Alán Aspuru-Guzik is using artificial intelligence, robots and even quantum computing to create the new materials we need to deal with climate change.

When Alán Aspuru-Guzik, a Toronto-born chemist who was born in Mexico City, was studying climate change models, his eyes were drawn to error bars, which show the surrounding area of ​​any given forecast The range of uncertainty. "As scientists," he said, "we have a responsibility to consider the worst-case scenario." If climate change goes as expected, humans may have decades or so to develop materials that do not yet exist: enabling us to be fast and cheap. To capture carbon molecules, and batteries made of substances other than lithium, an expensive and difficult-to-mine metal used to store global renewable energy supplies. 

What if the situation becomes worse than we expected? The demand for new materials will range from urgent to extremely urgent to terrible. Can we quickly come up with what we need? 

This story is part of our November 2021 issue

Aspuru-Guzik (one of the 35 innovators under 35 in MIT Technology Review 2010) spent most of his life working on a version of this problem. Material discovery—the science of creating and developing useful new substances—usually progresses at a frustratingly slow pace. The typical trial and error method, that is, scientists produce new molecules and then test each molecule in order to obtain the desired properties. It takes an average of 20 years, which makes most companies pursue too much cost and risk.

Aspuru-Guzik’s goal, which he shares with more and more computer-savvy chemists, is to shorten this time interval to months or years so that humans can quickly accumulate resources to deal with climate change, such as Battery and carbon-capture filter. The goal is to revive the dying materials industry by incorporating digital simulation, robotics, data science, artificial intelligence, and even quantum computing into the discovery process. 

Imagine using precise knowledge of the electronic structure of molecules to create newly designed computer programs; imagine robots that make and test these molecules. Imagine software and robots working together—testing molecules, adjusting designs, and then testing again—until they produce materials with the characteristics we are looking for.

At least this idea. Actual implementation is another matter. The structure of the molecule is incredibly complex, and chemical synthesis is usually more of an art than a science, ignoring efforts to automate the process. But advances in artificial intelligence, robotics, and computing are bringing new vitality to this vision. 

Aspuru-Guzik co-hosted a seminar held in Mexico City in 2017. 133 participants (including Nobel Prize-winning scientists and representatives from 17 governments) gathered to bring the global research community to the attention of this goal. The conference was a pivotal moment that helped transform the field of accelerated material discovery from a research niche area to a global priority for many attendees. After the event, countries such as Canada, India, and the European Union began to invest in initiatives to accelerate material research. 

The work itself is ambitious and technically difficult because it spans so many disciplines. But as a chemist, software engineer, artificial intelligence pioneer, quantum computer programmer, robotics enthusiast, and serial entrepreneur, Aspuru-Guzik may have the right combination of computational expertise and imagination, which can add as much as necessary to achieve this goal. Kinds of tools. He has become one of the most persuasive evangelists of new chemical methods.

"Allen can go beyond what people think is possible," said Joshua Schreer, a chemist and frequent collaborator at Fordham University. He is the kind of innovator, Schreer said, he changed the way everyone around him practiced science. 

For Ryan Babbush, head of Google's quantum algorithm team, Aspuru-Guzik's most prominent personality trait is his creative agitation. "Allen spends his time and energy on the latest things, the most unknown areas," he said. "He will not persistently focus on incremental development." 

Considering the time and hard work required to bring new materials to the market, this may be a problem-this work requires tenacious, narrow research and endless commercial patience. But in the end, Babbush said, Aspuru-Guzik was interested in reimagining the process of material discovery, equipping scientists in the community with the computing and automation tools they need to speed up their work. 

Today, Aspuru-Guzik is setting up a laboratory in Toronto where artificial intelligence algorithms design new molecules, and robots quickly manufacture and test them. The laboratory is a prototype designed to show how material discovery might work in the future. "I want to open a whole new era, an era of on-demand materials, where every laboratory can easily create new compounds," he said. He hopes that in the future, we can better respond to the next global crisis. "The world's problems need molecules," he added. "And now, we make them terrible."

Aspuru-Guzik speaks enthusiastically, off topic, and speaks quickly. When I visited his office at the University of Toronto for the first time, he showed me a series of lucha libre (Mexican wrestling) masks-bright blue, green and pink balaclavas decorated with Aziz Turk pattern. "Mask is a human tool," he said. "You bring a Nobel Prize winner or an executive from Hitachi to your office, chat for a while, stop and say'take off the mask'. Just take a selfie.' "It's hard not to take a selfie. The mask is seen as a metaphor for many aspects of his life. 

Aspuru-Guzik grew up in a family of semi-Catholic, semi-Jewish writers, musicians and architects. As a 19-year-old chemistry student at the National Autonomous University of Mexico, he returned from a night out in Cuernavaca when his car deviated from the road and crashed. The surgeon had to open his abdomen to repair his intestines and burn the broken blood vessels, leaving him with a scar, like a midline, extending to the center of his abdomen. 

After passing by with death, he began to devote himself to a life of intellectual adventure. If a certain field of research arouses his interest, he will pursue it, even if it is esoteric or beyond his expertise. 

At the time, people were very excited about the possibility of using computer-based modeling to design molecules with desired properties without the need for slow and tedious experiments. Scientists talked about a new era of virtual chemistry, but it didn't work well. The computer is too slow and the molecule is too complicated. 

While browsing journals in the university library, Aspuru-Guzik stumbled upon a paper about molecular chemistry challenges in computers. In 1926, physicist Erwin Schrödinger published an equation to predict the behavior of subatomic particles such as electrons and protons. If you can mathematically model molecules at the subatomic level, you can begin to make inferences about the resulting material: how it combines with other materials, how hard or soft it is, or how fast it decomposes. At least this idea. But for most materials, even today’s largest supercomputers, Schrödinger’s equation has become too complicated. 

To make mathematics feasible, Aspuru-Guzik set out to create versions of equations that required fewer approximations to make them more accurate-this project became the focus of his PhD research at the University of California, Berkeley. The goal is to simplify the calculations to the point where the computer can handle them, but not to make the model scientifically useless. Using Aspuru-Guzik's algorithm, researchers can model (ie, simulate) random molecules and immediately predict the properties of the resulting substance. 

Other scientists have also designed similar algorithms, but the algorithm proposed by Aspuru-Guzik when he was a graduate student was impressive, allowing him to find a job at Harvard after finishing his postdoctoral work at Berkeley. As an assistant professor at Harvard University and the leader of the Aspuru-Guzik research group, this 40-person team composed of computer scientists, biologists, engineers, physicists and chemists, he devoted himself to a project called Harvard Proposals for clean energy projects. Most solar panels use silicon to convert sunlight into electricity. However, is there any cheap, easy-to-manufacture organic substance that can accomplish this job? 

Aspuru-Guzik's hobbies (starting from the upper left corner) range from street art stickers to laboratory robots to Mexican lucha libre masks to automatic fluid handling.

For more than six years, Aspuru-Guzik and his team have simulated 2.3 million different organic molecules to see which ones might have photovoltaic properties. He is not the first researcher to practice virtual chemistry, but he is doing this work on an unprecedented scale. The increase in computing power in that era meant that a single molecule could be simulated in a matter of minutes; in the 1990s, such simulations took several days. Most importantly, he has access to seemingly unlimited server space, most of which is borrowed from other people's equipment. In a system similar to the old SETI@Home project, people who want to support the project can download a screen saver and temporarily lend their hard drive to Aspuru-Guzik and his team. "We have one of the largest supercomputers in the world," he said, "but it's distributed all over the earth."

Finally, Aspuru-Guzik discovered many organic materials that could theoretically be used in photovoltaic cells. The problem is that these successful molecules are too complex to make cheaply. "My mistake," he said, "didn't consult an organic chemist in the beginning to find out which molecules are easy to make."

Through clean energy projects, Aspuru-Guzik has basically been performing combinatorial chemistry in a computer rather than in a laboratory-the old trial and error method. Then, starting in 2012, researchers in Toronto and elsewhere have made a series of breakthroughs in deep learning and other machine learning methods. Like many chemists looking for new materials, Aspuru-Guzik turned to artificial intelligence, which allowed him to discover molecules in a faster and more thoughtful way. "The computer simulation is like a machine gun shooting randomly in the air, hoping to be hit," he said. "Artificial intelligence is a sniper. It chooses a target and takes aim."

First, he must train the neural network by providing it with data sets describing the molecular composition and chemical properties of 100,000 organic substances. The AI ​​program can begin to recognize patterns—that is, the correlation between a given molecule and the substance it forms. It can then use this knowledge to invent candidate molecules to be synthesized and tested in the laboratory. With the help of artificial intelligence, Aspuru-Guzik discovered a new type of organic light-emitting diode or OLED that is brighter than typical LEDs. He also identified new chemicals that will be used in organic flow batteries in the future. This large-scale industrial battery does not require metals such as lithium.

At the same time, he devoted himself to the emerging field of quantum computing. Schrodinger's equation is difficult to run on a classical computer, precisely because electrons and protons do not obey the laws of classical physics. Instead, they operate according to quantum mechanics: they can be entangled (even if they are not connected to each other), they can exist in a so-called superposition (occupying multiple opposite states at the same time). The modeling of these complex phenomena The required mathematical operations are also dazzlingly complex. But because the qubits in quantum computers also follow the laws of quantum mechanics, these devices are at least theoretically more suitable for simulating molecules. 

However, in practice, someone has to figure out how to make the simulation work. In 2014, Aspuru-Guzik and a group of researchers released Variational Quantum Eigensolver (VQE), a program for modeling molecules, even though it is on a small, error-prone quantum device, which is different from a general quantum computer. Actually exists today. Although the Schrödinger equation is an abstraction-a mathematical formula designed to describe subatomic particles-VQE uses qubits to simulate the behavior of particles in molecules, just as players in a reenactment might perform the Battle of Gettysburg . 

Over time, as the company develops more powerful quantum computers, VQE allows chemists to run extremely accurate simulations. These models can be so accurate that scientists don’t need to synthesize and test materials at all. "If we reach this point," Aspuru-Guzik said, "my work in materials science will be complete." 

When Donald Trump was elected president of the United States in 2016, Aspuru-Guzik's career was flourishing, but suddenly the prospect of remaining in the country no longer appealed to him. One week after the election, he started emailing colleagues in Australia and Canada , Looking for a new job. 

The University of Toronto provided him with a prestigious government-funded position designed to attract top researchers to the country and cross-appointed at the Vector Artificial Intelligence Institute, a non-profit company co-founded by machine learning pioneer Geoffrey Hinton , The company is rapidly developing Toronto is the global center of artificial intelligence. However, the biggest incentive was the commitment to establish a new material laboratory called Substance Laboratory, a project that Aspuru-Guzik has been dreaming of for many years.

"In the Substance Lab, we only started to solve the problem after asking three questions," Aspuru-Guzik said. "Is this important to the world? If not, then fuck. Has anyone else done it? If the answer is yes, then it doesn't make sense. Is it remotely possible?" Here, the word "remote" is the key . Aspuru-Guzik hopes to solve the challenges within the scope of feasibility, but almost none. "If a material is too simple," he said, "let others find it." 

The laboratory is located in a post-war brick building in downtown Toronto, unlike any other laboratory in the university. The ceiling is decorated with maroon and burgundy sound-absorbing panels, a tribute to the beloved Mexican architect Luis Barragán. Hidden in an inconspicuous corner is a typical laboratory bench—a table with flasks, scales, and beakers under a fume hood—where graduate students can practice chemistry, just like their grandparents’ generation did. Like that. People will feel that this workstation is not often used.

In the middle is a $1.5 million robot-a glass and metal shell filled with nitrogen, with a robotic arm that moves back and forth along a track. The arm can select powders and liquids from a series of tanks near the side of the housing and store the contents precisely in one of the multiple reactors. "The robot is like a tireless laboratory assistant, mixing chemicals 24/7," Aspuru-Guzik said. It can make 40 compounds in just 12 hours.

Two additional functions make the experimental setup of the substance laboratory unique. The first is a software designed by Aspuru-Guzik and his collaborators, called ChemOS. It includes an artificial intelligence system that generates candidate molecules and a program that interfaces with the robot to guide it to synthesize candidate molecules on demand. 

The second notable feature is the "closed loop" nature of the production process. To explain how this works, Aspuru-Guzik points to a pair of narrow hoses behind the robot. "This is where the pee comes out," he said. After the reaction is complete, the generated liquid will flow through a plastic hose into an analysis machine equivalent in size and shape to a mini refrigerator, which can separate unwanted by-products. The refined material will flow into another robot, which will test it to understand its characteristics. The robot will then feed back the results of the experiment to the ChemOS program, allowing the AI ​​to update its data and immediately generate new and better candidate molecules, until-after multiple rounds of prediction, synthesis, and testing-there is a winner. 

The idea of ​​an automated, closed-loop discovery system, partly because of Aspuru-Guzik's tireless advocacy, is becoming more and more popular among new chemical practitioners. Colleagues in Vancouver, New York, Champaign-Urbana and Glasgow are building similar facilities. These laboratories are intended to serve as general-purpose, automated spaces for molecular creation. This is why Aspuru-Guzik did not speculate too much about what the material laboratory will produce next. Such a decision may be determined by curiosity, or by the necessity of a global crisis.

In 2020, Aspuru-Guzik experienced a period of weight gain at the beginning of the pandemic, which caused his surgical wound to reopen. At the same time, he felt sleepy and bored with the 2D world called by Zoom, and frustrated that he could not roam freely in the laboratory. His busy work life hardly left room for that kind of aimless-or seemingly aimless-pursuit, which in the past promoted creative breakthroughs. He needs to change. 

A few months later, he started graffiti on his computer and drew a lucha libre mask similar to Screamin' Jay Hawkins, a rock pioneer known for his opera vocals and creepy stage antics. He named this character Bruho (a variant of "brujo", a wizard in Spanish) and decided to impose his artwork on the urban landscape. He bought a self-adhesive printer and began to paste Bruho's portrait on mailboxes and street lights. Soon, he became part of the bustling street art scene in the city.

Today, Aspuru-Guzik has two goals in the near future. The first is to design a modular and affordable version of his closed-loop system that can be used as a model for scientists around the world. He wants to build an all-in-one laboratory box that contains the ChemOS software package and synthesis and characterization robots. Using this device, the user will enter a description of a given material, and the system will immediately simulate and test candidate molecules. Aspuru-Guzik believes that if we are to create a new era of on-demand materials, this technology must be popularized and it must be easy to use.

His second mid-term goal is to leave his mark on the city of Toronto in art. 

After visiting the laboratory for a few days, I spent an evening with stickers and posters with him and his team. Just like his material work, this is also collaborative. Our group of eight includes Soap Ghost, an indifferent young woman with full sleeve tattoos; Urban Ninja, a sturdy middle-aged man who is pulling a cart and carrying a bucket of homemade liquid adhesive wheat paste; And life, a stubborn insomniac, had his hair split in the middle, half dyed golden, just like Crura Deville's. "I will walk until sunrise," he told me playfully. Everyone has their own designed bundle of stickers or poster rolls.

In Toronto, this kind of street art that does not require spray painting is fined (even if the police often disagree), so we acted quickly and secretly. The ninja took us along an alley to the exposed plywood wall of a wooden building. We landed on the wall with a brush, applied a paste on the surface and pasted an image on it-Buddha with a beard, bomb Ukulele mouse, Bruho's figure, robe like Jedi. This combination does not bring a lot of visual feeling, but it has an anarchistic beauty. In a very short period of time, emptiness gave way to diversity, and Aspro-Guzk was excited. "This wall was blank a minute ago," he exclaimed. "Look now.

Simon Lewsen is a magazine writer based in Toronto. 

The P vs NP solution can solve countless computational problems—or make them out of reach forever.

The Dutch company ASML spent 9 billion U.S. dollars and 17 years to develop a method to continue manufacturing denser computer chips.

The U.S. government is embarking on a generation-long battle to counter the threat to encryption posed by next-generation computers.

QuEra Computing, initiated by physicists from Harvard University and the Massachusetts Institute of Technology, is trying a different quantum approach to solve impossible and difficult computing tasks.

Discover special offers, popular stories, upcoming events and more.

Thank you for submitting your email!

There seems to be something wrong.

We were unable to save your preferences. Try to refresh this page and update them again. If you continue to receive this message, please contact us at customer-service@technologyreview.com and provide a list of newsletters you wish to receive.

Our mission is to make smarter and more conscious decisions about technology through authoritative, influential and trustworthy news reports.

Subscribe to support our news work.