What is a closed-book, take-home test? Do students affirm they will not open books? How does on choose not to take the final? Do you mean these students dropped the course? Can one drop a course on final exam day at Brown?
That was my immediate thought as well. What the hell does “take home exam” mean that doesn’t include “yea, go ahead and use the books, internet, friends and family, whatever, because I have no way of knowing what you used.”
That guy teaches at Brown and didn’t realize that is exactly what people will do?
So you’re saying there is no honor code at Brown University?
Or that even if there is no one should take either seriously, and no one who does not take it seriously could face consequences?
If I were to speculate about the professor’s logic, it would probably be roughly “yes, sadly, I except a non-zero number of students to cheat. That they cheat on this test and effectively harm the prospects of the others in the class is regrettable, but it is what it is. But I would never have expected 50 of them - 50! - to do so.”
Of course, it’s equally possible that he *did* expect that many to cheat and wanted to be able to identify and catch such cheaters and have them disciplined.
As I interpret the story, closed book was not important because the answers were not in the book. Either he did not realize that an AI could solve the problems or he expected that the students would not try because that would be cheating.
My guess is that Brown claims its students can be trusted not to cheat and tries to ignore evidence that they can't be. It sounds as though the university tried to ignore what had happened until he pushed and he still does not expect them to do much in the future about the problem.
But “closed book” unquestionably means you are not allowed to use an AI to generate answers. Exactly as it means you are not allowed to use search engines to help you generate answers.
Holding Serrano’s supposed questionable judgement or naïveté as any meaningful part of the problem here, as Dr. Hammer’s closing question implies, is what I was objecting to.
The objection would be fair imo if in fact it turned out there were only a moderate number of cheaters.
Perhaps you are unfamiliar with college students, but "honor code" does not really bite with them. If you give them work to take home you can fully assume they are going to work together, collaborate, look online for answers or how to solve issues, etc. For a longer essay type exam that will be somewhat harder, but assuming they are just going to sit down at their desk and conscientiously work all by themselves without cracking a book or doing a web search is stupid. The students should not do so, of course, as it is against the rules. However, the rules cannot be enforced at all, and everyone involved knows this. As such, there is no reason to believe the students will follow the rules, as they have a lot to gain and nothing apparently to lose.
Even at graduate level, the things you're taught in school are basics. You're being tested on solving problems that have already been solved. That isn't what employers want to pay for: they want to pay for applying that understanding toward solving problems that haven't been solved yet. You may be using an AI for those problems too, but it won't be one-shotting them, because if it could then there would be no reason to hire a human. Without an understanding of the academic basics, an employee will have no ability to prompt the AI appropriately or to evaluate its output.
There's a similar answer to your arithmetic analogy. Using a calculator for arithmetic is more efficient than doing it by hand, but if you can't do it by hand at all, then you don't understand arithmetic. And if you don't understand arithmetic, you won't be able to understand more advanced math.
If they don’t understand arithmetic, they are very unlikely to get the rest of the solution anyway. So unnecessary to test their understanding of arithmetic.
In the last paragraph I was speaking of elementary school education where the arithmetic is what's being taught and there is no "rest of the solution" beyond that. At this phase, ability to perform hand calculation should be tested, and calculators prohibited during testing. Once learners pass these tests and advance to higher math, calculators can then be permitted.
A long time ago, before the days of GUIs, I was tutoring a young woman in accounting. So I wanting to illustrate how to do entries for a particular transaction, and I proposed a base amount of a million dollars and a secondary amount of 1% of that (it might have been a discount). She said, "a thousand dollars?" So I asked her to refigure it, and she came up with a hundred thousand dollars. At that point, I suggested that she take out her calculator. On her calculator, she came up with $10,000, the actual amount. But if she had pushed the wrong button and come up with the wrong answer, I don't think she would have had any way of knowing it was wrong.
Learning to do arithmetic by hand will help you become familiar with what answers you ought to get, and recognize when one just doesn't look right. If you've always had technological help you may never acquire that sense.
Fewer people have a need to know poems, but I've memorized a few dozen, all but a few in English, and I can quote them, or spot when someone misquotes them. And it helps me think about what they mean. So there is some point to Socrates's concern about the effect of writing on memory.
I don't find it so myself; the poems I know mostly stimulate my thoughts or my emotions. But it has other uses. The other day I was playing in a friend's roleplaying game, and I quoted some lines from "Horatius" to describe our tactical situation. . . .
I would have said middle school, or even elementary school. Figuring percentages wasn't high school math in my day. And the time to learn arithmetic is ideally younger than high school age.
One of the skills that is being outsourced to LLMs seems to be copy editing, including developmental editing. I think that does need to be learned in high school. I expect that routinely trusting an LLM to do it makes it much less likely that people will learn to edit themselves, which is an important part of thinking through an idea.
"looks like evidence that being able to do arithmetic is no longer a useful skill hence no longer something worth testing."
Being able to do arithmetic is almost certainly less important than it was in the past, but I think it's still useful. Pulling out a calculator ruins your flow, like any task switching.
It's also nice to be able to spot errors at a glance, using a calculator is far more tedious.
AI likely already can spot many such errors now, and will likely spot even more of them in the future. AIs are already debugging software and fixing code.
What you're saying is true, and yet there are published papers with simple arithmetic errors.
As a software developer, I deal with annoying to find human errors all the time, you can have systems to mitigate it, but until AIs are much better than humans, we're going to be stuck with simple errors.
As a side note, I recently had the most expensive claude model mistakenly divide a yearly interest by 12 to convert it to a monthly interest rate. So it seems ai (for the moment) is prone to the same kind of human error we are.
“As a side note, I recently had the most expensive claude model mistakenly divide a yearly interest by 12 to convert it to a monthly interest rate.”
Perhaps the model “grew up” during the years after the GFC when interest rates were so low that dividing by 12 is a very accurate approximation… 😏
But I was not actually trying to make any strong claim about how good AI is on this subject right now. Only that it is rapidly improving, and that on narrow points like this it actually is likely to get so much better that it will be better than most humans at it.
I have always taken that there are two 'tests' other than projects.
The first are progressive tests for a teacher or lecturer to determine whether, as you pointed out, there are areas of the subject that the students have not adequately grasped. These are non-grading tests. These are mainly for the guidance of the teacher/lecturer.
The second are grading exams which are monitored to generally prevent cheating.
If a College/University is providing a degree (in my case Civil Engineering), there needs to be certainty that I achieved my degree honestly otherwise the degree is worthless to a future Employer if it became known that most students used AI or some other method of cheating instead of acquiring the knowledge by their own efforts.
What is the purpose anyway, you might ask, of a degree? It is the knowledge by a prospective Employer that the person applying for employment has been trained to think for themselves (in the case of Engineers in a logical manner).
Many Employers (for a variety of employment) only want to know that the graduate is capable of a certain quality of thinking shown by having (honestly) graduated from University with some sort of Degree.
When an Engineer graduates his real training begins. A Graduate Engineer has only shown that he/she knows where to find the information needed for the future work to be undertaken.
A Student who obtains a Degree by cheating with or without the help of AI is fooling him/herself and their future Employer. They are in effect a 'dumb' Graduate.
The danger of relying only on technology is they have no idea of the accuracy of the result. As the old saying goes, "garbage in, garbage out". I am certain that I wouldn't want to use a bridge or a building that was designed where the Engineer had no idea from experience that the beams etc provided by technology were the right size.
Many years ago, in my last employment as a Supervising Project Engineer in a large State Water Enterprise, the Enterprise no longer employed its own Design Engineers to do the actual design work, having outsourced its project designs to Consultants, where in most cases the Consultants had never built what they were designing. All plans, of course, were certified by the Consultants and accepted by the 'Design' Section without checking in-house. The Enterprise had, in effect, become a 'dumb' client.
Based on my 30 years of experience in Operations and Maintenance as well as Construction, I could not get the 'Design' Section of the Enterprise to understand what we had become. I was told to build the design and stay out of design work. Those designs cost substantial amounts for unnecessary work on sections of work where I did not have authority to change the design on site. While my complaints were based on work experience, the principle applies to learning to think about what you are designing/building and not relying only on technology such as AI.
Last spring 2026 semester, I gave 55 one-on-one, 30-minutes each, oral final exams via Zoom for two asynchronous online classes. I did so because it is no longer possible to give students summary course grades any other way for a totally online course, due to extensive availability and use of LLM Ai. Clearly, this solution to the problem is not scalable.
I also had ironclad evidence from Canvas quiz logs that more than 50 percent of the students used Ai to answer midterm exam questions. Rare is the student who does not now use LLM Ai to write any kind of essay that is not scrupulously proctored, which cannot be done reasonably outside a face-to-face venue.
DF, you are spot on correct in your observations. It seems to me that LLM Ai has forced the issue for online programming. Sadly enough, colleges and universities throughout the world have yet to admit that asynchronous online programs are a dead man walking. As I write, the university where I teach has yet to come up with a policy to address this issue.
Any graded activity must be done in a fully proctored venue; if it is not, the grade cannot be authentic. And yet, this upcoming fall 2026 semester has me teaching two asynchronous online courses. What am I to do? I'm all ears for anyone who has a reasonable solution.
I'm also a university professor. There is a fourth use for assessments/exams. That is, to actually get students to learn the material. Many/Most won't without a high-stakes test. AI on take-home assessments breaks that too. It is now possible on many assignments to take the assignment pdf or word doc, upload it to AI and say, "solve these problems" and then submit it. No learning at all and the student doesn't even see either the input or output document so nothing passes through their brain at all. I'm really interest in any thoughts besides "only in-person proctored exams". Projects that can be completed over days or weeks have real value in this fourth use.
Cellphones are not permitted in my exam room. We do not physically check students to verify that they are not in possession of a concealed cellphone. Our campus would not allow it. One of my fourth year mathematics students approached me after a midterm with 200 registered students and informed me that several classmates had sneaked out their cellphones hidden in their clothing during the exam photographed the midterm within Claude AI, then hid their cellphone between their legs and wrote down the correct answers prior to submitting the exam. The student demonstrated the cheat in my presence, opened up Claude AI, pressed the camera button, and then in less than one minute a midterm that would take one hour to manually complete was completely solved. The AI generated solution was perfect. The cheat required only required the students to press one camera shutter button. Claude AI automatically solved the midterm with perfect precision.
Presumably those grading exams going forward will know to do that very thing using the most popular LLMs and have that as background when grading each student’s exam.
And a fifth use of exams: to provide information to the instructors and the school generally on the effectiveness of the classes in conveying to students the information they purport to convey.
"If the purpose of the test is to generate information for potential future employers, why should they want the student tested without a tool that, if they hire him, he will have?"
AI isn't a tool, it's an agent. Using an AI to complete a task isn't like using a tool to help complete a task, it's like having a different person to complete the task for you. Why would I want to hire an employee who will tell an AI to do all their work, when I could just tell an AI to do the work myself and cut out the middleman?
Unless the AI operates independently once given a task, it is a tool, not an agent.
If I have a servant and tell him to cook dinner, I expect him to figure out that means get the ingredients beforehand and wash the dishes afterwards, all without any interaction with me. That's an agent. If I have to detail all those steps, he's a tool.
I think asking an AI to generate an essay for you is more like instructing a servant. A lot of students don't even bother reading the essays it generates.
Distinguishing among three uses for exams, or three audiences for the results of exams, is a very productive way of looking at things. With 20/20 hindsight, I was most interested in having students know what they knew and didn't know, not that that was conscious on my part. Letting the administration know who should fail and such was very much an edge question, though it did occasionally exist.
You identify the problem that because of lingual familiarity students tend to think they know more than they do. My students were wise asses, thinking they knew it all, so that counts double.
One place I taught at for many years was big on oral exams. I liked that because no cheating and I could be creative. Inter alia, I taught International Trade. One thing the people had to know was the results of the four theorems of the Heckscher-Ohlin Model, one of which is the Rybczynski theorem. For a weak, but overall deserving student, I would ask: How do you spell Rybczynski? Can't do that on a written exam!
I wonder sometimes if stories like these are somewhat downstream from the problem of people going to college who don't really want to be there and don't have a good reason to be there (other than someone paid for them and/or they were acculturated to believe they should go). Perhaps also of college programs that are not designed with clear goals or end states in mind.
Students who want to be in school, who worked hard to be there, who want to learn what is being taught, who have a strong sense of how it will help them in their careers, are unlikely to cheat even given the opportunity.
> Serrano could have asked students whose midterms were suspiciously good to explain some of their answers and failed any obviously unable to do so.
There’s a difference between showing an answer is correct and generating the correct answer.
For economics itself, it’s tough to be better than AI. I’m not sure academia is up to the task, but producing interesting results using AI is now more important than learning models so that’s what needs to be taught.
What is a closed-book, take-home test? Do students affirm they will not open books? How does on choose not to take the final? Do you mean these students dropped the course? Can one drop a course on final exam day at Brown?
That was my immediate thought as well. What the hell does “take home exam” mean that doesn’t include “yea, go ahead and use the books, internet, friends and family, whatever, because I have no way of knowing what you used.”
That guy teaches at Brown and didn’t realize that is exactly what people will do?
So you’re saying there is no honor code at Brown University?
Or that even if there is no one should take either seriously, and no one who does not take it seriously could face consequences?
If I were to speculate about the professor’s logic, it would probably be roughly “yes, sadly, I except a non-zero number of students to cheat. That they cheat on this test and effectively harm the prospects of the others in the class is regrettable, but it is what it is. But I would never have expected 50 of them - 50! - to do so.”
Of course, it’s equally possible that he *did* expect that many to cheat and wanted to be able to identify and catch such cheaters and have them disciplined.
As I interpret the story, closed book was not important because the answers were not in the book. Either he did not realize that an AI could solve the problems or he expected that the students would not try because that would be cheating.
My guess is that Brown claims its students can be trusted not to cheat and tries to ignore evidence that they can't be. It sounds as though the university tried to ignore what had happened until he pushed and he still does not expect them to do much in the future about the problem.
You and I are very much in agreement.
But “closed book” unquestionably means you are not allowed to use an AI to generate answers. Exactly as it means you are not allowed to use search engines to help you generate answers.
Holding Serrano’s supposed questionable judgement or naïveté as any meaningful part of the problem here, as Dr. Hammer’s closing question implies, is what I was objecting to.
The objection would be fair imo if in fact it turned out there were only a moderate number of cheaters.
Perhaps you are unfamiliar with college students, but "honor code" does not really bite with them. If you give them work to take home you can fully assume they are going to work together, collaborate, look online for answers or how to solve issues, etc. For a longer essay type exam that will be somewhat harder, but assuming they are just going to sit down at their desk and conscientiously work all by themselves without cracking a book or doing a web search is stupid. The students should not do so, of course, as it is against the rules. However, the rules cannot be enforced at all, and everyone involved knows this. As such, there is no reason to believe the students will follow the rules, as they have a lot to gain and nothing apparently to lose.
Even at graduate level, the things you're taught in school are basics. You're being tested on solving problems that have already been solved. That isn't what employers want to pay for: they want to pay for applying that understanding toward solving problems that haven't been solved yet. You may be using an AI for those problems too, but it won't be one-shotting them, because if it could then there would be no reason to hire a human. Without an understanding of the academic basics, an employee will have no ability to prompt the AI appropriately or to evaluate its output.
There's a similar answer to your arithmetic analogy. Using a calculator for arithmetic is more efficient than doing it by hand, but if you can't do it by hand at all, then you don't understand arithmetic. And if you don't understand arithmetic, you won't be able to understand more advanced math.
Agreed with all but your final logic.
If they don’t understand arithmetic, they are very unlikely to get the rest of the solution anyway. So unnecessary to test their understanding of arithmetic.
In the last paragraph I was speaking of elementary school education where the arithmetic is what's being taught and there is no "rest of the solution" beyond that. At this phase, ability to perform hand calculation should be tested, and calculators prohibited during testing. Once learners pass these tests and advance to higher math, calculators can then be permitted.
A long time ago, before the days of GUIs, I was tutoring a young woman in accounting. So I wanting to illustrate how to do entries for a particular transaction, and I proposed a base amount of a million dollars and a secondary amount of 1% of that (it might have been a discount). She said, "a thousand dollars?" So I asked her to refigure it, and she came up with a hundred thousand dollars. At that point, I suggested that she take out her calculator. On her calculator, she came up with $10,000, the actual amount. But if she had pushed the wrong button and come up with the wrong answer, I don't think she would have had any way of knowing it was wrong.
Learning to do arithmetic by hand will help you become familiar with what answers you ought to get, and recognize when one just doesn't look right. If you've always had technological help you may never acquire that sense.
Fewer people have a need to know poems, but I've memorized a few dozen, all but a few in English, and I can quote them, or spot when someone misquotes them. And it helps me think about what they mean. So there is some point to Socrates's concern about the effect of writing on memory.
I know a lot of poetry. Aside from my bardic circle at Pennsic it is useful for falling asleep.
I don't find it so myself; the poems I know mostly stimulate my thoughts or my emotions. But it has other uses. The other day I was playing in a friend's roleplaying game, and I quoted some lines from "Horatius" to describe our tactical situation. . . .
I found my slide rule experience was a tremendous help in developing that "doesn't pass the smell" test, since slide rules have no decimal points.
I wholeheartedly agree.
But that really argues for not allowing calculators in high school math classes, rather than in college economics courses.
I would have said middle school, or even elementary school. Figuring percentages wasn't high school math in my day. And the time to learn arithmetic is ideally younger than high school age.
One of the skills that is being outsourced to LLMs seems to be copy editing, including developmental editing. I think that does need to be learned in high school. I expect that routinely trusting an LLM to do it makes it much less likely that people will learn to edit themselves, which is an important part of thinking through an idea.
"looks like evidence that being able to do arithmetic is no longer a useful skill hence no longer something worth testing."
Being able to do arithmetic is almost certainly less important than it was in the past, but I think it's still useful. Pulling out a calculator ruins your flow, like any task switching.
It's also nice to be able to spot errors at a glance, using a calculator is far more tedious.
Spreadsheets properly designed aid in that now.
AI likely already can spot many such errors now, and will likely spot even more of them in the future. AIs are already debugging software and fixing code.
What you're saying is true, and yet there are published papers with simple arithmetic errors.
As a software developer, I deal with annoying to find human errors all the time, you can have systems to mitigate it, but until AIs are much better than humans, we're going to be stuck with simple errors.
As a side note, I recently had the most expensive claude model mistakenly divide a yearly interest by 12 to convert it to a monthly interest rate. So it seems ai (for the moment) is prone to the same kind of human error we are.
“As a side note, I recently had the most expensive claude model mistakenly divide a yearly interest by 12 to convert it to a monthly interest rate.”
Perhaps the model “grew up” during the years after the GFC when interest rates were so low that dividing by 12 is a very accurate approximation… 😏
But I was not actually trying to make any strong claim about how good AI is on this subject right now. Only that it is rapidly improving, and that on narrow points like this it actually is likely to get so much better that it will be better than most humans at it.
David
I have always taken that there are two 'tests' other than projects.
The first are progressive tests for a teacher or lecturer to determine whether, as you pointed out, there are areas of the subject that the students have not adequately grasped. These are non-grading tests. These are mainly for the guidance of the teacher/lecturer.
The second are grading exams which are monitored to generally prevent cheating.
If a College/University is providing a degree (in my case Civil Engineering), there needs to be certainty that I achieved my degree honestly otherwise the degree is worthless to a future Employer if it became known that most students used AI or some other method of cheating instead of acquiring the knowledge by their own efforts.
What is the purpose anyway, you might ask, of a degree? It is the knowledge by a prospective Employer that the person applying for employment has been trained to think for themselves (in the case of Engineers in a logical manner).
Many Employers (for a variety of employment) only want to know that the graduate is capable of a certain quality of thinking shown by having (honestly) graduated from University with some sort of Degree.
When an Engineer graduates his real training begins. A Graduate Engineer has only shown that he/she knows where to find the information needed for the future work to be undertaken.
A Student who obtains a Degree by cheating with or without the help of AI is fooling him/herself and their future Employer. They are in effect a 'dumb' Graduate.
The danger of relying only on technology is they have no idea of the accuracy of the result. As the old saying goes, "garbage in, garbage out". I am certain that I wouldn't want to use a bridge or a building that was designed where the Engineer had no idea from experience that the beams etc provided by technology were the right size.
Many years ago, in my last employment as a Supervising Project Engineer in a large State Water Enterprise, the Enterprise no longer employed its own Design Engineers to do the actual design work, having outsourced its project designs to Consultants, where in most cases the Consultants had never built what they were designing. All plans, of course, were certified by the Consultants and accepted by the 'Design' Section without checking in-house. The Enterprise had, in effect, become a 'dumb' client.
Based on my 30 years of experience in Operations and Maintenance as well as Construction, I could not get the 'Design' Section of the Enterprise to understand what we had become. I was told to build the design and stay out of design work. Those designs cost substantial amounts for unnecessary work on sections of work where I did not have authority to change the design on site. While my complaints were based on work experience, the principle applies to learning to think about what you are designing/building and not relying only on technology such as AI.
Last spring 2026 semester, I gave 55 one-on-one, 30-minutes each, oral final exams via Zoom for two asynchronous online classes. I did so because it is no longer possible to give students summary course grades any other way for a totally online course, due to extensive availability and use of LLM Ai. Clearly, this solution to the problem is not scalable.
I also had ironclad evidence from Canvas quiz logs that more than 50 percent of the students used Ai to answer midterm exam questions. Rare is the student who does not now use LLM Ai to write any kind of essay that is not scrupulously proctored, which cannot be done reasonably outside a face-to-face venue.
DF, you are spot on correct in your observations. It seems to me that LLM Ai has forced the issue for online programming. Sadly enough, colleges and universities throughout the world have yet to admit that asynchronous online programs are a dead man walking. As I write, the university where I teach has yet to come up with a policy to address this issue.
Any graded activity must be done in a fully proctored venue; if it is not, the grade cannot be authentic. And yet, this upcoming fall 2026 semester has me teaching two asynchronous online courses. What am I to do? I'm all ears for anyone who has a reasonable solution.
I'm also a university professor. There is a fourth use for assessments/exams. That is, to actually get students to learn the material. Many/Most won't without a high-stakes test. AI on take-home assessments breaks that too. It is now possible on many assignments to take the assignment pdf or word doc, upload it to AI and say, "solve these problems" and then submit it. No learning at all and the student doesn't even see either the input or output document so nothing passes through their brain at all. I'm really interest in any thoughts besides "only in-person proctored exams". Projects that can be completed over days or weeks have real value in this fourth use.
Cellphones are not permitted in my exam room. We do not physically check students to verify that they are not in possession of a concealed cellphone. Our campus would not allow it. One of my fourth year mathematics students approached me after a midterm with 200 registered students and informed me that several classmates had sneaked out their cellphones hidden in their clothing during the exam photographed the midterm within Claude AI, then hid their cellphone between their legs and wrote down the correct answers prior to submitting the exam. The student demonstrated the cheat in my presence, opened up Claude AI, pressed the camera button, and then in less than one minute a midterm that would take one hour to manually complete was completely solved. The AI generated solution was perfect. The cheat required only required the students to press one camera shutter button. Claude AI automatically solved the midterm with perfect precision.
Wow. Thanks for that story.
Presumably those grading exams going forward will know to do that very thing using the most popular LLMs and have that as background when grading each student’s exam.
And a fifth use of exams: to provide information to the instructors and the school generally on the effectiveness of the classes in conveying to students the information they purport to convey.
"If the purpose of the test is to generate information for potential future employers, why should they want the student tested without a tool that, if they hire him, he will have?"
AI isn't a tool, it's an agent. Using an AI to complete a task isn't like using a tool to help complete a task, it's like having a different person to complete the task for you. Why would I want to hire an employee who will tell an AI to do all their work, when I could just tell an AI to do the work myself and cut out the middleman?
Unless the AI operates independently once given a task, it is a tool, not an agent.
If I have a servant and tell him to cook dinner, I expect him to figure out that means get the ingredients beforehand and wash the dishes afterwards, all without any interaction with me. That's an agent. If I have to detail all those steps, he's a tool.
I think asking an AI to generate an essay for you is more like instructing a servant. A lot of students don't even bother reading the essays it generates.
The thing is, you are both correct.
Good point
Distinguishing among three uses for exams, or three audiences for the results of exams, is a very productive way of looking at things. With 20/20 hindsight, I was most interested in having students know what they knew and didn't know, not that that was conscious on my part. Letting the administration know who should fail and such was very much an edge question, though it did occasionally exist.
You identify the problem that because of lingual familiarity students tend to think they know more than they do. My students were wise asses, thinking they knew it all, so that counts double.
One place I taught at for many years was big on oral exams. I liked that because no cheating and I could be creative. Inter alia, I taught International Trade. One thing the people had to know was the results of the four theorems of the Heckscher-Ohlin Model, one of which is the Rybczynski theorem. For a weak, but overall deserving student, I would ask: How do you spell Rybczynski? Can't do that on a written exam!
I wonder sometimes if stories like these are somewhat downstream from the problem of people going to college who don't really want to be there and don't have a good reason to be there (other than someone paid for them and/or they were acculturated to believe they should go). Perhaps also of college programs that are not designed with clear goals or end states in mind.
Students who want to be in school, who worked hard to be there, who want to learn what is being taught, who have a strong sense of how it will help them in their careers, are unlikely to cheat even given the opportunity.
> Serrano could have asked students whose midterms were suspiciously good to explain some of their answers and failed any obviously unable to do so.
There’s a difference between showing an answer is correct and generating the correct answer.
For economics itself, it’s tough to be better than AI. I’m not sure academia is up to the task, but producing interesting results using AI is now more important than learning models so that’s what needs to be taught.
Paraphrasing T.S. Eliot: Yes, and it's the models that AI knows.
Who will make us new models?
Depends on what you mean by new. AI can currently collect and produce novel results, it's up to the human to curate and popularize them
> Testing used for grading should either be in-person and monitored or with cheating deterred by serious efforts to detect and punish it.
All exams where used for grading should be in person, to prevent cheating.