Despite technical breakthroughs like Elon Musk’s Neuralink, scientists still have no reliable model of how the brain actually works
The 19th-century villa, glass-walled conservatory, and gardens of the Carlsberg Academy in Copenhagen is hallowed ground in the history of science. It was originally the home of J.C. Jacobsen, founder of the Carlsberg brewery. Jacobsen decreed that the property should become “an honorary residence for a man or woman engaged in science, literature, or art.”
From 1932 to 1962, that resident was Niels Bohr, the Nobel Prize-winning Danish physicist who worked out how quantum mechanics determines the structure of atoms. This is where Bohr strolled and conferred with luminaries of science like Albert Einstein, Werner Heisenberg, and J. Robert Oppenheimer, discussing the essential physics problems that provided the groundwork for the nuclear age.
Today, the Carlsberg Foundation maintains the home and gardens for scientific conferences and symposiums. And for three days this past May, over a dozen scientists from around the world gathered there to share pieces to a puzzle as fundamental as the ones that occupied Bohr: how the brain works.
Mainly they talked about how they might begin to figure it out. How is information represented and processed in the spongy, fatty organ? How do the interactions of its 86 billion neurons, which vary greatly in their shape and other physical properties, facilitate reasoning, decision-making, and movements? What can cause these systems to falter, and what does it take for them to recover?
These are questions as big as any human beings have pondered, and the answers so far are sketchy and provisional at best.
“Let’s say we could actually record from 1 million neurons in a brain while it’s operating. You’d get a lot of data, but what would we look for? That is what we have to get some idea of.”
“We are making progress, but it’s a very difficult problem,” says Sara Solla, a neuroscientist at Northwestern University who attended this year’s Copenhagen meeting and a similar one there in 2016. “We are very far from the goal.” Another attendee, György Buzsáki of New York University’s (NYU) School of Medicine, writes in a new book, The Brain From Inside Out, that “neuroscience is still in its infancy,” capable of assembling a multitude of facts but struggling to determine the relationship between them.
The enduring mystery of the brain is evident in the fact that we have no clear pathways to cures for Alzheimer’s disease, schizophrenia, and other crippling neurological disorders. Scientists aren’t even clear why antidepressants work — or why they often don’t. But the challenges posed to the neuroscientists in Copenhagen are a reminder of just how deep the darkness goes, even while the brain has become the busiest scientific frontier of our time.
Billions of dollars are flowing into research and neurotechnology projects like the U.S. Brain Initiative, Europe’s Human Brain Project, and the China Brain Project. Neuroscientists recently revealed they made mice hallucinate by tweaking 20 of their neurons. Elon Musk says Neuralink, a company that he co-founded and has invested $100 million in, is on the verge of threading ultrafine wires into people’s heads to record from at least 1,000 brain cells, a step toward telepathic communication with computers.
And yet, while scientists might be able to hack into parts of the brain with greater and greater skill, without a comprehensive understanding of the entire organ — a true model of the brain, akin to the atomic models Bohr and his peers refined at that Copenhagen villa — will progress come incrementally, remaining dependent on trial and error, and some luck?
That’s an urgent issue for a society plagued by costly and debilitating neurological and psychological diseases. It also challenges any claims that the human brain can be mastered, upgraded, and melded with machines, or maybe even uploaded to servers in an immortal cloud.
There’s no doubt that technologies for peering inside the brains of humans and lab animals have been firming up neuroscientists’ grasp of how aspects of cognition, such as attention and memory, manifest themselves in electrical signals and biochemical changes.
But the descriptions don’t yet add up to something bigger. Explanations for how neurons interact to generate, say, the act of walking down the stairs, don’t necessarily apply to other aspects of experience, like your decision to walk down the stairs in the first place, or your recall of the staircase in your grandmother’s house. There’s no larger theory with predictive power as there is in physics, where the same principles of gravity explain why planets orbit stars and why your pen falls to the floor. “If we look away from single nerve cells — which are partly understood, not fully — nobody understands how the nerve cells work together to produce perceptions or thoughts or voluntary movements,” says Per Roland, a Danish neuroscientist.
Roland organized the Copenhagen workshops in May and in 2016, and in each of those meetings, he asked the visiting neuroscientists to discuss how threads in each of their specific lines of research might overlap and fit together. He stressed that a good model of how the brain works wouldn’t merely be an academic exercise. A theoretical framework, Roland argues, could dramatically boost neuroscience by tying together research that for now is happening in an ad hoc fashion. Without bigger hypotheses to test, neuroscientists aren’t necessarily in a position to generate profound new insights from all the amazing technologies for examining brains. No one wants a Neuralink that leads nowhere. “Let’s say we could actually record from 1 million neurons in a brain while it’s operating,” Roland says. “You’d get a lot of data, but what would we look for? That is what we have to get some idea of.”
Even though there is nothing close to an overarching theory of the brain, Steve Ramirez thinks he and his fellow neuroscientists might finally be on the right track to finding one. So after talking to some of the scientists who were in Copenhagen, I went to see Ramirez in the lab where he is pulling off some incredible feats.
The now 31-year-old Ramirez first drew attention as a graduate student at MIT in 2013, when he and a colleague remotely activated a bad memory in a mouse. When they turned this memory on, the animal froze up in a box it had no reason to fear. When they turned it off, the mouse scurried around the same box without hesitation. Later they got a mouse to show just as much fear of the box without even having to activate a memory the animal already had, because Ramirez and his collaborator had implanted the memory themselves. And then they erased it. They called that study “Project Inception,” after the 2010 Christopher Nolan science-fiction movie.
“Do we have to know every bit of detail of a phenomenon before we fix it? Sometimes we might just trip into the solution.”
Now an assistant professor of neuroscience at Boston University, Ramirez has continued tinkering with mouse memories, with ever-greater subtlety. This past May, Ramirez and his colleagues demonstrated that they could essentially adjust the volume of a mouse’s memory, making positive experiences more memorable and negative ones less so.
Ramirez is self-deprecating — a sign on his desk says “World’s Okayest Boss” — and gregarious. He seems still in awe that it is possible to manipulate brains so effectively. There are six empty bottles of champagne on a shelf high over his desk, remnants of celebrations of big publications or other milestones. When I visited in July, a bottle had been popped the previous night because the National Institutes of Health had awarded his lab team a highly coveted five-year grant.
If you didn’t know that they mess with memories in Ramirez’s lab, you might at first mistake it for some tech startup. Postdocs, graduate students, and even a few undergraduate researchers sit at computers in cubicles. But then you round a bend into a crisp white room where you can see a whole mouse brain in fluid at the bottom of a test tube, and a machine that is like the world’s finest deli slicer — it cuts brains into strips thin enough to go on glass slides that can be viewed under a standard microscope.
Down the hall are several rooms the size of large closets. These are set up for experiments involving optogenetics, which has become an inexpensive and widely used tool in neuroscience since it was developed around 2005 in the Stanford lab of Karl Deisseroth, who also led the recent study where mice were made to hallucinate.
Optogenetics begins with some deft genetic engineering. By injecting a virus into an animal’s brain, scientists selectively alter particular kinds of neurons so that they are sensitive to pulses of light and will glow when they’re active. These neurons can then be activated or deactivated with flashes from a laser plugged into a hole in the mouse’s skull. Meanwhile, the glow makes it possible to map which cells had been active in a particular interval.
When Ramirez and his original collaborator Xu Liu first used this technique to turn memories on and off, they did it by zeroing in on cells in the hippocampus, an area of mammalian brains known to be involved in memory formation and storage. That alone was profound. Even though we accept that all the intangible experiences that make up mental life must be rooted in tangible physical structures in the brain, it was nonetheless unsettling to watch a vital feature of consciousness be orchestrated from the inside.
But this research has a meaning that goes beyond the metaphysical, and it involves brain cells outside the hippocampus. Over the years, scientists have gotten adept at activating individual neurons rather than clusters of them, and at silencing and activating neurons at moments of their choosing. Now the glowing neurons in these experiments are yielding more detailed and more nuanced maps of brain activity. Ramirez hopes that within five years or so, he will be able to produce a complete 3D rendering of memory formation and recall in a mouse. It would show just how cells throughout the brain hold on to a memory. It would also reveal whether and how neurons sync up in meaningfully different ways for positive memories and negative ones.
His ultimate goal is not necessarily to get optogenetics working in human brains — that would require genetically engineering people’s neurons and shining lasers into their heads. Instead, he wants to see how extensively mouse memories involve structures in the brain that are similar in humans, and might play an underappreciated role in memory. If so, it might be possible to create drugs (or revisit existing ones) that target cells in those areas. Such drugs, Ramirez reasons, could restore or enhance someone’s powers of recall. Perhaps, in combination with psychotherapy or other treatments, such drugs could reliably dampen the trauma of bad experiences.
“Perhaps the only principle we know of is that the brain consists of brain cells.”
Could those kinds of breakthroughs happen without progress in the meantime on a holistic theory of how the entire brain works? Ramirez says he’s not sure. On the one hand, it’s possible to repair systems you don’t fully comprehend. “I played Super Nintendo and Nintendo 64 growing up. I don’t understand how sometimes the system didn’t work. But you know how everybody would take the cartridges and blow into the cartridge before putting it back in the Nintendo?” he says. “Or it’s like if your laptop isn’t turning on, or if it freezes and you can’t restart it or anything, you sometimes just close it. You walk away, have a cup of coffee, you come back, and then it starts again. And it’s through no understanding of computer science do you understand what just happened, but somehow you fixed it and it worked.”
It can happen in biology too. Serendipity and trial and error unlocked the world-changing powers of vaccines and antibiotics well before anyone understood how the immune system functioned. Ramirez can cause memories to form or be erased even though neither he nor anyone else has confirmed how memories are stored in the brain. “Do we have to know every bit of detail of a phenomenon before we fix it?” he says. “Sometimes we might just trip into the solution.”
Nonetheless, he agrees that scientists are far from understanding the brain’s organizing principles and that it probably would help to have such a model. He mentions a textbook called Principles of Neural Science. “It should be blank pages,” he says. “Perhaps the only principle we know of is that the brain consists of brain cells.”
Intriguingly, however, he adds that this might be changing, as neuroscientists come to fully appreciate what the brain really is.
Centuries of medical and experimental observations created the impression that many features of human experience derive from specific parts of the brain. After a piece of iron tore through a railroad foreman’s frontal lobe, he became a petulant asshole. After another man had his hippocampus removed, he could no longer form memories. Damage to your visual cortex can blind you.
But research in the past few decades has increasingly revealed how areas of the brain work with each other in multiple ways. Seeing is not just the reception of visual signals from the eyes; it’s tied up with interpreting the images based on other inputs, including memories of similar things you’ve seen in the past. This is why Ramirez’s efforts to map a memory will be so painstaking. Memories engage cells from all over the brain, not merely the hippocampus.
“The brain isn’t a waffle where one square is processing space, another one processes motion, and some other little square processes depression,” says Ramirez. “The brain has had 4 billion years to evolve into this pile of spaghetti that’s interactive in wildly complex ways. Everything is interacting with itself.”
Pushing further on that idea could be revolutionary for neuroscience, according to NYU’s György Buzsáki. At the Copenhagen meeting and in his new book, Buzsáki has argued that neuroscientists have been stuck in a rut that was first carved in 1890.
That’s when William James published The Principles of Psychology, a two-volume reference book in which the philosopher-psychologist characterized what he called “the science of mental life.” James wrote that he was interested in how “feelings, desires, cognitions, reasonings, decisions, and the like” could arise from the brain, and he devoted entire chapters to the nature of qualities such as attention, habits, reasoning, imagination, and perception of time. (From James’s chapter on time: “In hashish-intoxication there is a curious increase in the apparent time-perspective. We utter a sentence, and ere the end is reached the beginning seems already to date from indefinitely long ago.”)
As their science grew out of psychology, neuroscientists of the 20th century took up James’s ideas, trying to pinpoint the neurological mechanisms of the experiences he described. But while James’s categories may have been useful in describing various attributes of consciousness, Buzsáki points out that they are arbitrary: they don’t necessarily describe distinct patterns or states seen in the brain.
And so, Buzsáki argues, neuroscience should not just look “outside in” — taking some stimulus or experience in the world and trying to find neurological correlates inside. The point of his alternative “inside out” approach is to start by looking at patterns of activity across the brain and ask how they generate multiple aspects of human experience.
Buzsáki’s distinction between outside-in and inside-out is helpful for thinking about the prospects for brain research.
“I can do better experiments than Isaac Newton today, but it doesn’t make me Isaac Newton.”
For many neuroscientists, the name of the game is to use new technologies to get more and more fine-grained data, which might make it possible to someday intervene precisely at the cellular level. But the number of neurons recorded in any given experiment is not the metric that would matter most for understanding the brain. What might matter just as much, if not more, is how many areas of the brain are being measured simultaneously.
“We can record many more neurons and the information is better and better and better for the experimenter,” Buzsáki says. “But in order to know what is useful for the brain, you have to also record from the downstream structures and see whether [those signals] are utilized by other neurons. It [wouldn’t be] enough to say ‘I recorded from 5 million neurons in the hippocampus.’ It’s also very important how the neocortex interprets the outcome of those 5 million neurons. Neural information is in the interactions.”
Along those lines, if Elon Musk’s Neuralink can record from 1,000 neurons at once, as the entrepreneur boasted at a recent event, that is “an extraordinary thing for making a brain-machine interface,” says Buzsáki. The neuroscientist marvels at how the company has combined three technologies — flexible electrodes, machinery for injecting them into the brain, and a method for efficiently transmitting data across them. But what else might Neuralink’s product be useful for? It’s hard to tell. Recording from 1,000 neurons or even some higher number wouldn’t necessarily reveal something new about the brain.
Any new tool for peering into the brain opens “a new window,” Buzsáki says. “But that new window is useful only if you look at things a little bit differently. I can do better experiments than Isaac Newton today, but it doesn’t make me Isaac Newton.”
Buzsáki believes the outside-in framework reinforced a misconception that the human brain is passive, somewhat like a computer — that a sensory engine perceives stimuli from the outside world, represents those stimuli in some internal model that classifies the things being perceived as good or bad, and then decides on actions that motor-specific parts of the brain coordinate.
But even the smallest and simplest brains in insects and other creatures — brains too small to have separate sensory and motor centers — can act and react. What do those brains and human brains have in common? They allow their bodies to survive in their environments. Buzsáki argues that brains evolved to meet one main goal: to explore the world and continually learn “the consequences of successful exploratory actions” for use again next time. The difference between human beings and other organisms is that we’re better at it. Crucially, this can be the case even if there is not some decision-making center in the middle of brain activity, Buzsáki says. “Brains do not process information,” he writes. “They create it.”
Although this idea and other emergingbrain theories assume that the organ is not quite set up in the way neuroscience has typically imagined, they might have the counterintuitive benefit of clearing a path toward figuring it out. After all, if the experiences that underlie cognition are all in the service of the brain’s effort to generate and test hypotheses about the world, they might have common mechanisms that can be detected and explained — perhaps even described in equations, the basis of any robust scientific model. And at least some of Buzsáki’s fellow participants in Copenhagen see signs that this is true.
Adrienne Fairhall, a computational neuroscientist at the University of Washington, points out that when scientists measure the network-level behavior of many neurons at once, the patterns are “less complicated than they could be,” given how many cells and synapses are involved. It ought to be possible, she says, to encode such patterns in equations or algorithms like the ones used to describe other complex, dynamic systems, like populations or fluids.
And just as scientists use fluid dynamics to predict how liquids and gases will behave without measuring what every single atom of the liquid or gas is up to, it could be possible to model how large networks of neurons will behave without having to measure every last cell.
Sara Solla of Northwestern, who was a theoretical physicist before getting into neuroscience, says that millions of years of evolution probably allowed for lots of variation in neuronal activity, provided the larger systems worked well enough. “Evolution operates on the level of behavior,” she says. “Evolution is not going to tell neuron №52, `You’re not following the chorus line.” This was borne out in a recent experiment on roundworms that have just 300 neurons. Even in genetically identical roundworms performing the same movements, the activities of their individual neurons varied widely. Different combinations of cellular activity can generate the same higher-level behaviors.
This being the brain, though, these insights raise still more questions. Some networks of brain activity might prove to be discernible and calculable, but then how do they work with each other? Can that be modeled?
It’s a reminder that even with all of its new tools and new data points, neuroscience might be close to something huge and still might not be close at all. You can’t be sure what you’ll hit when you’re feeling around in the dark.
“We are in an era where a lot of dogmas are being challenged, and they’re forcing us to think more broadly,” Fairhall says. “There’s not yet this golden new concept that’s going to pull it all together, although there are, I think, glimmers of them.”