The Magic Lab

The morning after the 2016 presidential election, Alán Aspuru-Guzik woke up, shook off his sorrow and went for a run. Striding across the grounds of Harvard University, he tried to imagine what life might soon be like for him, a Mexican American professor of chemistry and computer science and a self-described “pinko commie guy at heart.” What kind of future did he have in a country run by Donald Trump, an anti-science nationalist who’d called Mexicans criminals and rapists and promised to build a wall to keep them out of America? As he was jogging, Aspuru-Guzik spotted a man in a red MAGA cap confidently strolling through campus. “I had never seen anyone at Harvard doing this MAGA shit before,” he says. To him, it was an omen: even in the ivory tower, he wouldn’t be safe.

At lunch, Aspuru-Guzik swapped worst-case scenarios with fellow Harvard prof and How Democracies Die co-author Daniel Ziblatt. They prophesied cuts to education and research and a rise in authoritarianism and violence. Aspuru-Guzik knew he’d be leaving the US, following in the footsteps of his Jewish ancestors who had left Russia during the pogroms. Later that afternoon, he and some colleagues bought, hung and bashed a Trump piñata. Related: Trump’s Loss, Toronto’s Gain—Meet the artists, professors, scientists and other luminaries ditching the US and moving north

Aspuru-Guzik visited universities in Australia and toured the University of British Columbia ­campus. But it was Toronto that captured his imagination. The chaos of the downtown core, jammed with cyclists, cars and pedestrians, reminded him of Mexico City, where he was raised. He liked the people and loved the food. Crucially, Toronto was also a wellspring of AI talent, which is what Aspuru-Guzik needed to advance his research into an emerging technology called self-driving laboratories, or SDLs—elaborate networks of computers, robotic arms and high-tech lab equipment that autonomously plan, conduct and analyze scientific experiments. When the University of Toronto offered Aspuru-Guzik a $3.5-million research grant to build the best SDLs in the world, he accepted.

The relocation could not have gone better. Aspuru-Guzik moved to Toronto in 2018, bringing 40 or so Harvard students and staffers with him. A few years later, he founded the Acceleration Consortium—a global network of universities, hospitals, government agencies, and corporate giants like IBM and Merck, dedicated to advancing SDLs—and secured roughly half a billion dollars in support, including a record-breaking $200-­million research grant from the Canadian government. To date, the AC’s growing team has built 32 self-driving labs, with which they’re developing life-saving drugs, designing new hip-replacement metals and looking into creating batteries made of organic compounds. Long-term, Aspuru-Guzik has even bolder ambitions. He says that, by using SDLs to automate and accelerate the discovery of new molecules, researchers may one day reach some of the holy grails of applied science: ultra-efficient batteries that turn the tide on climate change, artificial organs that prevent people from dying while languishing on transplant lists, patient-specific cancer treatment. “Finding new molecules,” he says, “will help us solve most of the practical problems humanity faces.”

Related: Why Geoffrey Hinton is sounding the alarm about AI

Because Aspuru-Guzik chose to conduct this potentially world-changing work in Toronto, the city has become the undisputed global leader in SDLs—and a magnet for anyone who wants to work with them. Since 2018, he has fielded calls from NASA, hosted scouts from multinational corporations—auto-parts manufacturers, pharmaceutical giants, even luxury fashion houses—and hired dozens of scholars from China, Iran, Singapore, Mexico and the US. Jason Hattrick-Simpers, a materials scientist who used to work at the National Institute of Standards and Technology, in Maryland, told me, “When Trump decided to throw a temper tantrum and close my lab just as we were making crazy progress, I didn’t talk to someone at Johns Hopkins. I didn’t talk to someone in Europe. I messaged Alán and said, ‘Get me out of here.’”

Aspuru-Guzik has been getting a lot of these calls lately. He says that what’s happening now with Trump—the unconstitutional executive orders, the indiscriminate sacking of world-class scientists, the dismantling of democratic institutions—is what he expected to happen eight years ago. Thousands of American academics now find themselves in the position he once occupied: wondering whether they should move north before they receive an email from DOGE with the subject line “Fork in the Road.”

To meet Aspuru-Guzik is to witness a human body try to keep up with a supersonic mind. The bald, bespectacled and occasionally bearded 48-year-old ricochets between topics: quantum computing and federal politics one minute, Mexico City’s food scene and Panamanian metal bands the next. His desk at U of T is a whirlwind of cords, papers and snacks; on the blackboard next to it, words scrawled sloppily atop half-heartedly erased equations provide additional evidence of a man in a hurry. A framed certificate awarded to him by the university’s former president sits in a box on the ground—Aspuru-Guzik hasn’t found time to hang it. It’s not apparent how he has time to do anything. He is a professor of both chemistry and computer science, the editor of the academic journal Digital Discovery, the organizer of an annual conference devoted to SDLs, and the co-founder of four AI and quantum start-ups. He has two young sons and moonlights as a street artist: five years ago, he invented a fictional luchador named Bruho, a character that embodies his Mexican roots and fighting spirit, and started plastering stickers bearing Bruho’s name and wrestling mask on street poles and hydro boxes in Toronto, Montreal, the Seychelles and beyond. I was both astonished and annoyed when Aspuru-Guzik told me that he does Pilates too. Related: “It started to feel a bit too much like Europe in the 1930s”: Music producer Bob Ezrin on why he moved from Nashville to Toronto

“He’s got that personality where he wants to go in 12 directions at the same time,” says his boss, Melanie Woodin, U of T’s president-designate and the former dean of the arts and sciences faculty. And most of the time, all 12 directions are worth exploring. Jorge Arturo Campos Gonzalez Angulo, a Mexican post­doctoral student in Aspuru-Guzik’s lab, says the hardest part about working for him is learning to say no. “Every week, Alán comes in with a new idea, and it’s so cool that you want to jump on it even if you don’t have time,” he says. “He’s looking for revolutions everywhere.”

Aspuru-Guzik traces his intellectual restlessness back to childhood. His father, a Basque electrical engineer and adamant atheist, was authoritarian, tolerating nothing less than immediate obedience to his unpredictable demands. But his mother, a psychoanalyst from a family of Mexico City Jews, was a driving force for Aspuru-Guzik and his younger brother, Daniel, and they credit her for their creative spirits. As a kid, he read voraciously: encyclopedias, astronomy books, issues of the New Yorker and Scientific American. His relatives, among them artists, engineers, writers, architects and musicians, encouraged an early affinity for science and technology, giving him a Corona computer, a dial-up modem and a Casio wristwatch. In his teens, he’d stay up until 4 a.m. “phreaking”: hacking foreign phone lines, pirating computer software and sharing it with other Mexicans via an online bulletin board system he created. He was a sort of digital Robin Hood, but he had plenty of juvenile fun too. When Mexico’s national soccer team beat Argentina in 1990, he and his friends exploited security vulnerabilities in the telephone network to place free long-distance calls to random Argentinians and rub it in.

An unremitting sense of urgency took hold of Aspuru-Guzik in 1995, while he was studying chemistry at the National Autonomous University of Mexico. That summer, on the drive home from a rave in Cuernavaca, his friend lost control of their car and crashed. Aspuru-Guzik, only 19, nearly died; his chest still bears scars from the surgery that saved his life. “I started thinking that life is very finite,” he says. “You could die tomorrow, so if you want to try something, you should try it now.”

---

When Harvard gave him tenure, Aspuru-Guzik—trailed by a mariachi band—led a parade through the chemistry department

---

He climbed the academic ladder quickly: an undergrad chemistry degree in Mexico City, then a PhD and postdoctoral position in quantum computing at the University of California, Berkeley. On the hunt for a faculty job in the mid-2000s, he mailed applications to 44 universities, including Harvard. “I was like, ‘Maybe I’m just wasting my postage with this one,’ ” he says. But the school hired him as an assistant professor, and within a few years, he was something of a campus legend. Part of his fame stemmed from the Harvard Clean Energy Project, a clever campaign in which participants downloaded a screensaver that allowed Aspuru-Guzik to borrow their idle machines’ computing power for his research into finding molecules that could be used to produce low-cost, ultra-efficient solar panel cells. Aspuru-Guzik was also known for his tradition of donning Mexican wrestling masks and taking selfies with his students whenever they published a paper. Fellow professors, including Nobel winners, stopped by to check out the máscaras; they usually stayed to talk shop. When Harvard made him a full tenured professor in 2013, Aspuru-Guzik—wearing a sombrero and red tribal boots, trailed by a mariachi band—led a parade through the halls of the chemistry department. As he told me, “It was my way of telling Harvard, ‘You have a Mexican. Deal with it.’”

Before Aspuru-Guzik left Harvard, he became enthralled with self-driving labs. As far as he could tell, very few had been built yet—and that only enticed him more. He wanted to put together a prototype, but he didn’t have access to a so-called wet lab where he could safely mix chemicals. So he came up with a workaround. He wouldn’t mix chemicals; he’d mix cocktails.

With help from a bewildered student, Aspuru-Guzik 3D-printed a frame the size of an espresso machine and outfitted it with a dozen simple parts: a microcontroller, a power supply, several small pumps. He connected the pumps to red Solo cups of tequila, various juices and vinegar (a negative control) and coded a rudimentary operating system that would allow the contraption to make margaritas and tequila sunrises. Aspuru-Guzik christened the prototype “Bob,” short for Bayesian Optimized Bartender, a nod to the influential statistician Thomas Bayes. The twist: Aspuru-Guzik didn’t give Bob recipes, nor was it able to access any online. Bob would need to learn the correct proportions itself.

To train Bob, Aspuru-Guzik invited his students to email their orders to bob.margaritor@gmail.com. Using natural language processing, Bob read incoming requests and pumped out random ratios of liquid. Its early beverages were ­undrinkable—pure vinegar, for instance. But, as students emailed Bob feedback (zero was perfect; four was terrible), the machine’s tiny silicon brain started to suss out what quantities of which ingredients belonged in each drink. The more cocktails Bob mixed, the better they got. Within a couple of hundred orders, it had mastered the margarita. “It was fucking awesome,” says Aspuru-Guzik.

It wasn’t the tequila that had him buzzing (he’s more of a mezcal guy)—it was the fact that Bob represented a promising proof of concept. If this primitive machine could learn to mix cocktails, he wondered, could a more sophisticated SDL teach itself to mix chemicals? Suppose he wanted to make a molecule rather than a margarita; could a similar set-up trigger, rank and repeat chemical reactions, trying different combinations of reactants under varying conditions until it pumped out a perfect molecule?

Such a machine, he thought, could revolutionize molecular discovery, the process that has yielded some of the most ­consequential inventions in human history: plastic, anti­biotics, the birth-control pill. Had scientists not discovered water-filtration molecules, humanity wouldn’t have enough clean drinking water. Without artificial nitrogen-fixation catalysts, a class of molecules crucial to agriculture, billions of people would starve. Penicillin, lithium-ion batteries, Covid ­vaccines—none would exist were it not for enterprising scientists seeking out new molecules.

The trouble with molecular discovery is that it’s glacially paced grunt work. Historically, searching for new molecules meant pipetting microlitres of chemicals from one vessel to another, waiting hours for a reaction, logging the results and then repeating the process ad nauseam. Using trial and error to search what’s known as “chemical space” is a monumental challenge. You could hire an army of scientists to work 16-hour days for an entire decade, and still there would be no guarantee of finding a molecule with the properties you were looking for. Using traditional methods, discovering a new molecule and turning it into a market-ready material takes an average of 20 years and costs roughly $100 million. But Aspuru-Guzik claims that, by automating that process and cranking it up to light speed, SDLs could eventually reduce that to just $1 million over a single year. “Imagine every chemist in the world becomes 10 times more productive and the cost of the work is lowered by 10 times,” he says. “Suddenly, it will be like we have 100 times more chemists in the world.”

Aspuru-Guzik’s ultimate ambition is to create an SDL that scientists will simply need to turn on; ask to find, say, an antibiotic that can target a previously untreatable bacterial infection; and provide with answers to a few questions about the request. They can then sit back and let the lab work its magic. The SDL would consult its training data—the properties of similar molecules, results of previous chemical reactions, simulations performed by a computer—to guess where in chemical space to start looking. Then the lab’s robotic arms would rapidly grab, pour, mix and measure different combinations of reactants, and machine learning algorithms would incrementally tweak the experiment’s inputs and parameters until it produced the winning molecule, the same way Bob inched toward the tastiest tequila sunrise.

SDLs aren’t quite there yet, but they’re not far off. When I visited U of T in February, I found myself staring into the belly of a multimillion-dollar steel beast. Its exposed innards—mechanical arms, coiled hoses and silver tracks—were quietly humming along, moving plastic vials to and fro with impressive speed and precision. Tubes and wires dangled out the back, connecting the hulking appliance to computers and other indiscernible bits of futuristic gadgetry. A couple of years earlier, in tandem with an international team of scientists, Aspuru-Guzik had used this state-of-the-art SDL to discover hundreds of new organic laser molecules—liquid substances that naturally glow different colours. The best and brightest of these molecules will give us more efficient screens and better sensors in self-driving cars. But, to Aspuru-Guzik, the lasers were the mere by-product of a more significant breakthrough: the lab itself. “This is the project I came to Canada to build,” he says.

---

“For us, success is when this is mainstream—when self-driving labs aren’t cool anymore”

---

There are now several SDLs scattered across the U of T campus. In the AI and automation lab, I found an Iranian staff scientist named Kourosh Darvish working through a complex computer-vision problem: training robots to recognize and handle translucent glass vessels under different light conditions. In another lab, this one walled by rows of sci-fi machines, Aaron Clasky, one of the AC’s rare Toronto natives, explained how he turns new formulations into materials—liquids, gels, creams, tablets and so on—and waxed enthusiastic about creating sustainable agrochemical solutions.

As I was wrapping my mind around these feats of science, I stepped into the office of professor Milica Radisic. A tissue engineer by training, Radisic uses human stem cells to create organoids, microscopic balls of flesh that are as gross as they are amazing. Organoids, as well as a similar technology known as organs-on-a-chip, are ideal for testing new drugs because they mimic human tissue. The catch is that creating and manipulating organoids was, until very recently, extremely laborious—a scientist would need to sit for several hours straight, arms held out in front of them, trying to manipulate tiny cells without making any mistakes. “We also had to come in on the weekends,” says Radisic, “because cells are living things and they need food every day, just like we do.”

A few years ago, Radisic got an unsolicited message from Aspuru-Guzik: how would she like to automate and accelerate organ mimicry? Intrigued, she attended a meeting with a group of computer scientists and roboticists, who outlined the potential of SDLs. “I had no clue we could do stuff like this at the University of Toronto,” says Radisic. “Very often, people will tell you 10,000 reasons why something cannot be done. But, in the Acceleration Consortium, there’s a culture of ‘yes we can.’”

Today, Radisic is a member of the AC’s scientific leadership team. She’s also 10 times more productive; where her lab once made 120 tissues per day, they now average 1,200. With gains like that, she argues, pretty soon it will be unusual for scientists not to use SDLs. “For us, success is when this is mainstream, when self-driving labs aren’t cool anymore,” she says. “We want to make these platforms so user-friendly that every industrial, biotech and academic lab can use them.”

Aspuru-Guzik is chasing such a future with characteristic haste. “We want Toronto to become a place where so many self-driving labs are being made that, every time we make one, it’s cheaper and faster than the last,” he says. To that end, the AC uses the open-source coding language Python, publishes many of its papers in free-to-read journals and offers inexpensive online courses to eager scientists who want to make their own SDLs. (Someone in Spain recently messaged Aspuru-Guzik on LinkedIn, proudly attaching a picture of a homemade SDL.) Part of the rationale for this all-aboard approach is that, the more SDLs there are, the more experiments they’ll run and the more data they’ll contribute to our ever-expanding understanding of chemical space. The deeper that understanding, the faster SDLs will be able to find, synthesize and commercialize in-demand molecules that will transform the world.

All of this is, in subtle and not-so-subtle ways, transforming Toronto too. As funding flows to the city’s universities and start-ups, quality jobs are opening up, brilliant newcomers are filling them and revolutionary discoveries are being made. It’s a snowball of success: as more breakthroughs come out of Toronto, the city will attract more bright minds, who will in turn make their own marks, and so on. Toronto’s prosperity is also a symbolic victory against Trump’s isolationism—proof that people from every country and creed can come together, share ideas, work hard and change the world for the better. “I feel like this is the right city for me,” says Aspuru-Guzik. “Unless we get invaded by the Americans. Then I want to be as far away as possible.”

This story appears in the May 2025 issue of Toronto Life magazine. To subscribe, click here. To purchase single issues, click here.