Patterns aren’t the only way to inspire coöperation. In 2018, Levin’s team attached a plastic cuff containing progesterone, a hormone that alters the behavior of ion channels, to the stump where a frog had once had a leg. They left the cuff on for twenty-four hours, then observed for about a year. Ordinarily, a frog that’s lost a leg will regrow a cartilaginous spike in its place. But the frogs in the experiment grew paddle-like limbs. About nine months later, little toes started to emerge. Levin thinks that, eventually, the same kind of cuff could be used on humans; you might wear one for a few months, long enough to persuade your body to restart its growth. (Ideally, researchers would find a way to speed development, too; otherwise, you’d be stuck with a tiny arm for years.)
Levin was wary of showing me any mouse experiments. He has grown tired of hearing his work compared to the sinister alchemy described in “Frankenstein.” “That story is about scientific irresponsibility,” he said. Although his research is in many ways unusual, it is ordinary in its treatment of animals—by some estimates, American researchers experiment on more than twenty-five million a year. “I get two types of e-mails and phone calls,” Levin told me. “Some of the people call and say, ‘How dare you do these things?’ for various reasons—animal rights, playing God, whatever. And then most call and they say, ‘What the fuck is taking you so long?’ ” From time to time, Levin receives a call from a would-be volunteer. “I’m going to come down to your lab,” he recalled one of them saying, “and I’ll be your guinea pig. I want my foot back.”
None of the developmental biologists I spoke with expressed any doubt that we would someday be able to regrow human limbs. They disagreed only about how long it would take us to get there, and about how, exactly, regrowth would work. Other projects explore growing body parts in labs for transplantation; 3-D-printing them whole, using tissue cells; flipping genetic switches (“master regulators”); or injecting stem cells into residual limbs. The solution may eventually involve a medley of techniques.
Levin’s vision isn’t confined to limb regrowth; he’s interested in many other forms of morphogenesis, or tissue formation, and in how they can be modelled using computers. He led me down the hall to a room where an elaborate, waist-high machine glowed. The device consisted of twelve petri dishes suspended above an array of lights and cameras, which were hooked up to a cluster of high-powered computers. He explained that the system was designed to measure tadpole and planarian I.Q.
In a study published in 2018, Levin’s team bathed frog embryos in nicotine. As they expected, the frogs exhibited a range of neural deformities, including missing forebrains. The researchers then used a piece of software called BETSE—the BioElectric Tissue Simulation Engine—that a member of the Allen Center, Alexis Pietak, had built. In this virtual world, they applied various drugs and observed their effects on both bioelectric signalling and brain development, hoping to find an intervention that would reverse the nicotine’s damage. The software “made a prediction that one specific type of ion channel can be exploited for just such an effect,” Levin said. The team tried the drug on real embryos that had been damaged by nicotine, and found that their brains rearranged themselves into the proper shape. The software, the researchers wrote, had allowed for “a complete rescue of brain morphology.”
The I.Q. machine gave them another way to measure the extent of the rescue. Inside it, colored L.E.D.s illuminate petri dishes from below, dividing them into zones of red and blue; when a grown tadpole ventures into the red, it receives a brief shock. Levin found that normal tadpoles uniformly learned to avoid the red zones, while those that had been exposed to nicotine learned to do so only twelve per cent of the time. But those treated with the bioelectricity-recalibrating drug learned eighty-five per cent of the time. Their I.Q.s recovered.
Researchers disagree about the role that bioelectricity plays in morphogenesis. Laura Borodinsky, a biologist who studies development and regeneration at the University of California, Davis, told me that “there are many things that we still need to discover” about how the process works, including “how the genetic program and the bioelectrical signals are intermingled.” Tom Kornberg, a biochemist at the University of California, San Francisco, studies another intercellular system that is similar to bioelectricity; it consists of morphogens, special proteins that cells release in order to communicate with one another. Kornberg’s lab investigates how morphogens move among cells and tell them what to do. “What is the vocabulary? What’s the language?” Kornberg said, in reference to morphogenesis. There is probably more than one.
Tabin, Levin’s former adviser and the chair of genetics at Harvard Medical School, told me that he is “agnostic” about how bioelectricity should be understood. Levin describes bioelectricity as a “code.” But, Tabin said, “there’s a difference between being a trigger to initiate morphogenesis versus storing information in the form of a code.” He offered an analogy. “Electricity is required to run my vacuum cleaner,” he said. “It doesn’t mean there’s necessarily an electric code for vacuuming.” The current flowing through the outlet isn’t telling the vacuum what to do. It’s just turning it on.
Levin thinks that bioelectricity is more complex than that. The right bioelectrical signal can transform a Dustbuster into a Dyson—or a tail into a head. Tweaking the signal produces highly specific outcomes—a head that’s spiky, tubular, or hat-shaped—without the need to adjust individual genes, ion channels, or cells. “You can hack the system to make the changes,” Levin said. “Currently, there’s no competing technology that can do these things.”
Levin’s work has philosophical dimensions. Recently, he watched “Ex Machina”—a sci-fi film, directed by Alex Garland, in which a young programmer is introduced to Ava, a robot created by his tech-mogul boss. Unnerved by how beguilingly realistic Ava is, the hero slices his own arm open in search of wires. Since childhood, Levin, too, has wondered what we are made of; having become a father himself, he enjoys talking about such questions with his sons, who are now teen-agers. Once, when his older son was six or seven, Levin asked him how a person could be sure that he hadn’t been created mere seconds ago, and provided with a set of implanted memories. “I didn’t really think about what the consequences for a kid might be,” Levin said, laughing and a little embarrassed. “He was upset for about a week.”
Our intuitions tell us that it would be bad to be a machine, or a group of machines, but Levin’s work suggests precisely this reality. In his world, we’re robots all the way down. A bioelectrical signal may be able to conjure an eye out of a stomach, but eye-making instructions are contained neither in the cells’ genome nor in the signal. Instead, both collectively and individually, the cells exercise a degree of independence during the construction process.
The philosopher Daniel Dennett, who is Levin’s colleague at Tufts, has long argued that we shouldn’t distinguish too sharply between the sovereign, self-determining mind and the brute body. When we spoke, Dennett, who has become one of Levin’s collaborators, was in bed at a Maine hospital, where he was recovering from hip surgery. “I find it very comforting to reflect on the fact that billions of little agents are working 24/7 to restore my muscles, heal my wounds, strengthen my legs,” he said.
In our discussion of Levin’s work, Dennett asked me to imagine playing chess against a computer. He told me that there were a few ways I could look at my opponent. I could regard it as a metal box filled with circuits; I could see it as a piece of software, and inspect its code; and I could relate to it as a player, analyzing its moves. In reality, of course, a chess computer offers more than three levels of explanation. The body allows more still: genetics, biophysics, biochemistry, bioelectricity, biomechanics, anatomy, psychology, and finer gradations in between, all these levels acting together, each playing an integral role. Levin doesn’t claim to understand the entire system, nor does he maintain that bioelectricity is the only important level. It’s just one where he’s found some leverage. He likens revising an organism’s body through bioelectric stimulation to launching software applications. “When you want to switch from Photoshop to Microsoft Word, you don’t get out your soldering iron,” he said.
In modifying the body, Levin is more whisperer than micromanager; he makes suggestions, then lets the cells talk among themselves. “Michael has these brilliant examples of how individual cells communicate with each other,” Dennett said. But the reverse is also true: when communication breaks down, cells can go haywire. Consider cancer, Levin said. It can be created by genetic damage, but also by disruptions in bioelectric voltage. In an experiment reported in 2016, Levin’s team injected cancer-causing mRNA into frog embryos, and found that injected areas first lost their electrical polarity, then developed tumor-like growths. When the researchers counteracted the depolarization, some of the tumors disappeared. In Levin’s terms, the cancer cells had lost the thread of the wider conversation, and begun to reproduce aimlessly, without coöperating with their neighbors. Once communications had been restored, they were able to make good decisions again.
Having built radios as a kid, Levin now hopes to assemble bodies from first principles. His ultimate goal is to build what he calls an “anatomical compiler”—a biological-design program in which users can draw the limbs or organs they want; the software would tell them where and how to modify an organism’s bioelectric gradients. “You would say, ‘Well, basically like a frog, but I’d like six legs—and I’d like a propeller over here,’ ” he explained. Such a system could fix birth defects, or allow the creation of new biological shapes that haven’t evolved in nature. With funding from DARPA—a federal research agency contained within the Department of Defense—he is exploring a related possibility: building machines made from animal cells. Recently, Josh Bongard, a computer scientist at the University of Vermont, designed a computer model in which small robotic cubes connect, creating microrobots that might someday clean up toxic waste or perform microsurgery. Levin took stem and cardiac cells from frogs and sculpted them into blobs that approximated the robot designs; they began working together, matching the simulations. Bongard likened Levin to a magician pulling rabbits out of a hat. “After a while, you start asking not just what’s in the hat,” he said, “but how deep does the inside of that hat go?”
On a warm afternoon, Levin and I drove out to Middlesex Fells Reservation—a twenty-six-hundred-acre state park with more than a hundred miles of trails. We set out through the woods along Spot Pond, a large reservoir where people sail and kayak in the summer. As we walked, our bodies worked up a light sweat. Occasionally, Levin stopped to wonder at fungi clinging to a tree trunk, or to look under a rock for creepy crawlies. Spotting an ant, he recalled trying to feed ants as a child and being surprised at their stubbornness. He noted that planaria can have different personalities—even clones of the same worm. He interrupted his comments on neural decoding to study a plant. “Look at the colors on these berries,” he said. “What the hell? I’ve never seen that before. It looks almost like candy. Let me get a picture of this.”
I jokingly asked Levin if, when looking at nature, he saw computer code raining down, as in “The Matrix.” “That’s a funny question,” he said. “I do not see the Matrix code, but I’m often taking pictures or kayaking or something, and thinking about this stuff.” I asked him if he saw squirrels and trees differently from the way others do. Not a squirrel, he said, because everyone recognizes it as a cognitive agent—a system with beliefs and desires. But a cell or a plant, for sure.
“I look everywhere, and I ask the question What’s the cognitive nature of this system? What’s it like to be a—” He paused. “What’s your sensory world like, what decisions are you making, what memories do you have, if any? What predictions do you make? Do you anticipate future events? Slime molds can anticipate regular stimuli. I look for cognition everywhere. In some places you don’t find it, and that’s fine, but I think I see it broader than many people.”
We stopped to look at a log and found a red splotch that appeared to be a slime mold.
“I don’t know what it actually is,” Levin said. “I’m not much of a zoologist.”
Bending down, he peeled off some bark: a second splotch. Researchers have found that, if a slime mold learns something and then crawls over and touches another mold, it can pass on its memory; in 2016, a pair of French scientists showed how one mold could teach another to find some hard-to-reach food through a gooey mind meld.
“That, I think about all the time,” Levin said. “What does it mean to encode information in a way that, almost like a brain transplant, you can literally give it to another creature?”
We left the log and continued on. Lichen spotted the rocks, and chipmunks chattered in the trees. There was electricity all around us. ♦