Pushing the boundaries of AI knowledge.

It started with a challenge.

Imagine a world where new drugs are designed completely from scratch by artificial intelligence and machine learning methods, potentially producing therapies for diseases that are currently incurable.

In this world, a completely automated lab designs and executes experiments, using the data from the experiments to refine a hypothesis, on repeat, never tiring, never stopping. The fully automated lab performs the experiments in a more precise and rational way than a human, exploring things that go beyond human knowledge and human intuition saving costs and time.

This is a world envisioned by Professor Michael Bronstein, who is optimistic about the applications of artificial intelligence as a force for good, that can push the boundaries of scientific knowledge and impact.

In the ever-evolving landscape of artificial intelligence, few researchers have left as indelible a mark as Professor Bronstein.

With a profound dedication to exploring the frontiers of machine learning and pushing the boundaries of what is possible, Professor Bronstein's work has transformed the field, offering new perspectives, methodologies, and applications.

His pioneering research has garnered widespread recognition and established him as a luminary in the world of AI.

When asked how it all began, Professor Bronstein says, ‘My interest in AI goes back to an undergraduate project at university.’ 

This kind of problem is studied in the field of computer vision, which is part of artificial intelligence. To solve this problem, the young Bronstein used methods from the field of geometry and the combination of these disciplines excited him a lot.

So much so, after successfully creating the algorithm, he decided to do his PhD and continue research in this field.

He still works on the intersection between machine learning and geometry, using ideas such as symmetry and the invariants applied to machine learning problems and vice versa to apply machine learning to geometric objects such as graphs or molecules or  3-D shapes.

Appointed as the University’s DeepMind Professor of Artificial Intelligence in 2022, Michael Bronstein had exhibited an early fascination with mathematics and computer science.  

The driving force behind his scientific curiosity he says is, ‘Being able to understand in a rigorous way how these things work.’

He completed his undergraduate and PhD degrees in computer science at the Technion – Israel Institute of Technology, under the guidance of renowned scientist Ron Kimmel.

His formative years as a student, marked by an insatiable curiosity and a penchant for solving complex problems, set the stage for his illustrious career.

Professor Bronstein brings a truly global perspective, having completed his studies in Israel, he then worked in Switzerland, and the US and has spent the last five years in the UK.

During that time he has held appointments at Imperial College London, Stanford, MIT, and Harvard.

He is the recipient of the EPSRC Turing AI World Leading Research Fellowship in addition to many other prestigious grants and awards.

In addition to his academic career,  Professor Bronstein is a serial entrepreneur and the founder of multiple start-up companies, including Novafora, Invision (acquired by Intel in 2012), Videocites, and Fabula AI (acquired by Twitter in 2019).

Most recently Professor Bronstein has been furthering his research goals at the University of Oxford.

He says, ‘as a result it attracts probably the best students and the best colleagues.’ 

‘I find that the students here are top notch and they have unique academic backgrounds that might be difficult to find elsewhere.’

‘So for me in particular, it's important to have a combination of both mathematics and computer science. I've been very happy here.’ 

Professor Bronstein believes machine learning has the power to transform the way that biological sciences are being done.

One of the topics that he is pushing the boundaries of knowledge in, is the field of geometric machine learning and its application to molecular modelling.

‘Molecules are these tiny things that more or less everything around us is composed of. Well, they are geometric objects so we can model them mathematically as graphs, or sometimes as surfaces or manifolds.’

For drug development, often you have a therapeutic target, he explains, which is typically a protein. The challenge is to find a molecule that will chemically attach itself or bind, as chemists say, to this target.

There are many possible candidates with better or worse binding properties, and then of course, a prospective drug molecule should not, for example, be toxic.

You might need other properties, such as it must be soluble, so you can deliver it. If, for example, you need a molecule that can penetrate the cell or membrane, if it is an intracellular target, or maybe it needs to be able to reach the brain if the target is in the brain.

There are many different considerations that you need to satisfy. It is probably fair to say that some parts of drug design have been driven by intuition.

Chemists know more or less the typical structures and then they try to combine them, but the number of molecules is huge. Estimates vary, but it is probably close to the number of particles in the universe.

Professor Bronstein and his collaborators are already at the stage where they can generate new molecules completely from scratch with machine learning techniques, based on certain desired properties.

Think of generative artificial intelligence programs that generate images from natural language descriptions, called ‘prompts’ such as Midjourney or Stable Diffusion.

Now imagine if you could use that technology to design new drugs that are designed completely from scratch by artificial intelligence, well, that is already happening.

Can we yet say that we have artificial intelligence? 

Despite his seminal contributions to the field of research in artificial intelligence and machine learning, expanding on the topic Professor Bronstein says we should probably start by saying that:

Typical examples traditionally belived to be the hallmark of intelligence is the ability to play games like chess, visual perception, language, and creativity.

Since its very beginning, artificial intelligence tried to mimick these human capabilities in a machine. If you look at the history of the field over the past ten or  20 years, we have had such a progress where a computer can now solve some these tasks on par with humans, or even better.  

Chess was probably the first complex intelligent game to surrender to a computer, in the famous match was Garry Kasparov that was won by the AI developed by IBM in the 1990s.

More recently, a DeepMind algorithm beat the champion of the Chinese game, Go.

In visual perception tasks, we now have self-driving cars that at least statistically work significantly better than humans. In language, everybody has probably heard of or used ChatGPT: when you talk to it, it gives the perception of being human, and is probably already beyond the famous Turing Test.

For creativity, generative AI models can produce stunning pictures that even win artistic competitions.

Professor Bronstein however believes that despite this groundbreaking progress, we are not anywhere close to having 'artificial intelligence' yet.

The creation of true, or 'general' artificial intelligence is somehow a kind of moving target, like the line of the horizon. We believe that we have made huge progress and yet we do not have true artificial intelligence. So, it is probably more a semantic question of what is true artificial intelligence.

Professor Bronstein believes this might be achievable. It might be very different from human intelligence, he continues. We do not necessarily need to imitate ourselves because in some tasks, computers are already way better than humans. In some others we do not even know how to approach the problems yet.

The scale and speed of the impact of AI could be comparable to the industrial revolution. 

Professor Bronstein believes, however, it is much faster, and the impact is much bigger, whether it is replacing us or augmenting us.

‘There are concerns about the impact of AI on society’, he says, ‘including scenarios of AI going rogue and killing us all. I personally believe these scenarios are unlikely.’

‘We see a glimpse of it already now, with tools like ChatGPT that many people already employ as a writing assistant. I think AI should augment human capabilities and human intelligence, acting as a tool rather than our foe.’

Professor Bronstein is actively engaged in shaping the future of AI research and advocates for responsible AI.

As AI becomes increasingly integral to our lives, Professor Bronstein is committed to ensuring that its development is in harmony with societal values and respects fundamental human rights.

He is also a proponent of interdisciplinary collaboration, believing that the fusion of knowledge from various domains can drive AI research to new heights.

His current projects explore the intersection of geometry, deep learning, and data science, with the aim of addressing real-world challenges in areas such as healthcare.

Ultimately, Professor Bronstein says, it will be up to us, as a society, whether we want AI to replace us, maybe we want to free ourselves from repetitive and more mechanical work, as this would enable us to do more complex, creative work.

He foresees that simple task requiring a mild cognitive effort, say under one second of thought, are more likely to be replaced, but other more complex and creative mental processes probably not, and maybe we do not want them to be replaced.

As the AI field continues to evolve, Professor Bronstein's influence endures through the many researchers and students he has inspired and mentored.

His work has not only led to cutting-edge algorithms and technologies but has also paved the way for a new understanding of how AI and geometry intersect.

With a vision for responsible AI and an unwavering commitment to excellence, Professor  Bronstein's legacy is one that will continue to guide and inspire generations of AI researchers to come.