Cheaters prospered before AI made it easy
The academy was broken before ChatGPT made essay assignments look pointless
For most of the past academic year, educators and academics have been grappling with how to handle the rise of powerful and effective large language models (LLMs), a form of artificial intelligence that can write essays better than the typical college sophomore. Methods for detecting AI-generated work are unreliable and are also generally likely to be obsolete after the next few development cycles.1
The problem with focusing on new AI tools like ChatGPT: Cheating was already a growing and largely unaddressed problem within the academy.
ChatGPT and other LLMs didn’t create a new problem; it disrupted a large and rapidly-growing industry in academic dishonesty that had paying clients at every level from Algebra I to post-doctoral tenure review. Cheating was already difficult to detect, and had already become widespread.
The naked emperor of education
Over the course of the past three or four decades or so, education and academia have systematically changed in multiple ways that have made it easier for cheaters to prosper. The emperor of education had already donned a set of non-existent new clothes. Proctored and structured assessment has been de-emphasized in favor of fuzzier measures. Take-home assignments have ballooned in volume and in importance. Grade inflation has steadily eroded standards at every step of the way. Students can and do earn degrees without mastering the material they’re supposed to; right now, most students cheat.2
I’ve had personal experience with many cheating students and even one cheating would-be-faculty-member, but the story that sticks with me the most comes from an older friend I got to know while I was going through graduate school. She had set up an interview between a freshly-minted Ph.D. and a prospective employer. It seemed like a perfect match, and the employer was very excited about the research the candidate had been doing for their dissertation.
The interviewer printed out a copy of the candidate’s dissertation in order to ask him questions about it. Unfortunately, the candidate hadn’t written the dissertation. In fact, he couldn’t read his own dissertation. His proficiency with written English was limited. He had successfully gone through an entire doctoral program at a top-flight American university paying someone else to write assignments and papers for him.3
The economics of cheating
The average cost of a college course at a four-year institution exceeds $1500 in tuition and fees alone, typically also involving about 50 hours of instruction and as much as 100 hours of out-of-class work. Accounting for time and living expenses, avoiding having to repeat a single college class can be easily worth $5,000 at a typical university. Successfully completing a degree can have a value of hundreds of thousands of dollars, possibly even over a million dollars.
Paying a better student to do take-home assignments has always been a viable option for some students who want to increase their grade. In a global internet-based service market, essay writing services are a click away. In Kenya alone, essay-writing services for students in the UK and US had combined revenues estimated at over a hundred million dollars a year in 2019. The essay-writing segment of the cheating industry was a multi-billion dollar industry before ChatGPT showed up.
There is no substitute for proctored assessment
Proctored in-person assessment has been systematically de-emphasized for a variety of reasons.4 The fact of the matter is that students can get outside help on any unsupervised assignment. Some primary and secondary school teachers expect that parents will help their children complete assignments correctly. The unfortunate fact is that many students (and parents) care much more about grades than learning outcomes, and are thus willing to cheat if the reward is greater than the risk.
Many wealthy parents are willing to pay significant amounts of money to ensure that their students get straight A grades. This is easy to do with take-home assignments of every kind. It is much more difficult with proctored assessments; desperate parents who pay for a ringer to sit and take a standardized test on behalf of their student run a high risk of getting caught.5 On the other side of the coin, university students who agree to take an in-person test for another student run a risk of serious disciplinary consequences if caught. Writing an essay for another student is a low risk activity.
De-emphasizing or removing proctored assessments means that cheating on take-home assignments has a larger impact on grades, which means cheating has a higher reward … and I expect this trend to accelerate in response to the recent Supreme Court decision banning affirmative action.
The detection problem
The biggest risk associated with cheating is that if caught, punishments can be severe, up to and including expulsion. One of the major reasons why cheating has been on the rise is that detecting and punishing cheating has become more difficult. Removing proctored assessments makes it harder to detect cheating because there is no independent basis for comparison; in courses with a mix of proctored and take-home assignments, high-quality assignments from low-performing students can be flagged for additional scrutiny.
Another major factor in higher education is adjunctification. Adjuncts are instructors whose job security usually depends on getting positive student evaluations. Detecting and reporting cheaters almost inevitably leads to negative student evaluations and complaints sent directly to multiple administrators.6 In fact, students tend to respond negatively to anything an instructor does to make cheating more difficult. Only tenured professors can afford to report cheaters.7
The nonlinearity of cheating
Social dynamics can (and do) lead to cheating en masse. If cheating becomes ten times as easy, it can become a hundred times as common.8 This has to do with social contagian, network effects, and the fact that an existing business will “push” its services to maintain or expand revenue. The fact that cheating leads to loss of learning also helps create habitual cheaters.
Students who cheat successfully will recommend this course of action to other students. Students who make money writing essays for a friend are likely to expand their business once it proves successful - and professional services will advertise openly on the internet. Student group chats are a major transmission vector of cheating, with students sharing solutions and resources (such as essay-writing services) with one another.
Another major component of the non-linearity of cheating happens on the individual level. Cheating also increases the degree to which students are not prepared for more advanced courses. A student who successfully cheats their way to an A is not well-prepared to master the material in the next course in a sequence, but is very likely to have the means, motive, and opportunity to cheat in the next course.
Where do we go from here?
The main reason that ChatGPT will not break the academy is because the academy was already broken. Rigorously honest assessment of students - proctored assessments such as standardized tests - have been systematically de-emphasized over a time period spanning multiple decades. For a variety of economic and technological reasons, cheating on take-home assignments has become far easier, and detecting cheating has become far harder.
Cheaters were already prospering on a large scale before the rise of large-language models like ChatGPT. College degrees had already become more of a marker of class and the willingness and ability to pay for tuition. ChatGPT has simply cast a stark light on an existing truth: Meaningful assessment of students requires proctored assessments. The alternative to high-stakes tests is rewarding dishonesty on a large scale.
For example, right now, fake citations are one of the most effective ways of detecting ChatGPT-generated work.
See this paper for an overview of a few surveys internationally.
It wasn’t clear whether or not he had done the research described in his dissertation. The candidate in question was Chinese, and it’s possible that all they paid for were translation services.
One of the simplest reasons is grade inflation. It’s a lot easier to inflate grades when they aren’t based on structured in-class assessments. Other reasons include fashionably trendy educational theories about multiple intelligences and learning styles as well as affirmative action, and sweeping attempts to generalize disability accommodations. I suspect there is also a role played by the inconvenient fact that most people going into the field of education are people who did poorly on tests; it is easy for researchers who “didn’t test well” to embrace the hypothesis that doing badly on proctored assessments isn’t meaningful.
In coverage of the 2019 scandal involving multiple celebrities paying large sums of money to try to get higher SAT scores for their children, one of the underemphasized facts is that every other aspect of the college admissions process is easier to cheat on. Most of those celebrity parents probably paid for tutors to correct their children’s homework before it was turned in for a grade, either paid money or traded favors to get glowing letters of recommendation written for their children, and probably had a professional write or rewrite application essays. Cheating on proctored exams is possible, but it is much more difficult and usually less effective.
I have seen this more than once in my teaching career, unfortunately.
Even then, preventing and punishing cheating usually negatively impacts the professor’s reputation with students and relationship with administrators.
The non-linearity is more on the order of a logistic curve - at the limiting high end, it’s not possible for more than 100% of students to cheat.