Starting the AI degree in the middle of the debate about its limits: “It’s hard to believe that it could destroy the world”

BarcelonaIt had only been two days since Pau, Jofre, and Nil had started classes when news resonated around the world: "Anthropic engineer resigns for fear of AI". While the head of alignment science at this company claimed that this technology "could kill us all by the end of this decade", the three of them were entering the Barcelona School of Informatics to face their second day as students of the artificial intelligence degree at the UPC.

During their first weekend as university students, tech giants urged slowing down the pace of AI model development and OpenAI postponed its IPO due to the dangers of AI. "I think we have to analyze well the whys of these news and these announcements... because perhaps behind this idea of agreeing to try to set regulations there are other motives, such as that their projects were not yet profitable enough," Nil suggests. He is also honest, however: "It is true that we have just started and, therefore, we do not know much more than a person who is interested in the topic".

Faced with the question of whether they believe AI can get out of control, Pau dares to speculate further. "Right now there is already a case that they tried to hide a little about some AIs that got out of control and started talking to each other... Luckily they found out and are focusing more on limiting where they can go," he explains. "It's a tool and in the end you can use it both to find a cure for a disease or to make a biological weapon... it all depends on who uses it, but I do believe it can get out of control," he adds.

In contrast, Jofre acknowledges that the controversy feels far away: "I am not very informed about this, but if it can get out of control perhaps first we need to think a little about where we want to go and monitor behavior to move forward with safe steps. What is true is that it is hard to believe that it can destroy the world".

All three comment that these first days of class they did discuss "a little" this news with the professors and that they have given them "some notions" of the dangers of AI. For example, "the fact that AI models talk to each other and how to limit what they do among themselves", among other things. "To be able to understand the news that comes out today and especially to understand how AI works, we are reading articles on how the whole discipline started," details Nil.

More than 50 years with AI

Studying the origins of AI, the three students have discovered that, although it seems like a very recent technology, artificial intelligence was born in the 50s. "When I studied computer science, there was already an AI course," explains Lluís Márquez, professor of the degree and vice-dean of the Faculty of Informatics (FIB) at the UPC. "This faculty has been quite a pioneer in incorporating it. In the late eighties, a specific doctoral program began, in the 2000s it continued with a master's degree, and more recently, five years ago, the degree was created," insists Márquez.

Regarding the boom of interest in AI, the professor attributes it to the development of different machine learning and optimization techniques, which are the basis of what form generative models. "What has happened is that a few years ago, generative artificial intelligence models began to be able to do things that until that moment had not been seen," he describes. Regarding the fear and catastrophic messages, Márquez assures that they have existed for years, but that, "like all technology," once it reaches maturity and we can regulate it and have security and development frameworks with control, "it does not have to be dangerous." However, he admits that the fact that the "spearhead" of its development is in the private industry makes it difficult for companies to agree on regulations.

"From the academic sphere, they already said that we had to see how we could develop it responsibly and ethically and avoid the possible harmful purposes for which some might want to use it", insists the professor. In this sense, the AI degree at the UPC, which has 50 places and a cut-off grade of 11.5, includes the subject of ethical and social aspects of artificial intelligence. It is 150 hours of class where, among other things, exercises are carried out on the application of ethical criteria in the design of systems, algorithms and the use of data. In this sense, they focus on aspects such as security, privacy, transparency, respect for human rights or the prevention of biases (of race, gender, religion, territory, etc.). Furthermore, the vice-dean explains that in some subjects such as software engineering, responsible development is also worked on, and that in the Final Degree Projects (TFG) students must justify that they have not used personal data or information.

Not knowing what it will be like when finished

Predictably, Pau, Jofre, and Nil will finish their degree in four years. Four years ago, the idea that AI would have entered our lives so much seemed almost impossible. "Not knowing what we study will be like in four years is an extra motivation to keep updating ourselves," says Nil. Jofre considers that "in four years there might be a very big leap," but that they will adapt to it. On the other hand, Pau is more direct: "Right now it is growing at an exponential level and I think they will have to slow it down a bit because otherwise it will get out of their hands."

The vice-dean, once again, calls for calm: "Computing has always developed quickly and in AI there is a core that is quite stable and that everyone must know. They are engineers, therefore, they must know mathematics, they must know how to program... And even if AI is capable of solving some of the tasks, we humans must continue to learn them... We cannot keep up with training profiles with these skills that remain the same and that are being demanded more and more each time," he states.