In conversation: Professor Lars Kotthoff
Voiceover: The University of St Andrews. Making waves since 1413.
Question (text over screen): How did you get into the field of Artificial Intelligence?
I studied Artificial Intelligence way back as an undergraduate and as many students I was sitting there during the lecture, and I was listening to the professor going on about artificial intelligence and how you do this and how you do that. And I remember sitting there and thinking, why are we spending so much time on this? This is easy stuff. This is trivial to do.
And then one day, I actually had to implement this myself and teach a computer how to do these trivial things that we as humans do all the time. And I realised that, wow, this is actually really, really difficult and this is when I thought, well, hey, this is maybe something interesting to get into more and to study more and then also eventually get into research because I wanted to really push the field ahead. But really my motivation for getting into artificial intelligence is because this is hard and I like solving hard problems.
Question (text over screen): What excites you most about taking on this new role as Chair in AI?
The thing that excites me most, and that I want to achieve here, is to facilitate more interdisciplinary collaborations that involve AI. And this is something that I've been doing in my previous roles as well. But it's really because AI is really everywhere.
Regardless of where you look, there will be some AI lurking around the corner or in the background, whether you realise it or not. And every discipline throughout the University will benefit by interacting more with AI and by using more AI for research and for teaching, and really to improve the institution as a whole. And this is something that I want to facilitate by talking to people and by figuring out how to use AI responsibly and ethically and in the right way in the particular domain that they're working in.
Question (text over screen): Where do you see the biggest opportunities for AI research in the next five years?
I think the biggest challenge going forward is really going to be to get away from the current mindset of scaling up AI systems, and in particular, I'm thinking of large language models and similar generative AI systems here, where the current approach is to essentially build them bigger and bigger and add in more and more data, and then hope that what comes out is something that will also work better in practice.
We are already starting to see limitations of this approach, and what's really missing is the fundamental understanding of how these systems work, why they work better, why they work so much better when we scale them up, why they do not work if we don't reach a particular scale, and how we can essentially make them work better in practice, even at smaller scales.
And I think I believe that a lot of this work will involve AI research that has been done in many, many decades past, in particular, lots of approaches about doing symbolic AI. So really teaching computer systems to reason with symbols rather than the current statistical approaches that include large language models, where essentially, we're identifying correlations on a very, very large scale and then leveraging these correlations to predict what's going to happen in the future.
What is the next word going to be in the answer that ChatGPT generates? I believe that a lot of what has been going on in AI research, up to the point where large language models were developed, has a lot of potential to make these and similar systems better.
Question (text over screen): This position is funded by an endowment. How does that support shape your vision for the role?
I take my main role to be really to facilitate the adoption of AI across the University and to facilitate its use in many other disciplines, and the support provided by the endowment is, I believe, what's really needed for this, because this enables me to focus on things that will not have an immediate return, but are really more strategic investments into the future of the institution and also into the future of research in particular areas where the use of AI might enable the human researchers to go that much further and to do that much more than they could do before, because they can now essentially give a lot of the work that used to be done by humans in the past to an AI system. And of course, there might be potential issues with that, because we still have to check that the AI system is actually doing the work properly. But at the very least, it might make researchers much more productive in particular tasks. And this will enable us as humans to do more interesting things. This is something that I would like to shape and that I would like to support through the support given to me through the endowment.
Question (text over screen): What will the creation of the Digital Nexus building mean for St Andrews?
I think that having a building where people who are interested in the AI space and in the digital space can come together and work together, is really going to enable us to go to the next level with respect to research, because physical presence, being able to be in the same space and meet there and work there, and bumping into your colleagues when you walk down the corridor is really something that enables much, much better collaborations than what we can do if we are based in different physical locations.
This is something that I hope the Digital Nexus building will really facilitate and really allow us to do better at St Andrews and then also facilitate more interdisciplinary collaborations, more collaborations within and without of Computer Science.
