Â鶹ÒùÔº

Artificial intelligence is here to stay. Most people probably recognize that by now. But there is still skepticism about the use of AI in many areas. One of them is education.

This may bring to mind students using generative AI tools that can answer all kinds of questions and practically write finished assignments. This can lead to concerns about a lack of learning and challenges related to using these tools to complete assignments.

Research from the Massachusetts Institute of Technology (MIT) has reignited the debate.

In short, the MIT research showed that there is less activity in the brain when using an AI tool such as ChatGPT to solve a task. EEG scans of the test subjects showed that fewer connections occur in the brain when using a standard generative AI as an aid instead of writing a text yourself.

The MIT research also showed that students who had used ChatGPT had more difficulty remembering what they had written themselves. In other words, there was less neural activation and less actual learning when the test subjects used AI tools.

The MIT article, which was to the arXiv preprint server, has been criticized in the major journal Nature, which conducted by Professor Guido Makransky and his colleagues at the University of Copenhagen. Here, they have taken a different approach.

ChatTutor

Professor Makransky from the Department of Psychology and his colleagues have collaborated with the company ChatTutor to design an AI tool specifically to be used by students in learning scenarios. Their studies have shown that this AI tool works better than other generic AI tools (ChatGPT) when it comes to learning. Their findings are in the journal Educational Psychology Review.

"Our work contributes to the discussion about the use of AI in education. A part of the debate surrounding AI revolves around how the use of artificial intelligence might erode cognitive processes that are important for learning. In our research, we simulate the role of a good teacher by creating a chatbot that stimulates reflection and deep cognitive processing. The chatbot scaffolds learning by adaptively prompting students to explain relevant concepts based on their specific knowledge gaps," says Makransky.

The tool was developed in collaboration with the company ChatTutor. And, as the name suggests, it was created specifically to support learning.

Specifically, artificial intelligence can help students by asking exploratory questions and encouraging them to reflect on their answers. Just as you would imagine a tutor or a teacher might.

"Our AI tool is based on theories and existing evidence about how people learn. We learn best when we are cognitively active. This means that we need to help students reflect on what they are learning themselves, rather than simply provide them with answers, thereby fostering their ability to understand concepts and apply them in relevant contexts rather than merely improving performance," says Makransky.

If you use ChatTutor, the artificial intelligence can analyze your answers to a given question and identify which parts of your understanding of the question are correct and which are incorrect. Moreover, it can support you and guide you in the right direction by asking clarifying questions that form the basis for reflection, instead of just giving you the answer.

"In this way, AI can simulate the teacher being present with each individual student at all times. The cool thing about this is that you can suddenly do this with 250 students at once, which I could never manage as a single teacher. However, it is important to have the teacher in the loop, that is, in our experiment, the instructors design the reflective questions, and students can involve their instructor in the student-AI interaction," says Makransky.

AI tested on UCPH students

Makransky and his colleagues tested ChatTutor on students from the University of Copenhagen. "We conducted an experiment with 175 students, comparing ChatTutor with ChatGPT and regular teaching without AI assistance. We measured how much the students learned, whether they liked using the tool, and whether they trusted it," says Makransky.

They subsequently repeated the same experiment with 234 high school students.

In both cases, it turned out that the students learned more by using ChatTutor compared to both the ChatGPT group and the control group, which did not have access to AI tools.

The researchers also found that the students in the ChatTutor condition reported significantly higher enjoyment when using the tool, and were more likely to intend to use the tool again compared to the other conditions.

"This shows us that we need to use generative AI in a way that supports learning, and not as a machine that can just provide us with answers," says Makransky.