Implementing artificial intelligence (AI) into the education system is as much a task for teachers and professors as it is for the system. If the task you give a student is one that can be solved by artificial intelligence, the question is whether that task is worth solving with anything other than AI.
Thus, the chief creative director of the company "Alicorn" and the executive director of the Montenegrin Association for Artificial Intelligence (MAIA) Milutin Pavićević For "Vijesti", he comments on the potential introduction of AI tools into Montenegrin, but also education systems around the world.
Pavićević explains that the task of a teacher is to prepare students for the "real world", and that large language models (LLM) and artificial intelligence will exist in it to a much greater extent when students "exit" the education system.
LLMs are a type of deep learning algorithm that processes and generates human-like text. These models are trained on massive datasets containing texts from various sources, such as books, articles, websites, customer feedback, social media posts, product reviews...
The primary goal of LLM is to understand and predict patterns in human language, enabling it to generate clear and contextually appropriate text.
"Vijesti" published a questionnaire in late December, the aim of which was to investigate potential misuses of AI tools in the education system, as well as to identify the best ways to use them and the benefits they provide. 130 people responded to the questionnaire.
Although this is not a representative cause, the same message runs through the majority of responses - students are using AI to do tasks for them, and in order for this to change and for artificial intelligence to be effectively implemented in the education system, a collective effort from state institutions and the educational community is needed.
Don't overestimate the capabilities of technology
Pavićević explains that the general public is at a point where they are starting to realize that AI tools are important in every sphere of life. He states that there are other parts of artificial intelligence that are already used in the education system, but that LLMs in particular are extremely useful today.
"They have experienced a huge explosion in their usability in the last few years and have proven to be incredibly useful in transforming content from one form to another and very useful in automating some processes," the interviewee says.
He says that, on the other hand, insufficient understanding of what LLMs do and what their limitations are leads to us "raising them to the sky" and overemphasizing their abilities.
"In other words, we have too high expectations of it, we trust them too much, and we end up with great risks in receiving inaccurate content, in entrusting them with things they shouldn't be trusted with, and the great desire of people who don't understand the limitations of all this to exclude humans from the loop in which decisions are made. Which, when it comes to this kind of stochastic artificial intelligence, is very risky and dangerous," said Pavićević.
It indicates that LLMs have advanced to the point that in certain matters, e.g. in complex mathematical questions that have already been solved and just needed to be used, if they can answer in the first two or three messages, the best of these models respond at the level of a PhD.
"If the conversation goes 20, 30, 40 messages further, for some questions where they are expected to solve problems that are not very obvious, that are not a direct combination of some knowledge, they are below the level of elementary school," adds Pavićević.
He points out that they are the best at transforming content and the worst at finding new solutions.
"They're very bad in situations where there are a lot of things, even if they're very simple and very easy to combine. If you have 200 or so parameters to analyze, they're simply unusable."
"Resilient" tasks required
When asked in which situations the misuse of AI tools in the education system most often occurs, those who filled out the "Vijesti" questionnaire had a clear answer - students use artificial intelligence to do tasks for them.
"Children use it to solve tasks that they just copy"; "By literally copying information obtained by asking questions"; "Creating written tasks in their native language, children do not read anything, vocabulary at the level of statistical error"; "Interruption of the research process, answers obtained on the fly, negative impact on the development of critical thinking, representing other people's opinions and attitudes as their own, merely copying the content obtained for homework, cheating on tests" - are some of the answers.
Pavićević states that it is up to professors and the system to encourage students to think critically, that is, to give them tasks that cannot be solved using AI.
"In every subject, there is a dose of enigma, where one must go through the so-called 'drill' task. In order for someone to gain the routine they need to think critically in a different direction, it is up to us to formulate these tasks in a way that artificial intelligence can be detected," emphasizes Pavićević.
It says that the professor's task is to check whether the submitted paper is plagiarized or not, i.e. whether it was written with the help of AI.
"We have plagiarism detection systems that can detect such things with a very high degree of certainty. We know how it is done for diploma, master's and doctoral theses. Today, those same systems are even more capable of detecting text written by artificial intelligence, they are 98 percent sure. For each paragraph of the text, it can know with 98 percent certainty whether it is artificial intelligence or not," he said.
Studying differently at colleges and schools
Artificial intelligence already exists as a subject at Montenegrin universities. In undergraduate studies at the Faculty of Electrical Engineering (ETF) and the Faculty of Science and Mathematics (PMF), as well as in master's studies at the Faculty of Information Systems and Technologies (FIST) of the University of Donja Gorica (UDG).
When asked whether artificial intelligence could be a subject in elementary and secondary schools, Pavićević replied that dealing with artificial intelligence in the ways they do at universities requires knowledge of mathematics that goes beyond what is taught in secondary schools, but that it is possible to educate students in AI through some other subjects.
He says that the question is whether this should be done through media literacy courses, or whether a separate course is needed that would deal with artificial intelligence.
"We see that it is part of everyday use and that perhaps children and students in other curricula should be trained in how to use artificial intelligence, what its advantages and disadvantages are, what is available to them, to learn the risks and advantages... And that is part of education," Pavićević points out.
Those who filled out the "Vijesti" questionnaire, when asked how to encourage students to use AI tools as support in solving tasks, instead of using them to automatically perform work, had various suggestions:
“AI should be used when processing topics that require internet sources, not during individual and independent work of students”; “To be used for checking results and interactive learning”; “Show them the possibilities and multiple ways of application, and point out and react adequately to any misuse. Every teacher can recognize work written in this way”; “Change in schooling and education in Montenegro. Conceptualize lectures and knowledge testing as preparation for a project through which the student would demonstrate acquired knowledge and allow the student to use AI tools to any extent”.
Development and regulation need to be harmonized
Speaking about the ethical use of AI tools, and how it can be regulated, Pavićević said that there are two directions in which this can be done.
"One is legal. Where does the problem come from, we know that the European Union is thinking in a very right direction, when it comes to the ethical use of artificial intelligence. But, on the other hand, there is a big race going on right now. You see that we have started a race of large language models, where Europe and America were there about two years ago. And now America and China (where AI is still not fully regulated) are far ahead of Europe," explains the interlocutor.
He believes that stricter regulations have slowed the development of AI.
"It's a big risk for us to say, okay, now we're going to slow down development so that ethical control can arrive. But I think that ethical use control and the ways in which it is done, somehow have to speed up the pace to catch up," said Pavićević.
The majority of those who responded to the question in the "Vijesti" questionnaire, what measures they would propose to encourage the ethical use of AI tools in the education system, believe that stricter regulation and greater engagement of educators are necessary for such a thing to happen.
“Engaging professors to read papers and indicate that they recognize the work that was not written independently. This way, everything passes and children do not give up and therefore do not develop critical thinking...just copy-paste”; “Detailed information and training in proper use for the purpose of better results”; “Until the current staff of professors is well trained in handling detection tools, at least at the rectorate level, automated checks should be introduced, and every seminar/exam paper of students should be run through it”; “Adopting legal regulations in the field of AI”; “Inclusion of the use of AI in references when writing essays and seminar papers. Encouraging creative use of AI technology for basic ideas and drafts that students would then improve. Rewarding honesty about the use of AI technologies (within reasonable limits)”; “Limit AI to a system that explains the process, not provides results”, are some of the answers…
Pavicevic: Don't let the machine value humans
Milutin Pavićević emphasizes that the use of artificial intelligence should not be in student assessment, but that it has great potential in personalizing learning.
"If you want to develop critical thinking in students, to let someone explain things in their own words, which is probably the goal of education, then it's very risky to teach that with a big language model, because the most creative students will somehow be discriminated against because of that," he said.
He emphasizes that a machine should never be allowed to participate in the evaluation of a human without a "human in the loop."
"And that's because sooner or later, they are very inclined, they are very prone to discrimination, or to developing those loops and rewards that will ultimately lead to discrimination," adds Pavićević.
He explains that AI work could tailor curriculum to each individual student.
"To identify where there is a knowledge gap in a large group of students, which may tell the professor that he or she explained something poorly. To turn uninteresting content into educational activities or games that will be more interesting to students. This is where large language models excel...", the interviewee states.
Abuse is an everyday occurrence
When asked in the "Vijesti" questionnaire whether they had witnessed a situation in which an AI tool was misused for the purpose of completing a task in the education system, almost everyone said yes, and among them were those who claimed to be teachers in primary and secondary schools.
They say that misuse of AI tools is an everyday occurrence:
“As a teacher of professional subjects in an electrical school, this is an everyday situation. In programming for the creation of mobile applications, they constantly use AI”; “Yes. It's an elementary school... children are given a topic for a written exam and they do a complete essay via Chat GPT”; “Yes. Since I am a high school teacher, students often submit papers that are not theirs but were done by AI”; “Yes, a simple situation, in a theoretical subject at the university, a good 70 percent of the entire class submitted AI-generated texts on given topics for the exam essay. The professor noticed an irregularity, he said they were weak”; “Yes, as a professor, I have been encountering this constantly in the last few years”; “Yes, as a teacher. From the form of the assignment, I realized that it was done with the help of AI tools”; “I hear these stories every day, especially from close people who are employed in education, that children are increasingly using AI to create written exams, even though they are creative, they do not have enough self-confidence in themselves and their work and rely on artificial intelligence”; "Students in an elementary school asked for and received a composition for writing. As it was in Croatian, the teacher recognized that they did not write", are some of the answers.
Bonus video:
