On the University of Glasgow campus, past a 17th-century entry gate, a grungy brick building houses the laboratory of the Regius professor of chemistry. Not much has changed since the first titleholder was appointed by King George III in 1818. Experiments are still conducted in glass flasks — although now by students in T-shirts and jeans.
Strolling through the building in a sporty tweed jacket and khakis, the current Regius professor proclaims that everything will soon be different. “In any physics or biology lab, there’s automation,” Lee Cronin tells me. “In chemistry, it’s all still done by hand.” Opening the door to an unoccupied room where chemical reactions are bubbling beneath a janky robotic scaffold, Cronin reveals that the automation of chemistry is already underway, with a goal set on far more than industry efficiency.
Cronin has devoted his career to repositioning chemistry as a 21st-century science. Since arriving at the university as a 29-year-old lecturer in 2002, he has built a 65-member research group, one of the largest in chemistry, funded with a budget close to $5 million per year. Roughly half of these resources have been funneled into the development of a “chemputer” — Cronin’s fanciful name for a computer-driven automated chemistry lab. Beyond the potential for his chemputer to custom-build specialized pharmaceuticals for personalized medicine, Cronin wants to chemputerize his field. He believes it’s the only way to successfully address two of the greatest outstanding challenges in science: to discover the origin of life, and to advance artificial intelligence by making a machine as intelligent as the human brain.
In his mind, these problems are related, because life and intelligence both emerged from prebiotic chemistry. Finding the chemical transitions that led from basic matter to Homo sapiens will require more experiments than what can realistically be achieved by a pair of hands pouring liquids into flasks. The scope of his work is compelling enough that the U.S. Defense Advanced Research Projects Agency (DARPA) supports one of his projects. The Templeton Foundation also awarded a $2.9 million grant to Cronin and several colleagues to figure out how life began. And he’s shrewd enough to know how to supplement this money by simultaneously developing practical applications for his chemputer.
“He’s making huge advances in pharma and all these other areas,” says Arizona State University astrobiologist Sara Imari Walker, one of his principal origin-of-life collaborators. “Something he does very well is strategically leverage other areas to get the fundamental science he wants to get done.”
If Cronin’s ambitions are fulfilled, countless other researchers will augment his research with breakthroughs on their own chemputers. He aspires for chemistry to “witness its own version of Moore’s Law,” the phenomenon in computing where capabilities double about every two years. Although his bombast faces some blowback from others in his field, who question whether automation will bring such a revolution, Cronin isn’t fazed by the doubts. Nor is the chemputer behind us in the lab, ignoring our conversation and single-mindedly assembling a molecule that few human chemists could synthesize by hand.
Chemistry As Carpentry
When Cronin was 8 years old, he ransacked his parents’ house in search of components to build a computer. In order to distract him, and to save the few surviving appliances, his father bought him a chemistry set. Cronin promptly set out to combine it with his scavenged electronic parts. He didn’t have the idea of a chemputer in mind — at least not quite — but he was already beginning the process of combining science and technology in ways that would determine his life’s work as a freewheeling experimentalist-inventor-entrepreneur. “I was always interested in reality,” he says.
That didn’t go over well in school in the eastern England town of Ipswich. The educational system had little tolerance for precociousness, and his teachers especially didn’t like students asking questions the instructors couldn’t answer. “Everyone said I was too stupid to do what I wanted to do,” he recalls. They considered his questioning of their lessons to be an evasive maneuver to avoid the real work of learning by rote. So he became increasingly disruptive in class, while spending his spare time teaching himself the mathematics of relativity. His grades dropped so low that he couldn’t qualify for the exams required for university. His father intervened once again. He paid the registration fee for the entrance tests out of his own pocket, with the understanding that he’d be reimbursed in the unlikely event his son passed. When the results came in, it became obvious that Cronin wasn’t, in fact, dumb. He matriculated at the University of York.
“I spent half my time dreaming about science, and half the time having to do chemistry, which I found really boring,” Cronin says. For the most part, his training in chemistry was akin to learning carpentry, mastering a set of chemical reactions that could be used in succession to construct new molecular structures. “Always I was thinking, ‘What is the minimum object I can assemble to make the most complex one and set off a cascade of events?’ ” he says. In other words, he was less interested in the minutia of chemistry than in what chemistry could create.
During this period, in the 1990s, Cronin also learned the practical skills he would need to support his unconventional career: Earning his Ph.D. and taking a faculty position at the University of Birmingham, he set out to become the best molecular carpenter he could. “I didn’t abandon my philosophical ideas,” he says. “I [just] realized there was no way I was going to be a successful scientist if I didn’t become a successful chemist first.” By the time he moved to Glasgow in 2002, he was able to build pretty much any molecule on demand.
To avoid being bound to benchwork, he started tinkering with simple robotic systems for moving liquids. Combining off-the-shelf hardware, basic open-source robotics and lab equipment, he made machines that could automate his experiments. Cronin finally came into maturity in Glasgow, where he tried to make the machine of his childhood dreams.
Back in the lab, I find Ph.D. student Przemyslaw Frei standing over a 3D printer, watching its nozzles extrude layer after layer of translucent plastic.
“This is the most complex reactor I’ve made,” he tells me. When it’s complete, it will be capable of combining chemicals to synthesize a new pharmaceutical with limited human handling. This integrated “reactionware” will be a streamlined version of the chemputer Cronin has prototyped with racks of Pyrex vessels, the difference between a sports car and go-cart.
If you have the design files to output the reactionware on a 3D printer, and you know the chemical procedure, “it’s foolproof,” Frei says. “This is an asset from a chemist’s point of view.”
Replicability is the basis of all good science, and especially important for a chemist with Cronin’s aspirations. “When you publish a paper, it is your ethical duty to make sure other people can reproduce it,” Cronin says. Replication in the lab has historically been a challenge because chemistry is artisanal. Although rigid in principle, procedures are akin to recipes in practice, often passed down from professor to student, and reliant on subtleties that go unrecorded because they’re habitual and practically unconscious. Do you proceed to the next step when the brew starts bubbling or after the bubbling subsides? It often depends on expertise, says Cronin: “A lot of the things that we do in the lab are not reproducible because our level of expertise is not well declared.” But in Cronin’s lab, the chemputer is the expert. Frei’s reactionware is reliable because automated systems must be told every step programmatically or they fail spectacularly.
Cronin has spent the past several years developing software — which he refers to as a “chempiler” — that automatically compiles each step of every chemistry lab procedure, as well as all the equipment and materials required. The chempiler can extract all of this from the ordinary language of a research paper, and flag places where the paper is ambiguous. After all vagueness is addressed, the chempiler code can run chemicals through the reactionware, or the clunky system of flasks and pipes that Cronin showed me when he first introduced me to the chemputer concept.
Cronin believes any lab could assemble this setup for under $10,000, using his freely available plans. Although the version using glassware is more primitive, he has held onto it out of pragmatism. He wants to provide as many options for researchers as possible, in hopes that the chempiler becomes “a universal programming language for chemistry,” standardized enough for everyone to collaborate. There’s a lot of inertia to overcome. “Chemists are quite grumpy,” he says. To generate excitement about the system, he demonstrated its capability by having the chemputer synthesize the active ingredient in Viagra.
The ability of the chemputer to make drugs on demand has already attracted serious interest from multiple pharmaceutical companies. While medications from aspirin to Viagra are mass-produced in factories, the industry sees an opportunity for chemputers to custom-make small batches of personalized medicine that treat diseases ranging from cancer to cystic fibrosis. DARPA has also expressed interest, and has provided Cronin with funding. The agency is excited about the possibility of making reactionware in the field, meaning that the military could synthesize any medicine or material anywhere by sending a digital file to a 3D printer.
From Cronin’s perspective, the chemputer will make even greater waves in the research lab. “Most chemists spend 90 percent of their time doing known chemistry,” he says. To synthesize the molecule they want to create, they go through numerous preliminary steps, like a chef preparing ingredients for a soufflé (except each step can take weeks and be highly toxic). If the chemputer could serve as sous chef — preparing any known molecule on demand — chemists could focus on innovation. In other words, they wouldn’t be distracted by the gruntwork Cronin had to endure in Birmingham.
Liberating the world’s estimated 200,000 bench chemists is just half of Cronin’s vision. The other half is to automate the discovery process in its own right. Chemists would not lose their jobs, he insists. Instead, research capabilities would be augmented. “If you had an infinite number of [chemicals] and an infinite number of people, you could also run an infinite number of experiments,” he says.
His basic idea is to connect a chemputer with a machine capable of analyzing chemicals instantaneously, add in some artificial intelligence and give the system a target. Then you let it run in a closed loop until it hits the bull’s-eye. The target could be as outlandish as making artificial life, or as practical as finding a drug that treats a disease with minimal side effects.
Other chemists are cautiously optimistic about this vision of automated chemistry. “Lee’s work here is important,” says University of Liverpool chemist Andrew Cooper, one of the pioneers in chemistry automation. He’s especially impressed by the chemputer’s scalability: A chemist could one day seamlessly move from research to production, making valuable new materials in quantity.
Alexander Godfrey, who developed early automation systems at the drug manufacturer Eli Lilly, and who now leads the National Institutes of Health’s automated drug discovery program, is even more invested in Cronin’s concept. He’s planning to build a chemputer of his own that integrates Cronin’s framework and innovations.
Godfrey notes that predictions about an automated future in chemistry have a checkered past. In particular, drug companies spent a lot of money in the ’90s on systems designed to do several experiments in parallel. It was “garbage in, garbage out,” he quips. But he believes this time will likely be different, primarily because AI has matured. Lee’s development will not only “impact drug discovery,” Godfrey says. It will revolutionize materials discovery, from more efficient batteries to more efficient biofuels. “By democratizing this, you’re bringing more ideas to the table [and you get] a more diverse group of thinkers."
Order From Chaos
To discover the origin of life, you might try building a motorized lazy Susan and let it run 24/7. At least that’s the approach taken by Dario Caramelli, a postdoctoral student working just around the bend from Frei in Cronin’s busy laboratory. “[We can run] thousands of experiments per day [because the machine] is always doing all steps,” Caramelli explains, pointing out how the armature above the spinning tabletop deposits chemicals in one dish while washing a second and drying a third. A camera monitors what happens inside. And if any supplies run out, “the robot sends an email.”
This lazy Susan is a variation on the chemputer, specially configured to explore how a random mess of simple chemicals can interact in ways that result in lifelike complexity — essentially a survey of the pathways chemistry might have taken on the way to Darwinian evolution. “What we’re doing is mixing up literally random formulations and putting them in a petri dish and videoing them,” says Cronin.
Image recognition software coupled with artificial intelligence calls out surprises, such as unexpected interactions. (In other setups, the camera is swapped with more sophisticated instruments, like a mass spectrometer.) The system generally operates on a closed loop. Notable behaviors can be automatically iterated to reach higher levels of complexity and more lifelike qualities.
In contrast to the chemputers used to discover new drugs, Cronin doesn’t set a specific target in advance. “I don’t know what I’m looking for,” he admits in a rare moment of humility. Since there’s no record of life’s beginning on Earth, his goal is to explore as many possibilities as he can without making assumptions, an approach to chemistry that’s feasible only with rapid-fire automation.
Walker, the ASU astrobiologist collaborating with Cronin, sees this approach as alluring because any life native to another planet is unlikely to have followed the same path as on Earth; understanding life in general can help researchers to identify it in alien circumstances by broadening the range of targets. From Cronin’s perspective, the creation of artificial life — or even lifelike behavior — is interesting in itself because it supports his other grand ambition: to create a chemical brain.
Cronin isn’t especially impressed with brains. He views intelligence, like life, as nothing more than a chemical phase transition. So to induce intelligent behavior in chemicals, Cronin is betting once again on random messes, and on tipping probability in his favor through the speed and efficiency of automation. “I thought, why not just take a blob of chemicals and connect them to an electrode array?” he says, leading me into a small locked room where he’s trying it out.
The general idea is to expose a gel to electrical patterns until the chemicals self-organize in a way that recognizes this signal, a rudimentary form of the pattern recognition governing animal and human behavior. The technique resembles what computer scientists use to train some AI algorithms, but Cronin is using a material as gelatinous as gray matter. Cronin argues that conventional computer emulation is too simple to ever grow very smart, and that it makes sense to create intelligence the way a real brain does it: with chemistry.
“People used to be worried that I was doing too many different things,” Cronin tells me as he turns out the lights on his future brain. “I said, ‘You don’t understand. I’m doing only one thing actually.’ ” From the origin of life to artificial intelligence, he was asking the question: “How do random chemical systems become information processing?”
I ask him what he means by information processing. He says that he’s talking about all the mind-boggling things that life does, from evolution to high-level decision-making — phenomena you don’t find when you look at the raw chemicals that compose life. He pauses for a moment to let the metaphysical mystery sink in, and then puts it another way: What are we overlooking in the stuff all around us that could be combined to build conscious machines?
Now he’s just waiting for a random mess of chemicals to show him.
Jonathon Keats is a writer based in San Francisco, and author of You Belong to the Universe: Buckminster Fuller and the Future.