Ida Rahu: “AI should never replace the scientist”
When Ida Rahu was 14, she sat in a chemistry classroom and made a declaration that would shape her life. “I literally said to the person next to me, ‘See, I will become a chemist.’”
That early conviction carried her through years of study in organic chemistry and, eventually, to a PhD. But the deeper she went into laboratory work, the more restless she became. “There’s a lot of creativity in the lab. You can synthesise all kinds of new molecules but I realised that that wasn’t where my passion lay. What truly excited me was solving problems on the computer.”
From Excel to data science
The turning point came when she found herself drowning in experimental data. “My supervisor said, ‘Here is an Excel, use this.’ And I was like, no, no, no – there has to be an easier way,” she recalls. “So I started taking programming courses, and while doing that, I realised this is exactly what I want to do.”
The same year she defended her PhD – during the height of the COVID-19 pandemic – the University of Tartu launched a new Master’s programme in data science. After completing her PhD, Rahu enrolled and discovered a new way of thinking about chemistry – through computation.
That unusual combination – chemistry, coding, and environmental questions – soon came to define her career. “Everything just aligned,” she says. She joined Stockholm University as a postdoctoral researcher in Professor Anneli Kruve’s group before applying for a position as Assistant Professor at the Department of Environmental Science. “I would never have applied if it didn’t say ‘data science tools.’ This position was literally what I was looking for.”
Genes, environment, and a dog named Lumi
The ultimate aim of Rahu’s research is to understand how genes and environmental factors interact to influence health. The idea is rooted in something personal: her dog, Lumi (“snow” in Estonian), a Rough Collie with a genetic disorder. “She’s very cute,” Rahu smiles. “But she is much more sensitive to some medications than other dogs. I almost lost her because of that.”
That experience shaped her scientific vision. “In toxicity studies, we often multiply one threshold value and say: this is safe. But individuals have completely different susceptibility. I don’t think this is the right approach.” By combining chemistry, genetics, and machine learning, Rahu hopes to move toward more individualised risk assessments. “That’s my goal. If this perspective could help inform future policy decisions, I’d be very happy.”
From toxicity to persistence
Rahu’s machine learning models are already making a difference. In a recent study of textile samples, more than 10,000 chemical features were detected – far too many for any researcher to identify manually. “It’s almost impossible,” she says. “But with my pre-trained model predicting biochemical activity, I narrowed this down to fewer than 1,000 compounds worth examining more closely.”
But toxicity is only one part of the story. For Rahu, understanding chemical safety also means looking at persistence – how long substances remain in the environment and how they move through different systems. “We are trying to identify the structural features that make certain chemicals more persistent in various conditions, for example, in rivers or in drinking water treatment plants,” she explains.
By extending her models beyond toxicity, Rahu is broadening her scope to capture a more complete picture of chemical behaviour. “A compound that isn’t highly toxic can still cause long-term harm if it doesn’t degrade,” she says. Her aim is to combine these aspects to better predict environmental and health impacts.
But as her models grow more complex and ambitious, so does her awareness of their limits.
The Black Box problem
“Some models work like a Black Box – you put your data in and get an answer out, but you can’t really see how it works,” she says. “For me, that’s a problem. If I want to use these models for real-life applications, I need to know whether it makes chemical sense.”
Her team is now working on Explainable AI, methods to “deconstruct the model” and make its reasoning visible. “Otherwise, users won’t trust it. And I see it as an opportunity – we might even discover new mechanisms, new explanations, from what the model highlights.”
Critical thinking and open science
During the interview, Rahu never refers to her work as AI. Yet she acknowledges that the label often follows her around – in project proposals, collaborations, and casual conversations alike. “People often say, ‘Oh, so you work with AI,’” she laughs. “And in a way, yes – but I prefer to be precise.”
For her, it’s not about creating intelligence but about using data-driven methods to deepen scientific understanding. “The goal isn’t to build an intelligent system from scratch,” she explains. “It is to make sense of complexity – to uncover patterns that help us understand chemical behaviour and environmental processes a little better.”
That careful distinction reflects a broader philosophy that runs through her work. Rahu values curiosity and innovation, but she also reminds herself that she has to follow scientific reasoning. “The most important thing is to stay grounded as a scientist,” she says. “If something doesn’t make sense in your head, it’s probably too good to be true.”
Transparency is another principle she holds firmly. “I have always published all my code,” she notes. “I don’t like this gatekeeping. If people can’t see the data or the code, it becomes very easy to manipulate results.”
Her critical mindset extends to teaching as well. While she encourages students to experiment with new tools, she worries about overreliance on generative models. “What I’ve seen is that students without programming experience put too much trust in them,” she says. “They miss the opportunity to develop real coding skills and often can’t judge whether the output is even correct code. For teaching and learning, I find that problematic.”
Teacher and mentor
What gets her out of bed in the morning? “It’s that curiosity that drives almost every researcher – the need to understand, to figure things out. Every day is different, but the goal is always the same: to know a bit more than you did yesterday.”
Just as important is teaching. “I love my discussions with master’s students. When I prepare a course, I feel like I can change the world a little. My teachers all changed me, even the ones I disagreed with. They gave me something to push against.”
Her dog, Lumi, still ensures she gets fresh air every day, a reminder that science begins in lived experience. And her research continues to be guided by the same conviction she had at 14 – that chemistry matters, but now powered by the tools of data science. “AI should never replace the scientist. But it can make our work more efficient, and maybe help us see things we couldn’t before.”
Interested in applying data science tools to environmental science? Explore the course we offer with Ida Rahu
Last updated: October 29, 2025
Source: Department of Environmental Science