AI4Science AMA (Ask Us Anything) featuring Chris Bishop, Bonnie Kruft, and Max Welling

Microsoft Research’s Bonnie Kruft, Max Welling, and Chris Bishop have reviewed your AI4Science questions, and they’re here with answers! Read on to see what they have to say. 

Why is AI4S considered the next paradigm and why is that important for the scientific community?

Max Welling: After experiment, theory, computer simulations, and big data, we are now seeing that computer simulations can be used as pseudo-data on which ML models can be trained. This can help to accelerate simulations next time around.

Chris Bishop: We believe there will be a transformational change over the next few years in the way fundamental processes in the natural sciences are modelled, driven by machine learning. The key difference compared to “conventional” machine learning is the use of synthetic data generated implicitly or explicitly by solving the fundamental equations of science, leading to simulation software that learns from experience and that is orders of magnitude faster than current approaches.

Bonnie Kruft: The fundamental equations of chemistry, physics, and biology are well known, however their solution is computationally extremely expensive. Over the past few years, we’ve seen a new way to use deep learning as tool to address this challenge. We think this will lead to an era of intellectually deep research in the scientific community and has a huge potential to impact the real world.

What aspect of AI4Science most excites you and why? 

MW: The prospect of accelerating scientific discovery, but also the fantastic opportunity to solve some of society's most urgent challenges, such as diseases, resistant bacteria, the need to accomplish the energy transition, and climate change.

BK: The potential to make a positive impact on the world. We have an opportunity to use AI to address some of the world’s most pressing challenges—from better medicines to treat disease, new substrates for capturing carbon, new sustainable materials and many more. 

What ways do you think AI4Science will most affect society? What will be the most impactful, and when do you think we would see it in action? 

MW: Modeling the climate and predicting the amount of global warming we will see at the end of the century given certain carbon emissions. Ab initio in silico design of new materials that can capture carbon, improve catalytic processes to produce hydrogen, synthetic fuels and fertilizers. We will see major progress over the next few years already I think.

CB: A deeper and more predictive understanding of biology could have important implications for healthcare including the discovery of new drugs and therapies.

BK: Progress in the way we model, predict and respond to climate change.

What is Microsoft currently working on with AI4Science? 

BK: We have multiple projects that span the impact areas of climate/environmental sustainability, drug discovery and life sciences, chemical reactions and large-scale physics simulations.

MW: Drug discovery, materials discovery, computational catalysis, PDE surrogates, accelerating DFT calculations, accelerating molecular dynamics simulations, whole cell modeling and many more things!

How can I get involved with AI4Science and Microsoft? What career opportunities are available and where?

MW: Please be on the lookout for new jobs or internships posted on our AI4Science website under career opportunities.

I’m a student, what should I major in if I want to work in your lab one day? 

MW: Either Natural Sciences, e.g. chemistry, biology or condensed matter physics with as many courses in coding and ML as you can take, or a CS curriculum with lots of AI and ML and with as many physics/chemistry/biology courses you can find. Unfortunately, there do not yet seem to be a lot of programs in AI4Science, but they may come one day.

What kind of partnerships does AI4Science participate in with the scientific community?

MW: We partner with universities, such as University of Amsterdam and Vrije Universiteit Amsterdam, Cambridge University and EPFL, to name a few by funding some PhD positions. We also collaborate with external universities on a more ad hoc basis when this is appropriate.

CB: Our partnerships with industry are key to seeing this new paradigm have real-world impact. We have a deep partnership with Novartis in drug discovery, and are currently forming new partnerships across multiple sectors in materials discovery and large-scale physics simulation.

BK: External partnerships are core to our strategy in AI4Science—we collaborate with both industry and academia to augment domain expertise and facilitate real world impact.

Where is the AI4Science lab located? 

BK: We are global team with locations in Cambridge UK, Amsterdam, Berlin, Beijing and Redmond.

What are the main priorities for AI4Science? 

CB: We seek to make fundamental and long-lasting advances in the use of machine learning for simulating physical, chemical and biological systems that also have transformational real-world impact. You can read more about our priorities on our FAQ page.

MW: We have a number of core projects around some big challenges: drug discovery, materials discovery, computational catalysis, PDE surrogates, accelerating DFT calculations, accelerating molecular dynamics simulations, whole cell modeling. 

BK: Advancing scientific research through machine learning; enabling Microsoft to create new technologies and services that empower every person and every organization on the planet to achieve more; to make an impact on the world’s most pressing challenges.

What type of background do you need to work on the team?

MW: ML/deep learning with a keen interest in science, or an expert in some domain like chemistry with coding experience and with a keen interest to disrupt these fields using ML.

I have a background in research and engineering? Are there opportunities to get involved without experience in the natural sciences?

MW: Certainly. If you are a stellar software engineer or if you are an ML researcher and you have a keen interest in science, then you can apply. I would recommend doing some projects that involve scientific data to support your case.

What are opportunities for PhD students in AI4Science? Are internships or fellowships available?

BK: Yes, we have opportunities for PhD students in many of our projects. You can find more by visiting our website.

Does AI4Science have any internships available? 

MW: Absolutely! Many in fact. Please check out the MSR AI4Science webpage under the career opportunities tab.

Do you have to be a researcher to work on the AI4Science team?

BK: We are also looking for engineers and program managers for AI4Science.

How do you see AI careers functioning in the future? 

MW: Either in academia or in industry, each with their own merits. In industry we tend to tackle large problems with interdisciplinary teams of researchers, which require significant resources to make progress.

What are some key business-use cases where AI for digital transformation projects can be used?

MW: PDEs are used in many places. For instance, to model airflow around airplanes or to model gas flow in porous media. DFT and MD calculations are also ubiquitously used in industry, such as the chemical industry to design new catalysts and materials. And we should not forget the pharmaceutical industry who are using ML to design new drugs and vaccines. 

BK: The pharmaceutical industry uses machine learning to improve the way that medicines are designed, made, tested and analyzed. Chemical and energy companies use molecular simulation to design new catalysts and substrates.

To learn more about Microsoft Research AI4Science, please visit our website: aka.ms/AI4Science