Eye-tracking is essentially a way of measuring where a person is looking and how their gaze moves. Our eye movements are rapid and largely outside our conscious awareness, so there is often a gap between what people can later report and what actually happened while they were looking at something. An eye-tracking system can continuously record where someone is looking and how their gaze moves over time.
The eye-tracking measure gives us a relatively objective record of where and how a person looked, while the person can help explain the ‘why’ afterward. Sometimes that explanation contains a lot of interpretation, which the eye-tracking data helps us identify. By combining eye-tracking with questionnaires, interviews or observations, I am able to address an enormous number of research questions. For me, that intersection of the objective and subjective is really the essence of the research work. I use this approach to primarily examine learning processes, methodological and design issues that are part of the learning process, and user experiences in the context of digital learning. Importantly, the aim is not to infer what a person is thinking from gaze alone, but to use gaze behavior as one source of evidence within a broader understanding of the learning process.
One of the studies I am particularly proud of focused on students’ critical thinking in the new era of Generative AI. This was a joint study with Prof. Kurtz, Dr. Segev, and Kahana and Fogel-Raz. We compared students who had received critical thinking training in identifying AI-generated images with students who had not yet received this instruction. We wanted to know not only whether the training improved their ability to identify deepfakes, but whether it actually changed the way they visually examined the images. This is where eye-tracking was particularly valuable, because it allowed us to examine the process leading to the decision, rather than only the decision itself. And we did see a difference. The trained students showed longer and more broadly distributed visual attention and were also more accurate in identifying deepfake images. However, they were also more likely to over classify images as AI-generated. Interestingly, both groups overestimated their ability to detect deepfakes. So, although the training changed both how students examined the images and how accurately they classified them, it did not give them a more realistic perception of their own detection ability. For me, that is one of the interesting contributions of the study. Eye-tracking allowed us to see that an educational intervention can change not only performance, but the visual processing strategies students use while making a judgment.
At the lab, our projects are very diverse, with many stemming from student or collaborator interests. For example, the COVID-19 pandemic saw a shift to online learning. This led us to ask, how does the visual configuration of Zoom affect the interaction between lecturer and student? Using eye-tracking, we examined how lecturers look at their screens while teaching on Zoom. We were interested in where they direct their attention and how that compares with what they believed they were doing. We found gaps between what lecturers perceived themselves as doing and what they actually did. For example, one lecturer, while answering a student’s question was actually looking at a presentation slide that contained nothing relevant to the answer. When we asked the lecturer afterward whom he had been looking at, he said he had looked at the student. The eye-tracking data showed that he had actually focused on the presentation. This highlights how Zoom is changing interpersonal communication, because the spatial relationship between teacher and students is no longer shared or fixed as it is in a physical classroom.
In another project, a collaboration with researchers at Sheba Medical Center, we are exploring clinical training practices. We use eye-tracking, for example, to examine how students divide their visual attention while examining clinical information and medical images. By seeing what they actually attend to, rather than relying solely on what they later report, we can better understand how they process information in complex clinical situations and use that knowledge to improve training procedures.
We are also beginning to combine eye-tracking with other measures, such as facial-expression analysis and Galvanic Skin Response (GSR), which can indicate changes in physiological arousal. In one virtual reality study, for example, we used 360-degree cameras together with GSR while students were learning in a virtual reality setting. We were particularly interested in whether specific moments in the learning experience were associated with increased physiological arousal, and then used the broader context and additional measures to understand what that response might represent.
I think eye-tracking, together with other non-invasive biometric tools we are already using in our lab, is moving beyond simply showing us where people look. I see the field increasingly moving toward understanding the learning process as it unfolds. Its real potential is in making processes visible that are otherwise very difficult for learners, or even researchers, to describe. When we can see where attention is allocated, where effort appears to increase, or where people may be struggling, we can use that knowledge to improve the learning environment rather than simply evaluate the learner. This is where I find the field particularly exciting. As these tools become easier to combine, and their software and machine-learning capabilities become more sophisticated, we can build a richer picture of the learning process as it unfolds and ultimately design learning experiences, technologies, and environments that better support learners.
Additionally, as in many other fields, AI is helping advance eye-tracking research, but it is also creating new and exciting research questions. I am particularly intrigued by how GenAI is altering digital learning processes. One question that interests me is how we read and learn when using GenAI. I am not necessarily referring only to the final outcomes of reading and learning, such as reading comprehension, but to the process itself: Where do students direct their attention? What do they choose to read, skip, or revisit? Where do they invest effort?
In a world where GenAI is changing how we learn, create, and consume digital content, I think eye-tracking will become increasingly valuable for understanding not just what people learn, but how that learning process changes. Understanding that process can, in turn, help us design technologies that better support learners.
