A Different Type of Student at Our Campus

Stroll through the Neuron building on the TU/e campus on any given afternoon, and you will notice students hunched over laptops, debugging code, analyzing datasets, or hurriedly finishing assignments before a Canvas deadline. As students, who have chosen STEM, we spend countless hours in MetaForum -1, trying to cram a quartile’s worth of Automata and Calculus into our brains.

Meanwhile, just beneath the surface of the technologies we study and use every day, a completely different kind of ''student'' is at work: suave, well-spoken, even somewhat charming, essentially the most popular kid at school. But how does it keep up with studying?. It never drinks coffee at Hubble, it never pulls an all-nighter, and it doesn't sit through lectures in the Auditorium. Yet, it is undergoing the fastest (and largest) education in history. In such a big magnitude actually, that there is little question left as to what the most disruptive technology of the 21st century will be. If you look closely at how modern machine learning actually works, it boils down to a deceptively simple acronym: B.E.H.O.L.D., a Big Engine for Harvesting Observational Learning Data.

What Does It Actually Mean for a Machine to "Observe"?

In traditional engineering, our predecessors have written explicit rules, calling it programming: if sensor X reads data above threshold Y, trigger actuator Z. Nowadays, we no longer program algorithms; we grow intelligence. To do that, AI systems need to "behold" large volumes of real-world data.

Let's take a closer look at Eindhoven. At our campus, "beholding" has taken on a very concrete meaning. Consider what happens inside TU/e’s SPIKE-1 AI supercomputer. In this project, researchers led by Associate Professor Fons van der Sommen fed roughly 250,000 3D CT scans and 75,000 radiology reports into a massive training pipeline. The AI "observed" hundreds of thousands of healthy and cancerous scans, each around 100 megabytes, until it learned to spot subtle anomalies on its own. Quite the difference to studying a textbook for numerous days. The results showcase that AI is on par with its tutors and trainers. What does it mean for the rest of us, students, remains to be seen.

An Anniversary

It is fascinating to think how much AI has grown since ChatGPT took the internet by storm almost four years ago, on November 30th, 2022. Back then, it felt like a clever parrot, for lack of a better word: a student who had read everything but understood little. Today, the same technology is writing a lot of the lecture slides judging by professors' sudden use of em dashes, drafting our own emails and, as SPIKE-1 shows, learning to read a CT scan by watching a quarter of a million of them. Yet, I can't help, but remain skeptical of this student's intentions. That is, if a product by humans can have such.

It is clear that the AI operates under no ethical guardrails, or at least no robust ones. While human students must watch out for plagiarism, cite their sources extensively, and buy incredibly expensive books (my last textbook, on Dutch law in urban planning, cost 75 euros on its own), with little chance of borrowing them from a library, their AI counterpart seemingly faces none of these issues. It flies under the radar, hallucinates sources, draws on undisclosed prior conversations and internal enterprise data, and sometimes (or by now almost all the time) even disembowels rare books at what seems on an industrial scale, books that will forever be lost in its context window. Heck, even this article will probably end up as a footnote to be used as it sees fit.

A Conclusion

The AI student has come a long way in a very short time. But its story is also a reminder of what makes our own education different. It learns by watching at a scale no human could match, while we learn by asking why, what and how; by getting things wrong (quite often), and by being taught by people, who care whether we understand (and here I hope and believe this applies to all professors on this campus).

Behold the AI student, by all means. Just don’t stop being a student yourself.