How AI and Quantum Could Help Fight Climate Change
Earth Day is the day to celebrate our blue planet. The day to remember that it’s the only home we’ve got.
During his very first days in office, President Joe Biden signed a flurry of climate change-related executive orders. He rejoined the Paris Climate Accord, pledged to double offshore wind-produced energy by 2030 and freeze new oil and gas leases on public lands. All in all, he’s committed himself to an ambitious goal.
Ambitious, yes, but realistic — especially with the help of cutting-edge science and technology. To make it happen though, academia and industry should join forces and change our established, traditional approach to the discovery of new materials. We should accelerate the rate of design of new advanced materials, crucial to create sustainable solutions to climate change.
The good news is, we already have the ingredients to make it happen. They are artificial intelligence and, soon, quantum computing.
Adieu to serendipity
Traditionally, we’ve been discovering new materials either by accident (think graphene) or using a lengthy and expensive trial-and-error process. As part of IBM’s Future of Climate initiative, IBM researchers have now successfully used AI to design new molecules for climate change-related applications much quicker that they would have with the traditional discovery methods.
“We’ve designed molecules that could lead to more efficient polymer membranes to filter off carbon dioxide better than currently used membranes in carbon capture technologies,” says Mathias Steiner of IBM Research Brazil, the lead scientist on the project.
That’s incredibly timely, too. The International Energy Agency (IEA) is forecasting a huge surge in CO2 emissions from energy later this year, when the pandemic finally starts easing off. While total energy emissions in 2020 will be a bit lower than the year before, CO2 emissions will increase by the second largest annual amount on record.
Typically, researchers rely on their knowledge and whatever they can find in published literature to design a molecule, hoping it will have the desired properties. Based on the initial design, they then follow many cycles of synthesis and testing of potential molecules until they create a satisfactory one.
The process often takes months, sometimes years, even with the help of computers to run advanced simulations. The most complex molecule we can simulate today is of the size of pentacene, with 22 electrons and 22 orbitals. Anything more complex, and computers stumble.
But the possibilities for molecular configurations are incredibly vast — there are more possible combinations for a new molecule than there are atoms in the universe. That propels the number of potential new materials to infinity. Equally vast is the ever-surging amount of data. In 2018 alone, about 450,000 new papers were published in the field of material science — impossible for any human to go through in a reasonable amount of time.
Enter artificial intelligence. Just five years ago, AI was mostly good at predicting characteristics of an existing material. Now, researchers are using it more and more to rapidly design brand-new materials with desired properties. “The application of AI to accelerated materials discovery is incredibly exciting and it will allow researchers to be far more efficient in their research,” says Stacey Gifford, a climate scientist at IBM Research.”As new technologies, like quantum computing, expand, the pace of discovery will only increase.”
From digital design to the lab
To design a new polymer for CO2 filtering membranes, Steiner and his team first had to outline the desired properties: permeability, chemical selectivity for specific gases, and durability. Next, an AI sifted through the past knowledge on polymer manufacturing — all the previous research tucked away in patents and publications. Then the researchers used predictive, so-called generative models to create a possible new molecule based on the existing data — a molecule that would make the polymer membrane more efficient in separating CO2.
The next step was to simulate this new molecule and the reactions interactions it should have with its neighbors on a high-performance computing cluster, to confirm that it performed as expected. In the future, a quantum computer could improve on these molecular simulations, but we are not there yet today.
Once everything is tip top with the design, the final step in molecular design is AI-driven lab tests to validate the predictions experimentally and create the actual molecules. This could be done using a tool like RoboRXN. Developed at IBM Research in Zurich, this ‘chemistry lab’ combines AI, a cloud computing platform, and robots to help researchers design and synthesize new molecules anywhere and at any time.
Steiner’s team hasn’t yet turned their digitally validated molecules into real ones, but other IBM researchers have done this last step for a different project. While not related to climate change, that study could help us make greener gadgets. IBM scientists used the same AI-boosted ‘accelerated discovery’ approach to create new molecules called photoacid generators (PAGs), important components of computer chips. The PAGs used today have recently come under enhanced scrutiny from global environmental regulators so the world is in need of more sustainable ones.
Sustainable hybrid cloud and AI
But material design isn’t the only way to help the climate. Another group of IBM researchers is working on making the company’s hybrid cloud more sustainable.
IBM is well-known for its hybrid cloud technology and OpenShift as the unified control plane on- and off- premises. A sustainable hybrid cloud enables companies to transparently assess the carbon footprint of their workloads, and reduce it if necessary. “To quantify and optimize the carbon footprint of cloud workloads, we are developing a carbon quantification and optimization method that attempts to make maximum use of renewable energy,” says the lead researcher on the project, Tamar Eilam. “IBM Research is also working on improving the overall efficiency of AI training by developing more efficient hardware.”
Another team, led by Shantanu Godbole at IBM research India, is using AI to help companies cut their carbon emissions associated with processes such as logistics, transportation, manufacturing, agriculture, and so on.
Meanwhile, IBM researchers led by Kommy Weldemariam are creating an AI to assess potential impacts of climate change on supply chain and infrastructure, from railroad lines to roads, bridges and tunnels. Dubbed the Climate Impact Modelling platform, the technology aims to improve regional climate modelling by bringing the model size down to about one kilometer. “At the moment, most climate models have a fairly low resolution, making it tricky to create accurate predictions,” says Weldemariam.
The researchers use physics-based and AI models to predict, assess and quantify the risks from extreme events — such as floods, wildfires or drought. The models can be integrated into enterprise processes, from the supply chain to the asset and infrastructure management, making it easier for companies to deal with a natural disaster.
While the ongoing research is promising, we are nowhere near the finish line. There is still a lot more to do to develop effective solutions to help our planet. And while Biden’s environmental goals are certainly ambitious, it’s almost certain that new technology will help us meet them.