury and conveniences of a text editor.
This last one is my biggest problem with it. It adds a great deal of difficulty and stress without adding any value to the evaluator.
Can we just go back to calling this shit Algorithms and stop pretending its actually Artificial Intelligence?
It actually is artificial intelligence. What are you even arguing against man?
Machine learning is a subset of AI and neural networks are a subset of machine learning. Saying an LLM (based on neutral networks for prediction) isn’t AI because you don’t like it is like saying rock and roll isn’t music
If AI was ‘intelligent’, it wouldn’t have written me a set of instructions when I asked it how to inflate a foldable phone. Seriously, check my first post on Lemmy…
https://lemmy.world/post/1963767
An intelligent system would have stopped to say something like “I’m sorry, that doesn’t make any sense, but here are some related topics to help you”
AI doesn’t necessitate a machine even being capable of stringing the complex English language into a series of steps towards something pointless and unattainable. That in itself is remarkable, however naive it may be in believing you that a foldable phone can be inflated. You may be confusing AI for AGI, which is when the intelligence and reasoning level is at or slightly greater than humans.
The only real requirement for AI is that a machine take actions in an intelligent manner. Web search engines, dynamic traffic lights, and Chess bots all qualify as AI, despite none of them being able to tell you rubbish in proper English
The only real requirement for AI is that a machine take actions in an intelligent manner.
There’s the rub: defining “intelligent”.
If you’re arguing that traffic lights should be called AI, then you and I might have more in common than we thought. We both believe the same things: that ChatGPT isn’t any more “intelligent” than a traffic light. But you want to call them both intelligent and I want to call neither so.
I’m with you on this and think the AI label is just stupid and misleading. But times/language change and you end up being a Don Quixote type figure.
I think you’re conflating “intelligence” with “being smart”.
Intelligence is more about taking in information and being able to make a decision based on that information. So yeah, automatic traffic lights are “intelligent” because they use a sensor to check for the presence of cars and “decide” when to switch the light.
Acting like some GPT is on the same level as a traffic light is silly though. On a base level, yes, it “reads” a text prompt (along with any messaging history) and decides what to write next. But that decision it’s making is much more complex than “stop or go”.
I don’t know if this is an ADHD thing, but when I’m talking to people, sometimes I finish their sentences in my head as they’re talking. Sometimes I nail it, sometimes I don’t. That’s essentially what chatGPT is, a sentence finisher that happened to read a huge amount of text content on the web, so it’s got context for a bunch of things. It doesn’t care if it’s right and it doesn’t look things up before it says something.
But to have a computer be able to do that at all?? That’s incredible, and it took over 50 years of AI research to hit that point (yes, it’s been a field in universities for a very long time, with most that time people saying it’s impossible), and we only hit it because our computers got powerful enough to do it at scale.
deleted by creator
Intelligence is more about taking in information and being able to make a decision based on that information.
Where does that come from? A better gauge for intelligence is whether someone or something is able to resolve a problem that they did not encounter before. And arguably all current models completely suck at that.
I also think the word “AI” is used quite a bit too liberal. It confuses people who have zero knowledge on the topic. And when an actual AI comes along we will have to make up a new word because “general artificial intelligence” won’t be distinctive enough for corporations to market their new giant leap in technology….
I would suggest the textbook Artificial Intelligence: A Modern Approach by Russell and Norvig. It’s a good overview of the field and has been in circulation since 1995. https://en.m.wikipedia.org/wiki/Artificial_Intelligence:_A_Modern_Approach
Here’s a photo, as an example of how this book approaches the topic, in that there’s an entire chapter on it with sections on four approaches, and that essentially even the researchers have been arguing about what intelligence is since the beginning.
But all of this has been under the umbrella of AI. Just because corporations have picked up on it, doesn’t invalidate the decades of work done by scientists in the name of AI.
My favourite way to think of it is this: people have forever argued whether or not animals are intelligent or even conscious. Is a cat intelligent? Mine can manipulate me, even if he can’t do math. Are ants intelligent? They use the same biomechanical constructs as humans, but at a simpler scale. What about bacteria? Are viruses alive?
If we can create an AI that fully simulates a cockroach, down to every firing neuron, does it mean it’s not AI just because it’s not simulating something more complex, like a mouse? Does it need to exceed a human to be considered AI?
If it shouldn’t be called AI or not no idea…
But some humans are intel·ligent and let’s be clear…say crazier things…
Inflating a phone is super easy though!
Overheat the battery. ;) Phone will inflate itself!
I am arguing against this marketing campaign, that’s what. Who decides what “AI” is and how did we come to decide what fits that title? The concept of AI has been around a long time, like since the Greeks, and it had always been the concept of a made-made man. In modern times, it’s been represented as a sci-fi fantasy of sentient androids. “AI” is a term with heavy association already cooked into it. That’s why calling it “AI” is just a way to make it sound high tech futuristic dreams-come-true. But a predictive text algorithm is hardly “intelligence”. It’s only being called that to make it sound profitable. Let’s stop calling it “AI” and start calling out their bullshit. This is just another crypto currency scam. It’s a concept that could theoretically work and be useful to society, but it is not being implemented in such a way that lives up to its name.
Who decides what “AI” is
Appearently you.
The field of computer science decided what AI is. It has a very specific set of meanings and some rando on the Internet isn’t going to upend decades of usage just because it doesn’t fit their idea of what constitutes AI or because they think it’s a marketing gimmick.
It’s not. It’s a very specific field in computer science that’s been worked on since the 1950s at least.
The issue is to laypeople the term AI presents the idea of actual intelligence at a human level. Which computer science doesn’t require for something to qualify as AI
Leads to lay people attributing more ability to the llm than they actually posses.
Agreed. That said, I am uncomfortable with the idea that policing language is the correct or only solution to the problem.
Do the LLMs of the current craze meet that specific definition?
Yes.
Fair enough then.
Who decides what “AI” is and how did we come to decide what fits that title?
Language is ever-evolving, but a good starting point would be McCarthy et al., who wrote a proposal back in the 50s. See http://www-formal.stanford.edu/jmc/history/dartmouth/dartmouth.html
Techniques have come into and gone out of fashion, and obviously technology has improved, but the principles have not fundamentally changed.
Maybe machine learning models technically fit the definition of “algorithm” but it suits them very poorly. An algorithm is traditionally a set of instructions written by someone, with connotations of being high level, fully understood conceptually, akin to a mathematical formula.
A machine learning model is a soup of numbers that maybe does something approximately like what the people training it wanted it to do, using arbitrary logic nobody can expect to follow. “Algorithm” is not a great word to describe that.
Please let’s not defame Djikstra and other Algorithms like this. Just call them “corporate crap”, like what they are.
This thinking just feels like moving in the wrong direction. As an elementary teacher, I know that by next year all my assessments need to be practical or interview based. LLMs are here to stay and the quicker we learn to work with them the better off students will be.
And forget about having any sort of integrity or explaining to kids why it’s important for them to know how to do shit themselves instead of being wholly dependent on corporate proprietary software whose accessibility can and will be manipulated to serve the ruling class on a whim 🤦
It’s insane talking to people that don’t do math.
You ask them any mundane question and they just shrug, and if you press them they pull out their phone to check.
It’s important that we do math so that we develop a sense of numeracy. By the same token it’s important that we write because it teaches us to organize our thoughts and communicate.
These tools will destroy the quality of education for the students that need it the most if we don’t figure out how to reign in their use.
If you want to plug your quarterly data into GPT to generate a projection report I couldn’t care less. But for your 8th grade paper on black holes, write it your damn self.
Putting quarterly data into ChatGPT is dangerous for companies because that information is being fed into the AI and accessible by its creators, which means you’re just giving away proprietary information and trade secrets by doing that. But do these chucklefucks give one single shit? No. Because they’re selfish, lazy assholes that want robots to do their thinking for them.
as someone with wrist and hand problems that make writing a lot by hand, I’m so lucky i finished college in 2019
Chat GPT - answer this question, add 4 consistent typos. Then hand transcribe it.
Wouldn’t it make more sense to find ways on how to utilize the tool of AI and set up criteria that would incorporate the use of it?
There could still be classes / lectures that cover the more classical methods, but I remember being told “you won’t have a calculator in your pocket”.
My point use, they should prepping students for the skills to succeed with the tools they will have available and then give them the education to cover the gaps that AI can’t solve. For example, you basically need to review what the AI outputs for accuracy. So maybe a focus on reviewing output and better prompting techniques? Training on how to spot inaccuracies? Spotting possible bias in the system which is skewed by training data?
That’s just what we tell kids so they’ll learn to do basic math on their own. Otherwise you’ll end up with people who can’t even do 13+24 without having to use a calculator.
people who can’t even do 13+24 without having to use a calculator
More importantly, you end up with people who don’t recognize that 13+24=87 is incorrect. Math->calculator is not about knowing the math, per se, but knowing enough to recognize when it’s wrong.
I don’t envy professors/teachers who are hacing to figure out novel ways of determining the level of mastery of a class of 30, 40, or 100 students in the era of online assistance. Because, really, we still need people who can turn out top level, accurate, well researched documentation. If we lose them, who will we train the next gen LLM on? ;-)
end up with people who don’t recognize that 13+24=87 is incorrect
I had a telecom teacher who would either allow you to use a calculator, but you had to get everything right.
Or go without and you could get away with rougher estimates.Doing stuff like decibels by hand isn’t too bad if you can get away with a ballpark and it’s a much more useful skill to develop than just punching numbers in a calculator.
When will people need to do basic algebra in their head? The difficulty between 13+24 and 169+ 742 rises dramatically. Yeah it makes your life convenient if you can add simple numbers, but is it necessary when everyone has a calculator?
Like someone said. It’s not just about knowing what something is, but having the ability to recognize what something isn’t.
The ability to look at a result and be skeptical if it doesn’t look reasonable.
169+742. Just by looking I can tell it has to be pretty close to 900 because 160+740 is 900. That gives me a good estimate to go by. So when I arrive at 911. I can look at it and say. Yeah. That’s probably correct, it looks reasonable.
Yes.
How so?
That sounds like ot could be a focused lesson. Why try to skirt around what the desired goal is?
That also could be placed into detecting if something is wrong with AI too. Teach people things to just help spot these errors.
In my experience, it’s so much more effective to learn how to find the answers and spot the issues than to memorize how to do everything. There’s too much now to know it all yourself.
37? In a row?
Training how to use “AI” (LLMs demonstrably possess zero actual reasoning ability) feels like it should be a seperate pursuit from (or subset of) general education to me. In order to effectively use “AI”, you need to be able to evaluate its output and reason for yourself whether it makes any sense or simply bears a statitstical resemblance to human language. Doing that requires solid critical reasoning skills, which you can only develop by engaging personally with countless unique problems over the course of years and working them out for yourself. Even prior to the rise of ChatGPT and its ilk, there was emerging research showing diminishing reasoning skills in children.
Without some means of forcing students to engage cognitively, there’s little point in education. Pen and paper seems like a pretty cheap way to get that done.
I’m all for tech and using the tools available, but without a solid educational foundation (formal or not), I fear we end up a society snakeoil users in search of the blinker fluid.
There are some universities looking at AI from this perspective, finding ways to teach proper usage of AI. Then building testing methods around the knowledge of students using it.
Your point on checking for accuracy is on point. AI doesn’t always puke out good information, and ensuring students don’t just blindly believe it NEEDS to be taught. Otherwise wise you end up being these guys… https://apnews.com/article/artificial-intelligence-chatgpt-courts-e15023d7e6fdf4f099aa122437dbb59b
It’s like the calculator in the 80s and 90s. Teacher would constantly tell us “no jobs just gonna let you use a calulator, they’re paying you to work”…
I graduated, and really thought companies were gonna make me do stuff by hand, cause calulators made it easy. Lol.
Why would any job requrie employees to do math by hand, without a calculator?
