Text by Ville Heirola

Originally from the Czech Republic, Filip Ginter got his PhD at the University of Turku, and now functions as a crucial component of the Turku NLP lab the same university. The TurkuNLP group is the other main hub of natural language processing expertise in the country along with the University of Helsinki’s NLP project. In addition to that, Filip Ginter has plenty of experience from the private sector. At Lingsoft, he is a technical architect and, at Silo.AI, a part-time AI Scientist. He balances his time between these roles, all the while making time for furthering his research at the University of Turku and mentoring his students.

As part of the UNICOM project, we talked to Filip Ginter about his experiences as an international talent who has moved between the public and the private sectors. We discussed his career path, differences between working at a university and a company, and how he would like to see the AI scene in Turku develop.

Luck and Hard Work

Filip Ginter has been working with natural language processing – a crucial subset of AI – for a while now. In 2001, he got his M.Sc. in Computer Science and, in 2007, graduated with a PhD. Since then, his research and teaching have afforded him the position of assistant professor of language technology. In that time, the TurkuNLP Lab has gotten its start and become a pioneering hub of natural language processing expertise.

Filip Ginter originally came to Turku for personal reasons and was lucky enough to find a thesis instructor at the university. He is quick to point out that his path may not be the most viable option for many international talents coming to Turku today.

— Many of the institutions I went through don’t exist anymore, both at the university and in terms of the authorities. From what I’ve heard from my students, it can be much more difficult now to deal with the various permits, grants, and other necessary papers. There’s also much more emphasis on knowing the right people and having your plan ready before you apply. In that sense, while I had good papers to apply with, I was also lucky, and my experience may not translate universally.

The Private Side of Things

While research and creating new knowledge is a matter of the heart for Ginter, a need to explore other avenues presented itself early on. Making the transition from public to private was surprisingly painless.

— I quickly rose through the ranks and got my own projects and staff to work on them. That’s when I wanted to try something else on the side. I love the hands-on, technical work so much, and wanted more of it. That’s why I became involved with both Silo.AI and Lingsoft. Again: networks played a huge role in my landing those jobs, as the people hiring knew my work and what I could do. And the companies work with the university as well on various projects, so the bridge was already there as we were aware of each other. Of course, I was also interested in simply finding out what the companies were doing.

According ot Ginter, both realms have their pros and cons:

— I went in to experiment. That was a a hugely positive experience, that the companies were supporting and enable me to do this more experimental work that’s hands-on and technical. Silo.AI and Lingsoft are very different from each other, too. Even more surprising were the great colleagues. However, there are some things I could still only do at the university. One is raw computation power. The university simply has much more of that available, which, in turn, enables me to do the kind of work that’s impossible elsewhere. In the private sector, the work is much more about applying what we already knew instead of creating new knowledge, which is, of course, absolutely fine.

However, Ginter says that Turku-area companies working with AI have gotten increasingly more interested in taking chances with new research.

—In recent years, the R&D side of things at companies has gotten more prominent, however. I think every company needs the kind of experts who have deep knowledge about how their products, methods and tools work. But developing completely new things takes time and resources and is much more trial-and-error than often though. The university model is great for that, and Turku has it good in that regard.

Ginter says that recent graduates tend to find employment rather quickly. This ties into the lack of skilled labour tech companies are facing right now. For Ginter, balancing between the university and the companies forms a perfect way of keeping up with the advances in the field.

— The bridge between universities and companies is working fairly well in my field. Machine learning, AI, natural language processing are all very hot properties right now, and many find employment in the private sector. Of course, we do need someone to stay and keep the programmes running. Yet, the model I have adopted, of working part-time as a researcher and part-time in a company, might not be ideal for all. At the same time, it’s an opportunity for me to learn new things and try things out in a field that’s constantly evolving.

Communication and Increased Understanding

While the potential of natural language processing is huge, many still have unrealistic or misguided notions about what AI can do and what it requires to work. All machine learning applications require a large dataset to work from and solutions do not just appear magically out of thin air.

— Right now, AI and machine learning are in a bit of a bubble, or have been so for years. There are a lot of inflated expectations and less-than-true promises, but AI has also achieved great things and it has so much potential if applied correctly and with the right expectations.

The consequences of current trends in AI are difficult to predict. Trial and error, both in companies and in universities, is crucial for pushing the field forward. Sometimes, serendipity plays a central role. Ginter says that with the TurkuNLP lab and many cutting-edge companies, the Turku region is a fantastic place to work with NLP and AI.

— For example, in 2013, we thought that it would be cool to gather a few billion words’ worth of Finnish. Nobody wanted to fund it at the time, and everyone thought it was ridiculous. Then we got some funding, and that corpus was key in developing the FinBERT model, which can be and has been used in all kinds of AI applications. And it began with just this idea of “wouldn’t it be cool if…”. I don’t think much of that could have been funded at the time. That was the kind of thing nobody expected.

Like Sepinoud Azimi in our previous interview, Filip Ginter also makes a point of emphasising the need for communication and increased understanding between everyone working with AI: universities, companies, students, professors, clients looking for AI solutions, you name it.

— Silo.AI used to hold these AI Meetups each month, where 50-60 AI professionals in the region got together to discuss what was going on. That’s the kind of interaction I’d like to see more of. Talks, seminars, get-togethers. It comes down to increasing communication and making sure everyone knows what AI as a tool is best-suited for and what needs may be met maybe even without using AI at all.