Much has happened at the prestigious Brown University (United States) since its professor Roberto Serrano reported in this newspaper in June a massive fraud with artificial intelligence (AI) in one of its exams, without the dean's office reacting. The news was replicated by the main media - even Nature magazine asked him for an article - Serrano has received "hundreds" of letters of support and the person who was in charge of the Academic Code Committee has been dismissed. But what moves the eminent Spanish economist most is that he believes his students have learned their lesson.
Since the news was made public, 53 of Serrano's students, one of the main promoters of the application of game theory, have confessed that they cheated with AI in the exam of an advanced mathematical economics course that, for exceptional reasons, allowed them to do from home. After canceling the grade for that test, Serrano decided to take into account only the grade for the final exam and asked that, if they were contacted by the committee studying the fraud, “they be honest.” Later, the 2024 King Juan Carlos Prize winner for Economics received a call from a student who, through tears, wanted to apologize for what she considered a “betrayal.” “I responded: ‘It is a vital lesson that you have learned: from now on you are going to defend honesty, integrity, decency,” says the professor by phone from Providence (Rhode Island).
Serrano's media complaint illustrates like few others the colossal challenge that the University faces: how to ensure that its students learn the basics and, if they don't, that they don't pass thanks to artificial intelligence (AI). Experts do not doubt that the generalization of language models endangers writing skills and critical thinking, while academic fraud and difficulty in evaluating real learning grow. But how to make these a complement to learning and not a shortcut to passing a subject?
At the moment, universities in Europe and the United States are inclined to avoid AI with other types of exams and essays, oral or practical, while in Asia it is integrated into the evaluation. The defense of these values and ethics tainted at Brown have become the great workhorse of the University. Princeton, for example, ended unproctored exams in May due to problems of violation of academic integrity, the pride of the institution for 133 years.
American universities have also detected a suspicious increase in the number of outstanding students, such as that identified by Serrano. And Harvard has chosen to cut corners: starting in the fall of 2027, no more than 20% of students will be able to obtain an A—the highest grade—although up to four optional grades can be added. Senén Barro, director of the Singular Center for Research in Intelligent Technologies (CITIUS) of the University of Santiago (USC), qualifies this news: “There is a reduction in the cost of high grades and it is being studied whether it responds to a relaxation of the tests, pressure from the students or because, in fact, they have more means to solve them.”
In Spain, official data does not quantify an increase in the average grade since ChatGPT emerged in 2022, but perhaps it is early because the latest figures are from the 2023/2024 academic year (7.43 compared to 7.45 the previous year). There has never been an honor code, like at Princeton, but measures are increasing to detect nano-ears, glasses or those used to copy with AI. In the last Selectivity, several communities purchased radio frequency detectors. Even the Polytechnic of Valencia, as its rector José Capilla tells this newspaper, is going to install arches at the entrance to the exam classrooms.
At the University of Valencia (UV) they are working on a regulation to apply AI and address cheating not only in exams, but in the essays that are part of the grade. The text states that AI detectors cannot be the only evidence for a student to be accused of academic fraud. “Because the use of AI, per se, is not bad, it depends on how it is regulated,” explains Joan Vila, rector's delegate for AI. “We are developing what we call an academic and scientific integrity framework to establish how AI is used in the learning process and how we are going to evaluate.” If students used AI as a “personalized tutor” to expand their knowledge, “it would be ideal, but we know that is not the case,” says Vila.
In its teaching guides, the UV is going to establish four levels, from not being able to use AI in final exams to evaluating the student's ability to use it. "In the final degree or master's degree projects we force them to sign a declaration. Students have to say what tools they have used, for what and, if they can, provide the prompts [the instructions given to the system]," lists Vila.
Barro, former rector of the USC, starts from a premise: "Tools specifically oriented toward education should be developed; because ordinary models are fantastic at developing responses, but they do not help learning." In Spain, according to a survey by the Knowledge and Development Foundation (Cyd) of 800 students, 61% of them stated that they have not received training from their center and only 41% believe that their institution promotes the use of AI.
Furthermore, only 24% of those surveyed consider that their teachers have already adapted their evaluation: 23% now undergo more practical tests and 18% oral exams. And only 54% assure that their teachers incorporate these tools in the classroom, which does not mean that they do not use them to prepare classes. In the same way as university students, who use AI to resolve doubts (68%), search for information (57%), summarize (54%) and review and correct texts (50%), according to FCYD.
Professors around the world who have already updated the evaluation intensify oral tests, many works now have a defense part - not just delivery -, they call for more face-to-face activities and put more emphasis on continuous evaluation. Practices that have always been recommended but "because they consume so much time, presentation and debate, they are limited," explains Barro. Although it is not impossible. In Italy, historically, most exams are oral at university, and in the UK dissertations are common. And there are even those who are trying to use a chatbot as an assistant.
A professor at New York University, Panos Ipeirotis, has decided to “fight fire with fire” and to do so he has created one so that his students from home can answer orally the questions that the computer program throws at them. Then, the teacher personally corrects each conversation.
In secondary school, AI is also pushing teachers to make changes that, to a certain extent, says Víctor Bermúdez, who teaches Philosophy at a public institute in Extremadura, can be positive. "The misuse of AI brings good: it forces evaluation tools to be more varied, individualized and demanding in terms of understanding what is learned. I, for example, continue to ask for research work, but now I prioritize the evaluation of teamwork, the oral defense of the project or transversal skills that can be measured very well in the classroom, such as the dialogic attitude or argumentative ability," he points out. They are more demanding formulas for the teacher, adds Bermúdez, and their use is limited by the ratios, as Barro denounces. “If you have many students, you cannot evaluate them as you should, and often the in-person exam is used as the only relatively fraud-free tool.”
In adapting to the change generated by the emergence of AI, VET teachers start with the advantage of the practical approach to studies, continues Maria Martorell, who teaches classes in the higher cycle of Early Childhood Education in a public institute in Mallorca. "If before you asked them to program [design] a game proposal to work with 4-year-old children, now you have to ask them for more concrete things. Not just a theoretical formulation, because they may have copied it from AI, but something more experiential. Like: now put it into practice, explain it, say where you got the information from. More like a defense in the classroom, individual or group, in which you ask them questions." In line with what Bermúdez proposes for ESO and Baccalaureate, Martorell admits, however, that with AI the final exam of a lifetime has regained weight; “something that we had left aside a little because daily activities were sometimes enough to evaluate.”
"The world is late and we can save specific cases, but not countries as a whole. We are all trying to see what we can do," Barro acknowledges. He praises the comprehensive teaching plans with AI that are being approved in universities like his own or Valencia, with training for teachers and a decalogue of what can and cannot be done. The Conference of Rectors of Spanish Universities (CRUE) aims to present a white paper in December.
AI as an ally
Meanwhile, in Southeast Asia, unlike the United States and Europe, expressly educational tools are being designed, they do not veto technology and use it a la carte, according to strengths and weaknesses. Just like Professor Ipeirotis does from New York. These tools have a Socratic profile: they challenge the student, ask them questions and force them to think. And the model, according to the answers, raises other questions that should help you learn. For example, they create bots (programs that imitate human behavior) to recreate a trial in which law students pretend to be lawyers and react to different situations.
IE University, with students from 160 countries between Madrid and Segovia, no longer asks candidates for a motivation letter because they distrust its authorship; but the entrance test is done with AI, although the choice is in the hands of a committee, humans. They follow the Western evaluation model, but with nods to Asia. For example, they let themselves rely on AI in an exam, and then refute it orally, emphasizing the positive and negative. “Now there is a lot of talk about critical thinking, but we also want them to have judgment,” says Martín Rodríguez, its general director of Executive Education & Digital Learning. "Above all, they know how to defend their ideas, their convictions. Before you sent an essay, and now, maybe, a written defense of your point of view on a strategic point and a panel of professors then asks questions."
Students cannot become guinea pigs, but universities inevitably have to continue experimenting with assessment in the age of AI and, in the meantime, trust their honesty. Like that of the 53 Serrano students who have learned the lesson.
