AI at UNamur: exploring, innovating, guiding

Typically, when conducting research, ChatGPT facilitates the creation of an initial state of the art. The first limitation is that it can only reproduce what it already knows, in other words, the knowledge available at the time it collected this information. Nor can we take what ChatGPT says at face value: it requires a critical eye on the part of the student. The fact that it presents itself as a conversational tool also gives a different perspective on the answers it provides: it feels like you are actually talking to someone, when in fact its response is nothing more than a sequence of words. So the sequence could be different and the resulting text could say the exact opposite.

The world of AI is currently based on a reactive intelligence system. I am trying to implement deliberative intelligence. One of the major limitations of most current AI systems is that they learn to solve a single task based on millions of examples. That is not intelligence. Having segmented knowledge is absurd! Intelligence is about solving new problems and adapting based on the knowledge you already have.

Personally, I am a big believer in the idea of mentoring in education, in one-to-one relationships: from this perspective, ChatGPT can be seen as an opportunity for students to have a personal digital assistant at home.

I believe it is necessary to introduce students to this tool and that it would even be "dangerous to ban it." We need to move away from stereotypes that teachers are necessarily old, out-of-touch fuddy-duddies and students are necessarily cheaters... Our job is to teach students how to evolve with the tools that exist because they are the ones who will change the professions in the future. This is all the more important given that many students, even if they are very comfortable with social media, do not have a digital culture: many do not know what a filter bubble is (editor's note: a system for personalizing search results) or that on Tinder, the algorithm matches attractive people with other attractive people...

With the arrival of the ChatGPT conversational robot at the end of 2022, all sectors realized that it was now impossible to ignore the challenges associated with AI. Departments of medicine, veterinary science, management, communication... everyone started asking me for help, but in the Faculty of Computer Science, we had been interested in AI for a long time and this topic was an integral part of our courses. However, over the past 10 years, we have seen AI permeate all strata of society at an ever-increasing rate. With ChatGPT, this acceleration has continued. In a way, it was no longer credible not to have a dedicated teaching unit.

ChatGPT can quickly lead students astray if they don't use it critically. That's why I suggest they use it and cite it as a source, just like any other source, by including the results in the appendices. This is a way of using the tool honestly, showing where it is right and where it is wrong. As with any investigative work, it must obviously be considered as one source among others and as an unreliable source since it works on the principle of probability rather than truth...

There is always enthusiasm for a new tool, but we must obviously remain critical. We went through a phase of analysis to understand the possibilities offered by ChatGPT, but also its limitations. It is a wonderful tool, but it must remain a tool, not the core of an educational system.

ChatGPT works on the basis of a probabilistic equation. It will string together the most probable sequence of words based on the question you have given it. This sequence of words will form sentences that make sense and are interesting because ChatGPT has been trained on a large amount of data, some of which is known: the entire English Wikipedia, the entire Reddit social network, two large book databases equivalent to a gigantic library, and the rest of the Internet, including Twitter.

It would be dangerous to believe that AI will simplify language learning. These tools can be useful for improving the quality of a text, but caution is needed. AI works relatively well for factual texts, such as contracts, but when it comes to academic texts requiring argumentation, it quickly shows its limitations. A text generated by AI can be spotted fairly quickly, as the author's voice is absent. Without this voice, there is no personal argumentation. I prefer to read a text in imperfect but authentic English rather than a perfect text produced by a machine. AI can be a good tool, provided you know how to use it intelligently. It's a bit like a dictionary or a speech manual. It can be a good learning aid, but it shouldn't be overused.