“What Are Dreams For? Converging Lines of Research Suggest That We Might Be Misunderstanding Something We Do Every Night of Our Lives”, 2023-08-31 ():
[sleep as offline reinforcement learning of motor function] In the late 1990s, a neuroscientist named Mark Blumberg stood in a lab at the University of Iowa watching a litter of sleeping rats. Blumberg was then on the cusp of 40; the rats were newborns, and jerked and spasmed as they slept. Blumberg knew that the animals were fine. He had often seen his dogs twitch [eg. hypnic jerk] their paws while asleep. People, he knew, also twitch during sleep: our muscles contract to make small, sharp movements, and our closed eyes dart from side to side in a phenomenon known as rapid eye movement, or REM. It’s typically during REM sleep that we have our most vivid dreams…Human adults spend only about 2 hours of each night in REM sleep. But fetuses, by the third trimester, are in REM for around 20 hours a day—researchers using ultrasound can see their eyes flitting to and fro—and their whole bodies seem to twitch. When a mother feels her baby kick, it may be because the baby is in REM sleep. Once born, babies continue to spend an unusual amount of time in REM, often sleeping for 16 hours a day and dreaming for 8…The videos attest to the apparent universality of twitching: not only do many animals twitch in REM but they start before they’re born.
Increasingly, these facts struck Blumberg as odd. In adults, dreams are offshoots of waking life: we have experiences, then we dream about them. But a baby in the womb hasn’t had any experiences. Why spend so much time in REM before you have anything to dream about? According to the dominant theory, the rats’ twitching eyes were supposedly looking around at dream scenery. But the rat pups were just days old; their eyelids were still sealed shut, and they’d never seen anything. So why were their eyes—and their whiskers, limbs, and tails—twitching hundreds of thousands of times each day?
Blumberg decided to put the dream-debris theory to the test. He surgically removed the rats’ cortex—the brain region, involved in visual imagery and conscious experience, where dreams were believed to originate—leaving only the brain stem, which controls subconscious bodily functions, intact. The sleeping pups continued to twitch exactly as before. “There was no way that twitching was a by-product of dreams”, Blumberg told me, when we spoke last fall.
Now in his 60s, Blumberg is the chair of the Department of Psychological and Brain Sciences at the University of Iowa. He has spent the past 20 years studying sensorimotor development—the process through which an infant’s brain links up with its body. Twitches had long been overlooked by sensorimotor researchers.
…By the late 1970s, the idea of a total “input-output blockade” between body and brain during REM sleep had emerged.
…After finding that sleep twitches in early development aren’t caused by activity in the cortex, Blumberg increasingly wondered whether it might be the other way around—perhaps the twitches were sending signals to the brain. Hardly anyone had considered this possibility, because it was assumed that the blockade would keep sensations out. It took Blumberg and his team years to build equipment capable of getting clean brain recordings from tiny, wriggling pups, but eventually, they were able to implant electrodes into rat pups’ brains, recording their neural activity while high-speed cameras captured their twitching.
The results were startling. “I could explain it in words, but it might help to see what it looks like”, Blumberg said, pulling up a video on his screen. It showed the front paw of a sleeping rat pup, hanging limp. “We assigned a different sound to each neuron in the brain that we’re recording from”, he explained. When he started the video, the paw began to twitch—and, with each twitch, musical notes resounded from different neurons in the brain. The effect was like a church organ playing underwater; chords rolled then subsided. An electrode readout made the order of events clear: first the pup moved, then the brain responded. Bursts of activity in the sensorimotor cortex, which coordinates movement and sensation, followed the twitches. The body and brain weren’t disconnected. The brain was listening to the body.
In a series of papers, Blumberg articulated his theory that the brain uses REM sleep to “learn” the body. You wouldn’t think that the body is something a brain needs to learn, but we aren’t born with maps of our bodies; we can’t be, because our bodies change by the day, and because the body a fetus ends up becoming might differ from the one encoded in its genome. “Infants must learn about the body they have”, Blumberg told me. “Not the body they were supposed to have.”
As a human fetus, the thinking goes, you have 9 months in a dark womb to figure out your body. If you can identify which motor neurons control which muscles, which body parts connect, and what it feels like to move them in different combinations, you’ll later be able to use your body as a yardstick against which to measure the sensations you encounter outside. It’s easier to sense food in your mouth if you know the feeling of a freely moving tongue; it’s easier to detect a wall in front of you if you know what your extended arm feels like unimpeded. In waking life, we don’t tend to move only a single muscle; even the simple act of swallowing employs some 30 pairs of nerves and muscles working together. Our sleep twitches, by contrast, are exacting and precise; they engage muscles one at a time. Twitches “don’t look anything like waking movements”, Blumberg told me. “They allow you to form discrete connections that otherwise would be impossible.”
…The theory, he pointed out, turned the rationale for REM paralysis on its head: the paralysis isn’t there to stop the twitches but to highlight them. It’s a process that’s most important in infancy, but Blumberg thinks this might continue throughout our lives, as we grow and shrink, suffer injuries and strokes, make new motor memories and learn new skills.
…In 2013, Blumberg published a paper in Current Biology titled “Twitching in Sensorimotor Development from Sleeping Rats to Robots”. In it, he asked, “Can twitching, as a special form of self-generated movement, contribute to a robot’s knowledge about its body and how it works?” As it happened, the idea was already being put to the test. Some years earlier, a team of roboticists including Josh Bongard, now at the University of Vermont, set out, with support from NASA, to create a robot that could adapt after an injury—an ability that would be extremely useful if it should get stuck or damaged on a distant planet. Early in the work, the team was struck by a dilemma. “If you’re caught in a rock slide or something really bad happens, most of the actions you could perform are going to make things worse”, Bongard told me. A stuck robot might be better off not moving—and yet it can’t get out of danger until it figures out what’s happened to it.
The roboticists came up with a clever solution: twitches. When it’s stuck, their 4-legged robot, nicknamed the “Evil Starfish”, moves the mechanical equivalent of one muscle at a time. Input from the twitches is used by its software to create different interpretations of what is happening; the software then orders new twitches that might help disambiguate the scenarios. If the robot finds that it’s suddenly tilting 30° to the left, it might entertain two interpretations: it’s either standing on the side of a crater, or missing its left leg. A slight twitch of the left leg is enough to tell the difference.
In work published in Science, in 2006, the team showed that their Evil Starfish robot could essentially learn to walk from scratch by systematically twitching to map the shape and function of its body. When the team injured it by pulling off its leg, it stopped, twitched, remapped its body, and figured out how to limp. Watching the robot twitch, a fellow-researcher commented that it looked like it was dreaming. The team laughed and thought nothing of it until the fall of 2013, when Bongard met Blumberg when he gave a talk on adaptive robots. Suddenly, the idea of a dreaming robot didn’t seem so far-fetched. “Dreaming is a safe space, a time to try things out and retune or debug your body”, Bongard told me.