How did I embark on a transformative journey into the realm of Responsible AI? In this article, I briefly outline the story of how I transitioned from an researcher in Climate Science into Data Science in industry, and then specialised on the responsible and ethical use of AI.

From Climate Science into Data Science

Following the successful completion of both Bachelors and Masters degrees in Environmental Sciences at ETH Zurich, with a specialized focus on Climate Science, I decided to pursue a PhD degree in the same domain. I examined the possible biases that emerge from combining different global climate models, due to their shared (atmospheric, land, or oceanic) components. With my collaborators, I identified a novel combinatorial optimisation methodology for forecasting future climate, which resulted in several publications. It was during my PhD that I transitioned from utilising MATLAB as my primary programming tool to Python, became comfortable with working with server-based environments, and handling large volumes of complex spatio-temporal data.

I thoroughly enjoyed my PhD at the University of New South Wales (UNSW) in Sydney, which is also when I got exposed to the field of Machine Learning. My passion for this field got stronger as I attended several Data Science Meetups, took several specialised courses on coursera, and was lucky enough to be able to attend lectures at UNSW by Trevor Hastie (one of the authors of “An Introduction to Statistical Learning”, who was on sabbatical in Sydney that year). I was immediately fascinated by the field of Data Science and its broad applicability in various domains.

After completing my doctorate degree and moving back to Switzerland, I knew that I wanted to apply my acquired skillset in industry. I came across the job opening for a Data Scientist role at Thomson Reuters Labs, which prominently emphasized the vast amount of high-quality data available in the job ad. Knowing the importance of labelled data in the context of Data Science, and being intrigued by the application of AI/ML in the legal, tax, and news domains, I was drawn to apply for the position. I ended up getting the job, which allowed me to build upon my foundations acquired during my PhD, and gain exposure to the domain of Natural Language Processing (NLP).

Over the course of several years, I worked as a Data Scientist, and subsequently Senior Data Scientist. I underwent substantial professional growth and learned a lot of important skills on the job, including interacting with non-technical product teams to define the project scope, working with design and engineering teams throughout the model development lifecycle, as well as innovative methodologies for evaluating NLP models leveraging human feedback.

From Data Science into AI Ethics

It was during my time at Thomson Reuters Labs that I first gained exposure to the growing field of Explainable AI (XAI). Engaging in an entity extraction project intended to be used as part of a legal editorial workflow, I soon recognized that complete automation was unfeasible, and the model’s role would primarily be in supporting human annotators to optimize their efficiency. This realization prompted crucial questions concerning the model’s efficacy in instilling trust among annotators and whether it genuinely expedited their tasks.

I was able to dive deeper on this topic by contributing to a publication on Explainability in the context of legal document summarization. This experience nurtured a profound interest in the ethical and responsible aspects of AI development and I was invited to contribute to Thomson Reuters’ first set of Data and AI Ethics Principles (has since been updated). I got exposure to other important aspects around the responsible design, development, and deployment of AI solutions, including fairness, privacy-considerations, as well as human oversight. I had the opportunity to co-lead a stream of the research program focussed on Human-Centered AI affording me the opportunity to propose novel research directions and oversee ongoing explorations within this domain. I was able to showcase my knowledge and insights through presentations at numerous internal and external events.

From AI Ethics into Model Governance

For me, this was just the beginning. After the AI Ethics Principles have been released publicly, I always wondered “What’s next?” and “What do those principles mean in practice?”. Through my mentor I learned about the creation of a central Data and Model Governance team at Thomson Reuters, designed to operationalize the aforementioned principles. Driven by my desire to contribute meaningfully to this transformative endeavor, I applied for an opening and was given the opportunity to join the team as a manager in AI Ethics, later transitioning into Model Stewardship. My team is now actively supporting model developing as well as consuming teams across the organisation with the adherence to Model Governance and AI Ethics expectations, turning my desire of making the principles actionable into reality.

In the future, I am hoping to share key learnings from my experience in Model Governance through an informative post.

Banner photo from USGS on Unsplash