Artificial intelligence models are widely used today, yet there is still limited understanding of how they reach their answers. At the University of Amsterdam, researcher Ana Lucic is developing methods to better understand the internal workings of AI systems and to make their outputs more interpretable. Her work aims to shed light on the so‑called “black box” nature of machine learning models and improve trust and transparency in AI.
When users receive AI‑generated responses, for example at the top of a Google search result, it is not always clear how those answers were produced or whether they are correct. Many AI models function as a “black box,” meaning that their internal decision‑making processes are hidden and difficult to interpret. This lack of transparency can undermine trust and make it challenging to assess why a model produced a specific result.
To address this issue, Ana Lucic, an assistant professor in artificial intelligence at the UvA and a MacGillavry Fellow, is developing interpretability methods that aim to open the black box and reveal what happens inside machine learning models. These methods strive to explain model predictions, but Lucic highlights a fundamental challenge: evaluating whether an explanation is correct. She explains that determining the validity of interpretability techniques remains an open problem in the research community.
In addition to her work on interpretability, Lucic is also extending her research to machine learning applications in weather prediction. During her time at Microsoft Research she was a lead developer of Aurora, a machine learning model for global weather forecasts. Traditional weather prediction models are both time‑consuming and computationally expensive, as they require continuous recalculation of complex physical equations. Lucic’s approach trains machine learning models on data generated by traditional systems, enabling much faster forecasts while still relying on the strength of conventional methods.
Lucic emphasises that improving interpretability is crucial for ensuring that AI systems are not only powerful but also reliable and understandable. She hopes to develop new tools that make AI behaviour clearer and ultimately enhance long‑term model accuracy. Her career has taken her between academia and industry, and her current position at the UvA allows her to collaborate with experts across diverse fields.
Lucic’s work highlights the broader scientific effort to make artificial intelligence systems more transparent and accountable, laying the groundwork for trustworthy deployment of AI technologies in research, industry and society.
Read also
