Interview with Juan Carlos Farah

You did your PhD within the REACT Group, which specializes in the design, implementation, and adoption of innovative information systems across an interdisciplinary range of domains, such as social media for digital education. Can you talk about your PhD journey and your research?

My PhD was centered on integrating chatbots into digital education platforms. At REACT, we develop Graasp, an interactive learning experience platform designed to facilitate the creation of online lessons. The goal was to study how chatbots could support these lessons.

My role was primarily focused on systematically integrating chatbots into interactive educational activities. I went from simple, rule-based conversations to integrating more advanced large language models as they became available.

The timing of it was also interesting—I began my thesis in February 2020, right before the pandemic hit. The sudden shift to remote learning was a real catalyst for exploring the added value of chatbots in educational settings. Especially then, when direct teacher interaction wasn’t possible.

We explored several dimensions. From using advanced language models to addressing privacy and ethical concerns, and streamlining the integration process for teachers.

How does that tie in with your work at the Laboratory of Cognitive Neuroscience?

I collaborate with Professor Olaf Blanke on integrating conversational agents into neuroscience experiments, particularly focusing on bodily self-consciousness and its relationship with brain functions.

There’s a clear overlap in the technologies we use. For instance, LNCO relies heavily on VR, and in education, there’s an effort to promote VR. Numerous technologies allow us to immerse people in similar contexts, which lets us quantify specific effects. This intersection of technologies provides an exciting opportunity: developing these technologies in parallel and customizing them for important domains like education and healthcare.

You started out with a degree in economics, can you share how your path led you to the field of education?

My long-term goal was always to get to build these products.

My journey started with undergraduate studies in economics and already back then I was always attracted to the analytics aspect… to the data science.

It was a time when online technologies were really starting to explode. Facebook had just started. YouTube was just around the corner; Google was getting its biggest push. There was this big emphasis on how technology was going to shift everything.

And then there was social media. The fact that you could get hundreds of thousands, millions of people around these communities… it kind of opened my eyes to how these technologies would profoundly change how we interact.

My long-term goal was always to get to build these products. That took me through a very long path where I worked in different industries, usually doing some data analytics as well as some research, up to the point when I decided to enroll in a Master’s in Computer Science at Imperial College London. That’s where I did my first real research project within a group that worked on computational neuroscience. So, I already had these inclinations towards neuroscience from back then.

Then I started thinking about what the next logical step was. I started looking at schools to do a PhD and I found EPFL: I applied, I got in, I was very excited and then I needed to find the lab where I’d do my PhD. Just a few days later, I got an email from Denis Gillet, Head of the REACT Group, saying that he was looking for people to work on a learning experience platform. This really piqued my interest because this was an opportunity to work on a big platform with hundreds of thousands of users and to bring together several of my research interests.

Graasp is an open-source platform that facilitates the creation and sharing of interactive lessons. It is used worldwide by universities and schools for blended active learning and digital knowledge sharing. You’ve contributed to the development of the project and its new iteration. A driving feature of the platform is that—provided consent—it is possible to collect data regarding teacher and student (learning analytics), which can be used not only for pedagogical awareness and reflection but also for evidence-based research. Can you talk about your role in this project?

While I was focused on developing my research expertise, my strengths were mostly on the software engineering side of things. And I realized that both were important for making good online platforms.

When I encountered Graasp, seeing the potential impact and Denis Gillet’s vision for the platform intrigued me. Despite my software engineering background, I quickly realized that research and user-focused approaches were crucial for impactful software development.

While I was focused on developing my research expertise, my strengths were mostly on the software engineering side of things. And I realized that both were important for making good online platforms. Graasp was partly developed within a big project involving many universities in Europe. That meant that there was a strong focus on functionalities that were broadly applicable across different borders and across different curricula. Talking to teachers who used our platform showed me how much impact our work could have. This made me see how important it was to understand what users need.

During my PhD, we did two main things regarding Graasp. We made a new and better version of it (graasp.org), which took a lot of time and involved many people. Now, only after a few years, we are starting to see the results.

We also sharpened the focus on open data, so everything is open source, and we have been very dynamic to be compliant with all the new privacy legislation. And all the data is based in Switzerland.

What’s your vision for the future of these technologies?

My long-term research is focused on our interactions with artificial agents because I feel that it’s going to become more and more common to have digital teachers or assistants around. We already widely use Siri and Google Assistant, for example. And I think that studying these interactions in the context of education is very important, as there’s a huge potential that we’re only starting to see just now.

It doesn’t only require massive breakthroughs in natural language processing, such as what we are seeing with GPT, BARD, etc. It also requires good design. So often people will say, “Look at what ChatGPT can do!” but they forget that there’s much more than the model that is behind that success. There’s all this feedback that comes in. And there’s also the identity that we interact with, in the case of ChatGPT through a website or an app. If these models are used to power, let’s say an artificial Albert Einstein, I think it would be very interesting to see what the effect of being taught a lesson by a virtual Albert Einstein would be.