Faster, chemist! Build! Build!

Alán Aspuru-Guzik just got $200 million from Ottawa to make cool things more quickly. He doesn't plan to dawdle.

Alán Aspuru-Guzik in the Matter Lab, July 11, 2023. Photo: PW

On one side of Alán Aspuru-Guzik’s lab at the University of Toronto, he makes molecules that shoot laser beams. On the other side he turns oil-sands bitumen into batteries.

Any sufficiently advanced technology, Arthur C. Clarke once wrote, is indistinguishable from magic. This stuff is getting close. But the most important product from Aspuru-Guzik’s lab might not even be the laser molecules or the clean energy from bitumen. It might be the way he makes them: with automated experimental equipment that is directed and monitored by artificial intelligence. Aspuru-Guzik and his colleagues call them “self-driving labs,” or SDLs, a name modeled after self-driving cars. They could open a new era of faster, cheaper, more widespread discovery in materials science.

The 47-year-old Mexican chemist left Harvard for Toronto in 2018. He’s the director of something called the Acceleration Consortium. It’s big, and growing. In April the AC, as its members call it, landed a $200-million grant from the federal government. It’s the largest research grant any Canadian university has ever received, from any source, for anything.

Often in this newsletter and in my old magazine job, I write about science policy — the administrative and spending decisions governments make in support of scientific research. The first thing that drew my attention to the Acceleration Consortium was just the eye-watering size of the grant. It’s as big as the initial investment in any of the five “Superclusters” the government launched with great fanfare in 2017. The program that administers the grant, the so-called Canada First Research Excellence Fund, is interesting too. CFREF was launched a decade ago in the late days of the Harper Conservative government, to promote more interdisciplinary “big science” of a kind that’s been more common in the United States, Europe and Asia than in Canada.

But policy isn’t everything. Once in a while I just start to wonder what the scientists themselves are up to. So at the beginning of July I spent a couple of days in Toronto with Aspuru-Guzik and his colleagues.

“The goal in this particular lab is at least two materials applications,” he said as we began walking around his team’s workspace, which he’s dubbed the Matter Lab. “Materials applications” means “things we’re making.” The first thing they’re making is laser molecules.

Most high-end smartphones already use screens with OLEDs, light-emitting organic molecules. “It’s incredible, a little powder that is spread in every pixel. And these pixels emit light,” Aspuru-Guzik said. But the light from OLEDs is old-fashioned, scattered, chaotic light, like from a light bulb. What if you wanted to beam the light directly and coherently, perhaps in some next-generation augmented-reality headset?

“Think about it. You need something that is very sexy: a transparent substrate that emits light to your eyes,” Aspuru-Guzik said. “You know, lasers are directional. So can you emit light directionally from an organic compound? The market is huge, billions of dollars, right?”

There have already been organic molecular lasers for several years. They’re the damnedest thing. Where once it took elaborate machines to transform traditional chaotic light into the tightly focused, coherent beam of a laser, now there classes of organic molecules — chains of carbon atoms with atoms of other compounds hanging off the sides and ends — that do the same work at a vastly smaller scale, just by the way they’re shaped and the way they interact with light.

There used to be 10 or a dozen such laser-making molecules, Aspuru-Guzik said. Now? “Let me show you 500.” He opened a cupboard to reveal row on row of vials:

“Now we’re fucking talking,” he said.

Now, these particular racks of laser molecules weren’t designed to be candidate products for a billion-dollar market, but to serve as a data set. The light-emitting characteristics of the slightly different molecules in each vial become the basis for an AI algorithm to learn what works and what doesn’t in the world of organic lasers. And because AI is designed to extrapolate from large numbers of known cases, when you teach an AI about organic lasers, it can start to predict which new molecules will work well and which won’t.

The Matter Lab’s template for potential laser-making molecules could, in theory, produce 160,000 different molecules. Most would be duds: too hard to make, unstable, wrong frequency of light, whatever. The predictive power of AI should allow chemists to skip the part where they make and test all the duds. They can go straight to the compounds with the best chance of working right. The savings in time and money would be huge.

So the Matter Lab’s main product isn’t the molecules, it’s the technique for making them. AI and automated production in a closed and iterative loop: run a reaction, analyze the outcome, refine the process, start again. The whole thing needs human researchers to design the system and load the chemicals, but then the machines take over. Set it and forget it.

And as the cost of the self-driving labs comes down, and the techniques for making them become more standardized….

Aspuru-Guzik introduced me to Han Hao, a post-doctoral fellow in his lab. Han showed me a self-driving lab he’d made to answer a specific question implied by a larger experiment. These robots aren’t much to look at, mostly just boxes and flasks connected by thin tubes. Here and there an off-the-rack robot arm. The total cost for the one Han was working on was $5,000. And how long did it take to make it? “It took me something like one afternoon,” Han said. “It took my undergrad two days, because she’s new to the system.”

“Our selling point is that we can make these available to anyone in the world,” Aspuru-Guzik said of the ungainly machines.

But if the labs do their self-driving somewhere else, won’t the next discovery be made by someone else? Aspuru-Guzik smiled. There are far more molecules in the universe than people to look for them, he said. “In other words, no worries, brother. If I do a laser, somebody else will do a laser, somebody else will do a — we don’t know [what they’ll make.] We are not jealous. We are the opposite.”

We rounded another corner in the Matter Lab and came to another set of automated experiments. Here was the lab’s second materials application.

 “I’ve been working on organic batteries for a very long time,” Aspuru-Guzik said. He showed me a new variation. “I’m going to take tar sands — organic molecules — instead of pouring them away, I’m going to convert them into a battery that is functional organic. That means, without metals.” You know all the debates about critical minerals, the rare metals whose supply is the limiting factor in so much clean-energy production? Aspuru-Guzik thinks it might be possible to do without them.

“I’m not going to go and mine a territory and pull out the metals,” he said. “I’m going to make a battery that’s fully organic. There are terrible problems mining chromium, nickel, in Africa, in Canada. So I could become a trillionaire, if this works and I had the patent.”

The laser molecules on the other side of the lab are proof-of-concept, a way to refine the SDL model, teach new researchers, improve the technology. “The organic laser is more like a gizmo,” Aspuru-Guzik said. “We needed a gizmo. This one is more like, ‘Let’s save the planet,’ right?”

Of course, there are many ways to save a planet. As the SDL technology spreads to other labs working on other puzzles, the possibilities multiply. “Downstairs we have a self-driving lab for organs on a chip. Another colleague of mine is doing alloys, like the most rust-resistant alloys. Since I got here I have started either converting people to the self-driving lab world, or recruiting. Four people since I got here.”

---

Here’s the problem with chemistry:

It’s really hard to do.

Chemists envy biologists. Which might sound odd, because biology is just a branch of chemistry in which strange and wonderful things happen, things we call “life.” But to chemists, it seems that most of the time, their brethren in the world of biology have it easy. Most biological processes happen near room temperature in environments that include liquid water, because most biology takes place in cells, which are basically just bags of water.

The rest of chemistry extends much further afield, to materials that are caustic or corrosive or toxic or hard to manipulate. The reactions that break chemical bonds between atoms and form new bonds might best be performed at high temperatures, or very low, or in a vacuum, or first one of those and then another. Most chemists have to devote much of their lives to becoming ninja experimentalists. They start training early. They put in huge amounts of practice time in labs. And no matter how good they get, they can’t escape an iron rule: what most chemistry experiments require, more than any other ingredient, is the attentive presence of a human chemist.

Over in the balmy water-bag world of biology, there’s been breakthrough after breakthrough in high-throughput biology, which involves squirting thousands of almost-identical compounds into thousands of receptacles, producing thousands upon thousands of results, which can be analyzed by computer.

This practice of blitzing all the possible permutations of an experiment has paid spectacular dividends. In 2000, it cost $100 million to sequence a single person’s entire genetic code, or genome. By 2018, that cost had fallen below $1,000, a far faster rate of progress than would have been predicted by Moore’s Law.

Great. Wonderful. Congratulations. Try doing that with the gnarlier chemistry that happens outside cells. By the early 2000s, some chemists were trying a work-around: “virtual chemistry,” which involved using computers to predict the outcomes of imaginary reactions. The process involved lots of approximations, but you could always check your results later, in a lab.

As an assistant professor at Harvard, two decades ago, Aspuru-Guzik decided to go hunting for organic photovoltaics, carbon-based materials that could power solar panels without the need for silicon or rarer substances. Over six years, he and his colleagues supervised computerized evaluations of 2.3 million candidate molecules. A daunting data set, even for a supercomputer. So he farmed the work out to thousands of laptops.

“I was very naive,” he recalled. “I created a screen saver called the Harvard Clean Energy Project.” People would download the screen saver and then, while their computers were otherwise idle, they would be running simulations on candidate molecules for solar panels. “I was already thinking big. I said, ‘I’m going to use all the screen savers in the world.’ We had tagged it the largest computational project in history.”

Remember the part about how hard chemistry is? All those laptop screensavers produced dozens of candidate molecules for a new generation of solar panels. In theory they should have been able to transform light into electricity. But in practice most of them couldn’t be produced in real life, not cheaply, which was supposed to be the point.

With organic flow batteries, the magical compounds he now hopes to synthesize from bitumen, “I was much smarter. I did computations only on about 10,000 molecules. My colleague synthesized it and we published a paper in Nature in 2014: ‘Here is the first organic flow battery.’”

The next step in finding the most productive pathways in chemistry is to let an AI algorithm help find them. This idea came to a bunch of people within a few years of one another: Lee Cronin in Glasgow, Martin Burke at the University of Illinois in Urbana-Champaign, Bartosz Grzybowski who divides his time between Poland and South Korea, and others.

One thing that’s striking about this emerging field is how cooperative it seems to be. Everybody I just named is a member of Aspuru-Guzik’s Acceleration Consortium. Many have their own startup companies, marketing processes and materials they’ve developed in their labs. But nobody owns the notion of a self-driving lab, and the leaders in the field compare notes constantly.

“Honestly, it's such a huge opportunity,” Martin Burke told me when I called him at his Molecule Maker Lab outside Chicago. “It's transformative. The world's going to be different. And there's so much to do. So it doesn't make any sense to slice up the pie. I think we've kind of unofficially decided to grow the pie together, and look forward to telling our grandkids about it.”

Why’s the world going to be different?

“There's not a problem facing society today that couldn't benefit from a new molecule,” Burke said. “You look at history of society — for, say, 2 million years of recorded human history, we’ve been making tools to create the world around us. That was arguably, the transition point into our human-ness. And that's how we create our society, right? We create tools.

“We've only been making molecular tools on purpose for 200 years.” Roughly since Wöhler synthesized urea in 1828. “That’s very inspiring, because that means we've barely scratched the surface, right? Literally 200 years of molecular toolmaking. And then you think about everything that molecules can do. I mean, molecules can do amazing things. Literally, you and I are having a conversation right now because a bunch of molecules are flying around in our heads. It's very humbling and inspiring to think about the large — I mean, massive, astronomical — untapped potential of molecular tools to help us create a better world.”

---

You’re probably wondering about the name.

Alán Aspuru-Guzik was born in the United States but grew up in Mexico City. His parents were Mexicans with family roots abroad: His father, Aspuru, from Basque sailor stock; his mother, Guzik, from Central European Jews.

As a teenager he represented Mexico at the International Chemistry Olympiad in Oslo. “I studied Swedish as an undergrad, because I was very interested in living in Scandinavia. Like, I was very interested in this idea that poor people had a chance. In Mexico, when I was walking to college, you saw people in the street. Literally, when NAFTA was signed, all the farmers lost their jobs because corn was very expensive compared to the United States. There was a huge migration to Mexico City. And I just happened to be commuting to college, seeing 25-year-old, 30-year-old people sleeping in the streets with nothing. That was one of the life moments when I was like, I have to do something about this. That’s a person. This is not okay.”

He studied at the National Autonomous University of Mexico and at the University of California, Berkeley, both sprawling public universities with huge student bodies. But as a post-doc — an early-career researcher, the next step after his PhD — he published a paper that opened a door for him at Harvard.

“Harvard is like a brand, it’s like Coke, it’s name recognition. It’s very powerful: ‘Wow! Harvard!’ But to be honest, it’s a university like any university.

“But Harvard shaped me in one very important way. They told me, ‘Alán, you have to create a sub-field of chemistry to get anywhere here.’ That’s the bar.” In other words, you couldn’t just master the fundamentals of a craft someone else had developed. You had to break new ground.

At first he mostly broke ground with his sociability. “I brought from Mexico a bunch of lucha libre masks,” he said, referring to the Mexican brand of professional wrestling with its lurid, colourful leather masks. “And I started, with my students, every time we published a paper, taking selfies with them.” Pretty soon other members of the chemistry faculty would drop by. Harvard being the sort of place that it is, sometimes those colleagues were Nobel prize-winners. They’d try on a lucha libre mask, have a laugh, stick around for a beverage and some shop talk.

On a shelf in his office are some Mexican tribal boots, apparently a comic fashion trend from a decade ago. “I always had 10%, 20% Mexicans in my lab, my grad students,” he told me. The day he got tenure at Harvard, his students gave him the pair of botas picudas. A Mexican cleaning lady in the chemistry department gave him a sombrero. Thus decked out, he led a parade through the hallowed halls of Harvard Chem. He was set for a rewarding career. With colleagues he launched startup companies in materials science and quantum computing.

Then Donald Trump was elected President of the United States. “My kids were in school shooting drills, where the kids had to be under the desk. In Cambridge, MA, where I was living, halfway between MIT and Harvard, there were lockdowns in the school three times a year because, I don’t know, some crazy guy was passing next to the school and they made a lockdown. So my wife was not happy. But me too, I was like, ‘Why am I here? Remember Scandinavia, remember that idea that you had?’”

So the day after the 2016 election Aspuru-Guzik started calling contacts in Australia and Canada. “I said, ‘Hey, can I move?’” These new potential destinations weren’t quite Scandinavia, but his wife didn’t speak Swedish.  The director of CIFAR, my friend Alan Bernstein, introduced me to a very high level at UBC and at the UofT.”

CIFAR is a sort of Canada-based, global research university without a campus. Under Bernstein, who was director for a decade until 2022, the organization and the Trudeau government in 2017 launched a pan-Canadian AI strategy.

With Bernstein’s guidance, in 2018 Aspuru-Guzik took an appointment at the University of Toronto, thanks to a seven-year federal grant at $1 million a year to run the Matter Lab.

Why Toronto? “First of all, UofT is very similar to where I did my PhD, UC Berkeley. In fact it’s the second public university after UC Berkeley, ranked globally.” As I was reminded when I wandered around the campus for a couple of days, Toronto, like Berkeley and Mexico City, impresses through its size and comprehensiveness more than through ostentatious refinement. But that’s the sort of vibe that makes Alán Aspuru-Guzik feel at home.

He launched the Acceleration Consortium in 2019 with $5 million from a UofT initiative to launch large multi-disciplinary research efforts. Landing the really big grant, the CFREF grant, was a strategic goal, but to do that, the university had to put its own skin in the game: $130 million for an expansion of the university’s chemistry building, and $300 in matching grants from other partners.

The $199 million from CFREF is to be paid out over seven years. Nearly $180 million of that amount goes to UofT; about $20M will go to AC’s lead partner, the University of British Columbia. Much of the money will pay for recruitment. AC will have 30 staff scientists, paid more than what university faculty usually get. Aspuru-Guzik was looking forward to seeing Oppenheimer. He has the Manhattan Project on the brain.

“Let’s call this a consortium. Why [would he want to call it] an institute? I want industry, I want government, I want startups. This is not just academia. We want to make Toronto the capital of materials for the world, of AI for materials.

“I am sitting on a big responsibility. We got the largest grant ever awarded. And I am doing everything possible so that this grant is not wasted. The worst-case scenario is, we’re training shitloads of people. Like, the worst thing that can happen is, we’re training hundreds and hundreds of scientists that will help Canada. But that’s, like, boring.”

What’s less boring is the installation of a new model for focussing talent on big problems. In the United States, such things are called Focused Research Organizations, or FROs.

After I got out of his hair, Aspuru-Guzik was going to have meetings with a construction company to discuss using self-driving labs to find lower-emitting construction materials. Another meeting in his calendar was with a mining company, to find treatments for mine tailings that could make them far more efficient at carbon capture than forests are.

The number of projects that could benefit from self-driving labs is, potentially, really large. But of course, looking isn’t the same as finding, and chemistry is still hard even if robots and AI reduce the scope for failure. But to Aspuru-Guzik, this model for building interdisciplinary teams on big challenges has its own value.

“I mean, think about my worst-case scenario. Like, we produce three materials and seven years later you write in your Substack, ‘Hey, this guy sucked.’ But maybe we brought the FRO model to Canada and then another FRO succeeds.”

A few years ago, Aspuru-Guzik invented a cartoon character with a lucha libre mask. He called his alter ego Bruho, a corruption of the Spanish word brujo, for sorcerer. Sometimes at night he goes out on the town with friends, putting Bruho stickers on light posts and back alleys. Where he travels, Bruho stays. When friends travel, they take Bruho with them. Bruho has made his presence known in many places, from Montreal to the Seychelles:

For an immigrant kid who grew up in tough times in a changing Mexico, of course Bruho is an outlet, a kind of symbol. “In some sense I was like a luchador – I got punched and I came up,” Aspuru-Guzik said. “The luchador effect is that kind of fighter in me. It also represents my Mexican side, my people that I’m very proud of.”

Is the scrappy cartoon luchador a metaphor, too? Aspuru-Guzik is working on a technology that might pop up anywhere, even where you least expect it. Sure it’s a stretch, but so is building a world-beating new science colony two countries north of where you started.

“I think,” Aspuru-Guzik said, “what I share with Bruho is ambition.”