AI an administrative support for the doctoral student
For KI doctoral student Eric Åhlberg, AI and Microsoft Copilot (formerly Bing Chat Enterprise) offer tools for statistical analysis work and for managing certain administration. His long-term goal is for AI to become a validated tool for high-quality research results.
How did you first become aware of AI as a tool in your work?
"I'm researching the immune system's reaction to COVID-19 and influenza infection. By investigating immune responses throughout the course of the disease, we want to find early biomarkers to understand the development of severe disease," says Eric Åhlberg, doctoral student at the Department of Medicine, Solna, Karolinska Institutet. "My introduction to AI took place in connection with the planning of my PhD project. My supervisor and I realized that the amount of available research results was so large that it was difficult to capitalize on the data with traditional statistical methods."
"I think it's positive that Microsoft Copilot (formerly Bing Chat Enterprise) is offered to all employees at KI. Microsoft Copilot automatically cites and is based on a powerful language model. I've had mostly good experiences, but there are also limitations such as the number of interactions per day and text lengths accepted."
How do you use AI solutions today?
"I've just started implementing AI and I'm learning how to use it responsibly in my research. It is important to gradually evaluate the performance of AI, in parallel with traditional methods, in order to validate AI-generated outcomes. I use language models such as Microsoft Copilot and ChatGPT for R programming and R packages to apply machine learning algorithms to my research data."
What are the benefits of AI in research?
"At the beginning of my doctoral studies, I was a beginner at programming and had to devote a lot of time to this. ChatGPT and Microsoft Copilot helped me understand and improve code from open-source forums, which was an effective educational tool."
"Smart chatbots make it easier, among other things, to write program code for statistical analyses and data management. It benefits researchers with programming skills but who are not yet programming experts, by improving and refining their code. However, it's important to understand the code's strengths and weaknesses."
"AI also frees up time for researchers to focus on their main task: research. This is done by assisting in administrative tasks outside of the research itself, such as creating presentations and managing documents."
"However, you shouldn't expect miracles on the first question to the AI tool, or prompt. In my experience with AI-based chatbots in the design of presentations, for example, is that it can take many interactions before the results are useful."
What are the challenges?
"There are several risks associated with applying AI in research. When you blindly trust a chat model's recommendation for statistical calculations, long program codes, or article summaries, sooner or later you will step on the wrong foot. Many of us have experienced that language models tend to give an answer with varying degrees of truth to a question, rather than answering "I don't know". AI does not deprive us of critical thinking, but on the contrary, it increases the importance of it."
"Another risk is that researchers input sensitive information that is then used to train the algorithm, information that may become available to individuals for whom it was not originally intended."
What opportunities do you see with AI over time?
"My hope is that AI will increase my ability to achieve research results of high scientific quality. This can be done in a direct way, for example by brainstorming ideas for statistical methods and codes for programming. It can also be done in an indirect way by reducing the time that researchers need to spend on other tasks," says Eric Åhlberg.