Agenda

Interview to Prof. Matteo Bottai, visiting professor at DAIS thanks to Department of Excellence support. 
1. What motivated your visiting period at DAIS?
Sometime ago, your statistical faculty contacted me with an offer to spend time at your department as visiting professor. The potential for collaboration and joint research made it an excellent opportunity for me and for the Karolinska Institutet, my home institution. The level of research and depth of the collaborative relationships that have been established during my stay far surpass any expectations and hopes. As I approach the end of my visit, I wish to express my gratitude to my hosts for their dedication that is hard to beat in organizing structured activities. 
2. What are the main methodological challenges your research aims to address?
My research in theoretical and methodological statistics has grown over the years in several directions. During my stay, we focused on developing statistical methods to assess the probability of observing unusual values of variables of scientific interest, such as massive rainfall, high atmospheric carbon dioxide concentrations, high blood pressure, and short children’s height. Among the multiple methodological challenges, a key one is the need for statistical methods tailored to the specific practical problems at hand, as opposed to popular, readily available, off-the-shelf methods.
3. Why is it important to move beyond average-based models when studying environmental and health data?
While assessing averages of variables is useful in many practical problems, it may be inadequate in many others. For example, in clinical practice knowing that that a patient’s blood pressure is higher than the population average is generally insufficient, as values naturally ranges among individuals and not every healthy person is expected exactly to measure at the average. To make informed clinical decisions, one must know whether a measure is unexpectedly high, not just above average. Similarly, unexpected high carbon dioxide levels, rainfall, or soil pollutants concentrations, are of greater scientific and policy making interest than averages. 
4. How can these methods support environmental and public health applications?
Decision making processes are complex and suitable statistical methods can greatly aid them. It is essential to enable analysists and policy makers to quantify probabilities of unusual air pollution or health outcomes and identifying vulnerable population or areas.
5. How do statistical inference and AI complement each other?
Large language models are useful tools for scientists, helping structure complex languages, like English or mathematics and accelerating the process of scientific communication. However, they remain computer software that simply recycle words but do not know or feel like humans do. One can argue that is also what humans do, but whether that is philosophically true or not is irrelevant. Language itself cannot fully express the structure of the world, because words all short of conveying emotional or sensory experiences. In that respect, art and music may be better suited. The connection to the physical world is what humans must seek and explore.
6. What are the main opportunities and challenges in combining modern statistical models with AI methods?
The rapid growth of data availability and technology creates great opportunities to advance science and understanding of the world. Historically, advances in technology have always driven knowledge, from the printing press to modern digital tools. Today, large language models are promising, they also bring challenges and risks. In my field, I see an increase in analytical simplification and computer automation, which in my opinion is detrimental and should be closely monitored.