Professor of Computer Science Kai Puolamäki uses AI to investigate new ways of modelling atmospheric particles. His research focuses on how to make machine learning models understandable, how to assess their uncertainties and how researchers can use them to discover new knowledge.
You have likely used AI, whether in your free time or at work. But do you really know how it works? There is no need to feel embarrassed if you don’t.
“No one fully understands how complex, self-learning AI models actually work,” says Professor of Computer Science and Athmospheric Sciences
“The software involved is so extensive and complex that even the people who develop it don’t know everything about it.”
Even so, Puolamäki is working to uncover what is happening inside these models for a very practical reason.
“Picture a study where AI analyses data on the researcher’s behalf. If the researcher doesn’t understand what’s happening, it’s difficult for them to judge how reliable the AI’s conclusions are.”
This is why we must understand AI better to use it effectively.
From particle physics to machine learning
Puolamäki has long been interested in the collaboration between humans and machines. After completing his doctoral thesis in particle physics at the University of Helsinki, he moved on to the Helsinki University of Technology, now part of Aalto University, where he began studying data mining, machine learning and human–computer interaction.
“I think we need better ways to visualize the data generated by algorithms. It should be something just as easy for people to grasp as information presented as points on a coordinate grid,” he says.
Using AI to advance aerosol research
At the University of Helsinki, Puolamäki collaborates extensively with atmospheric scientists, developing AI-based tools to improve our understanding of how aerosols, or tiny particles in the air, form at the molecular level.
While the bonds involved can be calculated using quantum chemistry, doing so is time-consuming; machine learning can help accelerate the process. Puolamäki has developed approaches where AI deduces aerosol molecule formation by comparing them with the structures of similar, previously identified molecules. He has also developed ways to understand the workings of these machine learning models and assess their uncertainties.
Will AI take over from coders?
It seems clear that we have entered a new era of AI. Concerns have been raised that it could take jobs away from professionals, including coders. Puolamäki does not view this as a serious threat.
“Simple coding work has long been outsourced beyond Finland’s borders. And even if AI ends up writing code in the future, we’ll still need people who can monitor and guide it.”
Still, the future remains hard to predict. Puolamäki points to social media algorithms as one example, noting how they have fuelled the growth of filter bubbles and deepened societal polarization.
“Twenty years ago I still idealistically believed online discussion platforms would help democratise and improve the quality of public debate. Things didn't quite work out that way.”