Prof Jonathan Shock‘What we need are people who can think deeply about complex questions, people who are technologically savvy and understand the value of the scientific method, and people who have a desire to make a difference. I have seen time and again the evolution of students as they come from undergraduate, into graduate studies and become well-rounded scientists, and I am ever proud of the work I see them do. We need more people like this, and I cannot understate the amazing privilege of being able to study the workings of the universe at all its scales, as a career.’

These are the words of Dr Jonathan Shock, Associate Professor in the Department of Mathematics and Applied Mathematics at the University of Cape Town (UCT) and interim director of the new UCT AI initiative.

‘My broad research questions relate to how we can use machine learning to better understand the way organisms survive in a complex world, and conversely how we can improve machine learning systems by studying simple and complex organisms.’

He explains in more detail: ‘It includes research in reinforcement learning – a field which has evolved from psychology, control theory and machine learning. It is the study of how computers can learn to perform tasks using trial and error learning, receiving feedback signals from the environment in the form of rewards.’

He continues: ‘My specific research in this area focuses on multi-agent systems, offline learning (where the computer doesn’t have direct access to the environment), and meta-learning (learning how to learn). I am also currently working on active inference, which takes a Bayesian perspective on the brain as an inference engine, with only indirect access to the world through our observations, and evolutionarily determined preferred states that promote survival. On this topic, I work closely with Professor Mark Solms and a team investigating what are the factors that lead to affective consciousness.’

The Shocklab ecosystem
Additionally, Dr Shock also uses machine learning to study neuroimaging, ‘including for the purpose to understand the aging process and identify brain regions that may be particularly important for detecting pathological aging signatures.’

His work with statistical methods to understand the brain initially sparked his interest in machine learning and lead him to utilise the Coursera platform to develop a range of skills including deep learning, probabilistic graphical models, and natural language processing. At the heart of his projects is the Shocklab ecosystem comprising extremely talented graduate students who collaborate across numerous departments at UCT and beyond. This includes physics, statistics, computer science, psychiatry, psychology, computational chemistry, education and engineering.

‘Many of these students have also worked very closely with Instadeep, an industry partner, working in both pure research and applied areas of machine learning, and who have been instrumental in mentoring and guiding these students to publish at the top machine learning conferences in the world. Instadeep is very rare as an industry partner in both their desire to take on and mentor students, as well as the opportunity they afford to do essentially pure academic research which then leads into their applied groups.’

The future for AI, machine learning and pan-disciplinary approach
‘I believe we are at a crossroads, not just for the field, but potentially for humanity. There is great excitement, but also well-founded fear, about what we may have created,’ Dr Shock says.

‘It is vital that we continue with caution but with an openness to working together and for sharing insights. We must get out of our silos and work as a global community so that we can leverage the power of AI for good and mitigate the clear and present dangers as much as we possibly can. At the same time, it is vital that Africa is not left behind as it has been during previous industrial revolutions. The teams of researchers around the continent must speak up and show the world the richness of what we have to offer, intellectually, culturally and philosophically. The 2024 IndabaX (the local version of the Deep Learning Indaba – the continent’s largest machine learning conference) was held in over 40 African countries. This shows that there is passion, expertise and a drive to be heard.’

He adds that ‘we have a long way to go to educate those around us in what AI is. There is a belief by many that AI is ChatGPT, but of course it entails far more than this and includes applications as diverse as climate modelling, drug discovery, radiological examining, species identification, crop management and so much more.’

‘In this context,’ Dr Shock says, interdisciplinary research ‘allows us to leverage expertise across fields, often revealing how ideas from one area can advance another with minimal adaptation. While working within specialized domains can produce amazing results, there’s immense value in learning from other fields when we invest the time to understand one another’s perspectives, methodologies and importantly, languages. We can accelerate our own research by both learning about others’ work and sharing our own expertise as widely as possible.’

In true pan-disciplinary spirit, Dr Shock, who is an adjunct professor at the Institut National de la Recherche Scientifique (INRS) in Montreal (Canada), works with a INRS group on how to use machine learning for material discovery. ‘We are investigating ways to accelerate the discovery of materials useful for direct air carbon capture to mitigate climate change. Traditional approaches use time and compute-intensive DFT (Density Functional Theory) to predict material properties. Our work with graph neural networks trained on DFT data has already achieved an order of magnitude speedup in calculating the ground state of these important materials.’

Turning to the role of NITheCS in this context, Dr Shock as ‘As Associate of NITheCS I’ve seen firsthand how vital such institutes are for strengthening local talent, facilitating interdisciplinary teams, and nurturing the next generation of researchers. NITheCS plays a crucial role in showcasing South African research globally through platforms like YouTube, and initiatives like SATACS that are moving to address historical resource imbalances in our country.’

Background
Dr Shock talks briefly about his background, saying his father, a chemical engineer, showed him what it meant to be a scientist and guided his early development. ‘My mother, an artist, provided a complementary perspective that, I think, has enabled my broad approach to research. At high school I was extremely lucky to have a passionate physics teacher, Thomas Garnier, who fostered my growing interest in the subject and helped me to overcome my often very messy approach to problem solving.’

He then studied physics at Bristol university, became fascinated with string theory, and earned a PhD in the subject at the University of Southampton. ‘Finding a postdoc position was not easy, but the Institute for Theoretical Physics in Beijing gave me an opportunity to become a researcher there, and I had two incredible years in China. I grew in confidence in terms of my research, and discovered a new culture. Two more postdocs later I came to UCT, found myself in front of 400 bright-eyed firsts year students, and fell in love with teaching. UCT has given me a great deal of freedom in the work that I do, and I have been very lucky to have massive support from the faculty of Science and the Department of Maths and Applied Maths.’

And importantly…
‘I love to teach – but when not working, I can be found in the kitchen, with my camera, or on the rowing machines. My one piece of advice to others is to be curious and keep asking questions.’