“I like taking simple research ideas and turning them into something that can actually be used by people.”

You work as technical lead in the AI for Knowledge group at the IBM Research Zurich laboratory. What does your average day look like?

I work with people all over the world, so in the morning I usually start by catching up with what happened in other time zones. I have meetings with my group or calls with clients. Then there is time dedicated to working on our projects where I might for example do some coding.

What projects is your research group working on?

When I first joined after finishing my PhD in theoretical physics in 2017, we just started out with one key idea: knowledge exploration. The vision was to build a system that could collect the knowledge from publications and render it searchable. This turned out to be a bit more challenging than anticipated. One of the first problems we encountered were PDF documents. Publications are usually in PDF format, and while those are good for displaying content on a page and printing, it is not easy at all to extract information from them. In a PDF, you have lots of little boxes which don’t necessarily correspond neatly to lines or words. And don’t get me started on tables! 

Since PDFs are so widely used, a technology that will convert them well is valuable. Hence, we started working with clients straight away. In 2018, we even had the opportunity to work with the product team at IBM to turn some of our research into part of an official IBM product. That was exactly the sort of experience I was looking for in my work. I like taking simple research ideas and turning them into something that can actually be used by people. 

PDFs are tricker than I thought! Do you have an example for another challenge that is keeping you busy these days?

The key idea has remained the same since we started, but of course new developments come with new challenges. Last year, we experienced a huge disruption when it comes to Artificial Intelligence. Large language models are enabling us to do things we previously couldn’t. Together with many others at IBM, our group started working on solutions for some of the issues with generative AI.

When you ask a question to an AI, it will tell you a nice, but not necessarily true, story, using the data it was trained on. If you are a company and you would like to use AI to work with internal information, the model will probably not have been trained on that, so its answers won’t be that useful to you. Moreover, you might not want your data out there for anyone to access! We managed to find a way to store your internal data such that the AI can search it and include the relevant pieces in its generated answers. This way, you also know where the AI is taking its answers from, making them more trustworthy.

Participants of the Physics Olympiad visiting IBM Research in Zurich during the 2024 second round. (Image: Sebastian Käser)

Some of the young researchers in the Science Olympiad community are probably wondering whether they should pursue an academic path in a university or enter the industry. What has your experience been like?

The thing is that IBM Research truly feels like the middle ground. We apply research in the industry, but are less product-oriented and deeply connected to the academic world. We even collaborate in research projects funded by the Swiss National Science Foundation. There are twelve IBM Research labs in the world and our mission is really to be the organic growth engine of the company. Not only do we strive to bring innovations to the market but we also explore the future of computing itself, for example by researching quantum computing.

The IBM Research Europe – Zurich lab covers four research areas: Accelerated Discovery & AI, Hybrid Cloud Research, Security Research, and Science of Quantum and Information Technology. The campus also includes the Binnig and Rohrer Nanotechnology Center with a cleanroom and noise-free labs. Four researchers were already awarded two Nobel Prizes for their seminal work at IBM Research - Zurich. Learn more: IBM Research Europe | Zurich

In your work, are you mostly assigned to solve specific problems or do you get to come up with and pursue your own ideas?

There is a lot of freedom. If you have an idea you want to publish or present at a conference, go for it! When working with a client, such as a private company, obviously you are assigned to solve the client’s problem. However, we really get to bring in our expertise and clients appreciate our input. Sometimes, we have so-called joined development agreements with clients, collaborating to solve a problem together.

Which three tips would you give to readers who are just starting their career?

One: Interact with your colleagues and integrate in the teams that you join. It’s always appreciated if you put yourself in a role where you are helping others, not just asking them to help you. Two: Don’t just do things for their own sake but think about how they could be of service. There are many ways to make your work useful to others: You could offer your code open source, you could build a community of people with shared interests or make your work into a product. Three: Volunteer! This will get you lots of opportunities to experiment and build new skills. Leadership, teamwork, accounting, running a website… you will use this in your career eventually.

Michele Dolfi (left in the front) as part of the Swiss delegation at IPhO 2006 in Spain. (Image: Alfredo Mastrocola)

You volunteered for the Science Olympiad after participating in the Physics Olympiad in 2005 and 2006 and even attending the very first OlyDay with former federal councilor Pascal Couchepin. What impact did the Science Olympiad have on your career?

I’m almost sure everything would have turned out differently if I hadn’t been involved in the Science Olympiad. During my PhD at ETH, it turned out during a coffee break that many of my colleagues and even one of the professors participated in various Science Olympiads. I was not the only one who shared this experience. When we organized the International Physics Olympiad in Switzerland in 2016, we made the tool “OlyExams” which is now widely used. I still use the tools I used for “OlyExams” while working for IBM.

IBM Research Europe - Zurich shares the Science Olympiad's mission to promote and develop intellectual talents. The Zurich lab allows participants and volunteers to explore interesting activities across a range of scientific fields, connecting different disciplines, theory and practice.

From 2016 to 2020, you were president of the Science Olympiad. In the 20 year history of the umbrella association, it was one of the most dynamic times: a new corporate identity, a new co-director, new associations… Is there anything you would do differently looking back?

We wanted to push the sentiment that, rather than ten Science Olympiads, we are really united as the Science Olympiad. This was very well received by some people, especially those active in the board, but the message was also controversial and easily diluted as it trickled down to the members of each association. I suppose we could have done more, but in the end, you can never do everything. 

For the future, I would encourage the Science Olympiads to recruit their members not just among former participants, but teachers as well. A single teacher can attract several classes of students to the competition and do so consistently over many years. It would be invaluable to have more teachers as volunteers.