To his own surprise, Arnoud Kuijpers, teacher of Dutch at the Coornhert Gymnasium in Gouda, became the object of research into ChatGPT. Or rather, the guinea pig: “My pupils had to submit a reflection, a text intended to make the reader think. One of the pupils writes for the website scholieren.com, and together with the editors, he came up with a plan to have his assignment created by ChatGPT. Just to test how well it works.” The pupil fed the text requirements into the AI tool, and within a few seconds, he got back an automatically generated reflection. “I gave him an 8.3.”
A clear pass, for sure. But this was not as simple as pressing a button. The pupil had added some quotes of his own, and he had manually rewritten several paragraphs. So he did put some time into it in the end. This may also be the best way for the educational system to deal with this technology, Kuijpers thinks. “I don’t think we should ban it. I think it’s better to teach students how to use it to improve their writing assignments. Some teachers want to go back to handwritten assignments. I don’t think that’s the right way to go.”
Uses in education
“Make it fun!” says Radboud University education innovation expert Inge Molenaar. She has decided to make ChatGPT into a topic of research for her students. “Let it write the first draft of an essay. How exactly is this kind of text from ChatGPT structured? What goes well, and where does the program make mistakes? Can you make it do things that don’t make sense on purpose? These are interesting experiments. In my opinion, it's a wonderful tool to use in education. Let’s see if it can improve students’ writing skills.”
According to Molenaar, there is no point in getting defensive. She finds complicated measures, such as the return of pen and paper at Australian universities, to be old-fashioned. “This is just a consequence of automation. In the past, many jobs already disappeared due to the advent of robots, which took over work that we no longer wanted to do: repetitive tasks, reproducing things. The added value of humans lies mainly in higher cognitive skills: controlling, assessing, and connecting with feeling. This is also what we should focus on in education, rather than on reproducing knowledge.”
Race and gender
ChatGPT is not completely new, as we explain here. Actually, the computer program is an evolved language model that has access to a huge number of texts written by humans. It has learned from all these texts which words belong together, and can use this information to continuously predict the best next word. This allows ChatGPT to form linguistically correct sentences and come up with answers that are correct – or at least seem correct.
For scientists like computational linguist Raquel Fernández (University of Amsterdam), ChatGPT is primarily a next step in the long-term development of this kind of language model. “But we’re also impressed with the quality of ChatGPT. With what it can do, and especially with the high quality of the texts it produces. It really is more than just a fun toy. Still, I don't think it’s going to radically change our lives. I’m mostly interested in it from a scientific perspective: the fact that we as humans can build something like this. In turn, it can also help us learn more about language and technology.”
ChatGPT has been programmed to avoid discriminatory or racist statements, but according to Fernández, these limitations do not resolve the problem. After all, the language robot itself does not sense this sensitivity – it is still a machine. “For example, if you ask it to write a program that selects job applicants based on race or gender, it will do it.” In addition, ChatGPT texts sound like they express an irrefutable truth. “The texts it creates are so convincing, so human, that people may be quick to believe what the texts say.”
This danger is comparable to that inherent in other online applications. After all, people are already allowing themselves to be fooled by anything they see on the internet. “The difference is that you don’t see what the source is.” If you find some information on a website, you can at least judge for yourself whether it is reliable. Fernández: “It is a major limitation that ChatGPT does not disclose its sources and the developers do not disclose the technology behind it.”
Moral risks
According to cognition philosopher Anco Peeters (Radboud University), ChatGPT is so popular mainly because it makes AI accessible. Everyone can talk to it and do whatever they like. This may impact our work, just as email, bar codes and search engines have also changed our work. Writing professions may even disappear due to the advent of ChatGPT and its successors. “But is this really a bad thing? There are people now who manually create the descriptions of shop items in web shops. These kinds of professions may become obsolete.”
For the more content-based writing professions, this technology does not form such a big threat, Peeters thinks. After all, writing a good novel, film script, scientific or journalistic article is about more than just stringing words together. “ChatGPT is good at parroting, as one of my colleagues said. It predicts what answer you want to hear, but it doesn't really understand it. If you ask for a connection between two events, it will look for this connection, whether it exists or not. It is good at inventing information, and bringing it across very convincingly. This is a paradox, because you have to be an expert to know whether something is true.”
ChatGPT may also reinforce certain power relations, Peeters warns. “The traditional relationship between men and women, or between rich and poor, these are ideas that it just spits out in its texts now. What’s more, this technology is typically something that benefits people who have easy access to new technologies. With a poor internet connection – in developing countries, for example – you cannot do so much with it.”
From calculator to typewriter
Teacher Kuijpers compares ChatGPT in education to the calculator: initially forbidden, now mandatory. “The quality of the texts ChatGPT produces leaves much to be desired, but that is not really relevant. Clearly, it is only going to get better. Nor can we check these texts for plagiarism: they are too unique for that.”
Therefore, according to the teacher, it is better to accept and embrace this technology as we did calculators, Wikipedia, and the internet in general. Because in a way, the biggest risk of ChatGPT is similar to the risks of the internet. “It fits in with the wider danger of the unreliability of the internet. ChatGPT can also talk nonsense and concoct plots. But it often sounds trustworthy. We need to find a good way to deal with this in education.”