Dr David Watson is a Lecturer in the Department of Informatics who obtained his doctorate from the University of Oxford. His research focuses on Machine Learning and Causality.

What first attracted you to the field of Artificial Intelligence?

I was initially drawn to artificial intelligence by my fascination with human cognition, which is so complex and varied. That said, my academic trajectory has been a bit unusual - I originally started my studies in philosophy. I took a philosophy of computation course as an undergraduate. The class got me totally hooked.

Is there a scientist in history, or today, who is your biggest inspiration/role model? And why?

I suppose among all time scientists we are more or less obligated to state the obvious – Einstein was an unrivalled genius, who combined visionary creativity with mathematical rigour to achieve revolutionary results. However, to cite someone a bit closer to my own field, I would say that Judea Pearl’s work on causality has been a major inspiration. His writing is exceptionally clear, and the ideas he’s introduced – the causal hierarchy, do-calculus, structural causal models – are absolutely fundamental to contemporary practice.

Tell us about something you are working on at the moment - what is exciting about it?

I recently wrote a paper with some colleagues at the University of Bremen introducing something we call “adversarial random forests”. Basically, we propose a new approach to generative modelling – the same class of algorithms that powers ChatGPT, DALL-E2, etc. – specifically designed for unstructured, tabular data. Text and image data, by contrast, is highly structured.

I think the project is super exciting for a lot of reasons. Perhaps the most important is that the method is so fast and simple, practitioners with little to no machine learning expertise can easily use it on their datasets. This will hopefully bring generative modelling into new domains such as healthcare and economics, where deep learning approaches are often too complex and data-hungry for practical use.

Please give us an example of AI enhancing everyday life in 2023 that you particularly like.

I’m a big fan of text-to-image models such as DALL-E2 and stable diffusion. I’ve never been an especially gifted artist, much to my chagrin. But I absolutely love playing with these tools to create new images and bring random ideas to life. For instance, we used a DALL-E2-generated logo for our adversarial random forest software package.

What do you think is the biggest misconception people have about AI?

I’m very concerned about our tendency to anthropomorphise AI. I think that even when we “know” the model isn’t thinking or feeling things, we are still tempted into pretending that it is, especially with chatbots doing such a good job of mimicking human dialogue. This misconception can be ethically dangerous when models are deployed in high-risk settings such as healthcare and finance.

It’s important to remember that AI is just a tool, and the choice of how to use this technology is always up to humans. We can and should impose high standards for algorithmic fairness, accountability, and transparency.

Population Health Seminar with David Watson

At this event

Dr David Watson

Event details

Related departments