Meet Dr Denny Vrandečić - Head of Special Projects, Wikimedia Foundation

AI is transforming the way we create and access knowledge. In your view, what is the most exciting – or the most concerning – change we’re seeing right now?

AI understands natural language on a level that we haven't seen in any previous technology. AI systems like large language models (LLMs) can transform the user experience, and help many people access knowledge and use it in many new modalities and devices. I think this is extremely exciting, and that we can help a lot of people with these new technologies. 

On the other hand, what is a bit concerning is the current architecture of these systems, which was a major part of my talk. Because they're so fluent, we tend to associate fluency in natural language with intelligence, and intelligence with knowledge, and and so on.

LLMs appear extremely competent when they give an answer, even when the answers are wrong, which happens quite frequently. We haven't yet developed any methods for being cautious around the systems.

They are great for many different tasks. For example, If I'm asking it to rewrite a piece of text, help me formulate an e-mail or assist with creative writing, they can be extremely useful and helpful. But for accessing knowledge it seems the current architectures are just not there yet.

Dr Denny Vrandečić delivered a thought-provoking talk at a recent Informatics Departmental Seminar. You can now watch the full recording.

How do you see AI lowering barriers for knowledge sharing?

Right now, the English Wikipedia holds a very special role in the ecosystem of Wikipedias. Wikipedia exists in more than 300 languages, but when we talk about "Wikipedia", we often think about the English version. And when we talk about contributing to Wikipedia, we usually mean contributing to English Wikipedia, and so on.

Because English Wikipedia has the most visibility, the smaller-language versions often lose out on active contributors. People may feel that contributing to the English version has a bigger impact and reaches more people.

We are working on a project aimed at expressing Wikipedia's knowledge in a way that abstracts from concrete natural language. It's pure structure, and can be edited in any language. This can then be generated into natural language text for users to read in any language they prefer. This will help fill in the gaps in the current Wikipedia and lower the barrier for contribution.

I believe that we are smarter together as a collective than any one of us alone. You don't have to be smarter than Wikipedia, but if you know something that Wikipedia doesn't know yet, you can contribute in that space.

What has your experience been like as a Visiting Professor at King’s College London?

It's been very enjoyable. It's really pleasant to be around so many smart young people and to learn from them. I really enjoy the fact that King's has such a rich and diverse student body, which explains many interesting questions and projects on so many topics. 

If you could only take one piece of music, one tool, and one famous celeb (dead or alive) to a desert island, who or what would you take and why?

A single piece of music? Certainly not. It would get boring after a short time. There are a lot of songs that I really, really enjoy, but staying with just one piece of music… I don't know. However, taking an instrument is definitely something you can be creative with. It could be a computer, a guitar or a piano – one of those things so I can make my own music.

As for a famous celebrity, I'd choose someone who could be effective getting us both off the island alive. That would be very good. Someone I’d enjoy spending time with, though not a celebrity, would obviously be my wife. I think there are plenty of pleasant people in computer science to hang around with, but getting stuck with a single person for a long time isn’t fun. If I had to take a computer science book, I'd probably go for The art of Computer Programming by Donald Knuth. It’s a book I’ve started reading several times, and it would definitely take a lot of time to get through.

A tool I'd take is Wikipedia. I would definitely write about being stuck on some desert Island and I hope someone would eventually find it and rescue me. And if not, I’d still enjoy reading Wikipedia for fun and to learn.

Do you have any advice for young computer scientists looking to find success in an industry soon to be dominated by AI?

Something I’m seeing a lot in younger researchers is that they really dive deep into LLMs, deep learning, and model architectures. This often comes at the expense of understanding the foundational aspects of AI and computer science.

Do you need to be an expert in these foundational areas? Probably not. But I think it’s important to have at least an inkling of the basics. I'm not saying you should study things that bore you – that would be bad advice. When you’re studying something, I really, really hope it’s something that excites you, something you believe can help you grow and achieve what you want in life.

I had no clue where I wanted to go when I was your age. But one of the best pieces of advice I received back then was to try to figure out the one thing that only you can do. Maybe it's because of your cultural background, your interests, or the specific things you've studied, something that sets you apart. Figure out what that is, and lean into it.

Don’t just do what everyone else is doing, even if there are good reasons for following the crowd. If you have the capacity and opportunity to do something different, something uniquely yours, I’d recommend you go for it.

Oh, and have fun. I mean, that’s important too. Otherwise, what’s the point of all this?