The Future of the Mind
The first approach, simulating the brain using transistors and computers, is forging ahead by reverse engineering the brains of animals in a certain sequence: first a mouse, then a rat, rabbit, and a cat. The Europeans are following the rough trail of evolution, starting with simple brains and working upward. To a computer scientist, the solution is raw computing power—the more, the better. And this means using some of the largest computers on Earth to decipher the brains of mice and men.
Their first target is the brain of a mouse, which is one-thousandth the size of a human brain, containing about one hundred million neurons. The thinking process behind a mouse brain is being analyzed by the IBM Blue Gene computer, located at the Lawrence Livermore National Laboratory in California, where some of the biggest computers in the world are located; they’re used to design hydrogen warheads for the Pentagon. This colossal collection of transistors, chips, and wires contains 147,456 processors with a staggering 150,000 gigabytes of memory. (A typical PC may have one processor and a few gigabytes of memory.)
Progress has been slow but steady. Instead of modeling the entire brain, scientists try to duplicate just the connections between the cortex and the thalamus, where much of brain activity is concentrated. (This means that the sensory connections to the outside world are missing in this simulation.)
In 2006, Dr. Dharmendra Modha of IBM partially simulated the mouse brain in this way with 512 processors. In 2007, his group simulated the rat brain with 2,048 processors. In 2009, the cat brain, with 1.6 billion neurons and nine trillion connections, was simulated with 24,576 processors.
Today, using the full power of the Blue Gene computer, IBM scientists have simulated 4.5 percent of the human brain’s neurons and synapses. To begin a partial simulation of the human brain, one would need 880,000 processors, which might be possible around 2020.
I had a chance to film the Blue Gene computer. To get to the laboratory, I had to go through layers and layers of security, since it is the nation’s premier weapons laboratory, but once you have cleared all the checkpoints, you enter a huge, air-conditioned room housing Blue Gene.
The computer is truly a magnificent piece of hardware. It consists of racks and racks of large black cabinets full of switches and blinking lights, each about eight feet tall and roughly fifteen feet long. As I walked among the cabinets that make up Blue Gene, I wondered what kinds of operations it was performing. Most likely, it was modeling the interior of a proton, calculating the decay of plutonium triggers, simulating the collision of two black holes, and thinking of a mouse, all at once.
Then I was told that even this supercomputer is giving way to the next generation, the Blue Gene/Q Sequoia, which will take computing to a new level. In June 2012, it set the world’s record for the fastest supercomputer. At peak speed, it can perform operations at 20.1 PFLOPS (or 20.1 trillion floating point operations per second). It covers an area of three thousand square feet, and gobbles up electrical energy at the rate of 7.9 megawatts, enough power to light up a small city.
But with all this massive computational firepower concentrated in one computer, is it enough to rival the human brain?
Unfortunately, no.
These computer simulations try only to duplicate the interactions between the cortex and the thalamus. Huge chunks of the brain are therefore missing. Dr. Modha understands the enormity of his project. His ambitious research has allowed him to estimate what it would take to create a working model of the entire human brain, and not just a portion or a pale version of it, complete with all parts of the neocortex and connections to the senses. He envisions using not just a single Blue Gene computer but thousands of them, which would fill up not just a room but an entire city block. The energy consumption would be so great that you would need a thousand-megawatt nuclear power plant to generate all the electricity. And then, to cool off this monstrous computer so it wouldn’t melt, you would need to divert a river and send it through the computer circuits.
It is remarkable that a gigantic, city-size computer is required to simulate a piece of human tissue that weighs three pounds, fits inside your skull, raises your body temperature by only a few degrees, uses twenty watts of power, and needs only a few hamburgers to keep it going.
BUILDING A BRAIN
But perhaps the most ambitious scientist who has joined this campaign is Dr. Henry Markram of the École Polytechnique Fédérale de Lausanne, in Switzerland. He is the driving force behind the Human Brain Project, which has received over a billion dollars of funding from the European Commission. He has spent the last seventeen years of his life trying to decode the brain’s neural wiring. He, too, is using the Blue Gene computer to reverse engineer the brain. At present, his Human Brain Project is running up a bill of $140 million from the European Union, and that represents only a fraction of the computer firepower he will need in the coming decade.
Dr. Markram believes that this is no longer a science project but an engineering endeavor, requiring vast sums of money. He says, “To build this—the supercomputers, the software, the research—we need around one billion dollars. This is not expensive when one considers that the global burden of brain disease will exceed twenty percent of the world gross domestic project very soon.” To him, a billion dollars is nothing, just a pittance compared to the hundreds of billions in bills stemming from Alzheimer’s, Parkinson’s, and other related diseases when the baby boomers retire.
So to Dr. Markram, the solution is one of scale. Throw enough money at the project, and the human brain will emerge. Now that he has won the coveted billion-dollar prize from the European Commission, his dream may become a reality.
He has a ready answer when asked what the average taxpayer will get from this billion-dollar investment. There are three reasons, he says, for embarking on this lonely but expensive quest. First, “It’s essential for us to understand the human brain if we want to get along in society, and I think that it is a key step in evolution. The second reason is, we cannot keep doing animal experimentation forever.… It’s like a Noah’s Ark. It’s like an archive. And the third reason is that there are two billion people on this planet that are affected by mental disorder.…”
To him, it is a scandal that so little is known about mental diseases, which cause so much suffering to millions of people. He says, “There’s not a single neurological disease today in which anybody knows what is malfunctioning in this circuit—which pathway, which synapse, which neuron, which receptor. This is shocking.”
At first, it may sound impossible to complete this project, with so many neurons and so many connections. It seems like a fool’s errand. But these scientists think they have an ace in the hole.
The human genome consists of roughly twenty-three thousand genes, yet it can somehow create the brain, which consists of one hundred billion neurons. It seems to be a mathematical impossibility to create the human brain from our genes, yet it happens every time an embryo is conceived. How can so much information be crammed into something so small?
The answer, Dr. Markram believes, is that nature uses shortcuts. The key to his approach is that certain modules of neurons are repeated over and over again once Mother Nature finds a good template. If you look at microscopic slices of the brain, at first you see nothing but a random tangle of neurons. But upon closer examination, patterns of modules that are repeated over and over appear.
(Modules, in fact, are one reason why it is possible to assemble large skyscrapers so rapidly. Once a single module is designed, it is possible to repeat it endlessly on the assembly line. Then you can rapidly stack them on top of one another to create the skyscraper. Once the paperwork is all signed, an apartment building can be assembled using modules in a few months.)
The key to Dr. Markram’s Blue Brain project is the “neocortical column,” a module that is repeated over and over in the brain. In humans, each column is about two millimeters tall, with a diameter of half a millimeter, and contains sixty thousand neurons. (As a point of comparison, rat neural m
odules contain only ten thousand neurons each.) It took ten years, from 1995 to 2005, for Dr. Markram to map the neurons in such a column and to figure out how it worked. Once that was deciphered, he then went to IBM to create massive iterations of these columns.
He is the eternal optimist. In 2009, at a TED conference, he claimed he could finish the project in ten years. (Most likely, this will be for a stripped-down version of the human brain without any attachment to the other lobes or to the senses.) But he has claimed, “If we build it correctly, it should speak and have an intelligence and behave very much as a human does.”
Dr. Markram is a skilled defender of his work. He has an answer for everything. When critics say that he is treading on forbidden territory, he counters, “As scientists, we need to be not afraid of the truth. We need to understand our brain. It’s natural that people would think that the brain is sacred, that we shouldn’t tamper with it because it may be where the secrets of the soul are. But I think, quite honestly, that if the planet understood how the brain functions, we would resolve conflicts everywhere. Because people would understand how trivial and how deterministic and how controlled conflicts and reactions and misunderstandings are.”
When faced with the final criticism that he is “playing God,” he says, “I think we’re far from playing God. God created the whole universe. We’re just trying to build a little model.”
IS IT REALLY A BRAIN?
Although these scientists claim that their computer simulation of the brain will begin to reach the capability of the human brain by around 2020, the main question is, How realistic is this simulation? Can the cat simulation, for example, catch a mouse? Or play with a ball of yarn?
The answer is no. These computer simulations try to match the sheer power of the neurons firing in the cat brain, but they cannot duplicate the way in which the regions of the brain are hooked together. The IBM simulation is only for the thalamocortical system (i.e., the channel that connects the thalamus to the cortex). The system does not have a physical body, and hence all the complex interactions between the brain and the environment are missing. The brain has no parietal lobe, so it has no sensory or motor connections with the outside world. And even within the thalamocortical system, the basic wiring does not respect the thinking process of a cat. There are no feedback loops and memory circuits for stalking prey or finding a mate. The computerized cat brain is a blank slate, devoid of any memories or instinctual drives. In other words, it cannot catch a mouse.
So even if it is possible to simulate a human brain by around 2020, you will not be able to have a simple conversation with it. Without a parietal lobe, it would be like a blank slate without sensations, devoid of any knowledge of itself, people, and the world around it. Without a temporal lobe, it would not be able to talk. Without a limbic system, it would not have any emotions. In fact, it would have less brain power than a newborn infant.
The challenge of hooking up the brain to the world of sensations, emotions, language, and culture is just beginning.
THE SLICE-AND-DICE APPROACH
The next approach, favored by the Obama administration, is to map the neurons of the brain directly. Instead of using transistors, this approach analyzes the actual neural pathways of the brain. There are several components to it.
One way to proceed is to physically identify each and every neuron and synapse of the brain. (The neurons are usually destroyed by this process.) This is called the anatomical approach. Another path is to decipher the ways in which electrical signals flow across neurons when the brain is performing certain functions. (The latter approach, which stresses identifying the pathways of the living brain, is the one that seems to be favored by the Obama administration.)
The anatomical approach is to take apart the cells of an animal brain, neuron by neuron, using the “slice-and-dice” method. In this way, the full complexity of the environment, the body, and memories are already encoded in the model. Instead of approximating a human brain by assembling a huge number of transistors, these scientists want to identify each neuron of the brain. After that, perhaps each neuron can be simulated by a collection of transistors so that you’d have an exact replica of the human brain, complete with memory, personality, and connection to the senses. Once someone’s brain is fully reversed engineered in this way, you should be able to have an informative conversation with that person, complete with memories and a personality.
No new physics is required to finish the project. Using a device similar to a meat slicer in a delicatessen, Dr. Gerry Rubin of the Howard Hughes Medical Institute has been slicing the brain of a fruit fly. This is not an easy task, since the fruit fly brain is only three hundred micrometers across, a tiny speck compared to the human brain. The fruit fly brain contains about 150,000 neurons. Each slice, which is only fifty-billionths of a meter across, is meticulously photographed with an electron microscope, and the images are fed into a computer. Then a computer program tries to reconstruct the wiring, neuron by neuron. At the present rate, Dr. Rubin will be able to identify every neuron in the fruit fly brain in twenty years.
The snail-like pace is due, in part, to current photographic technology, since a standard scanning microscope operates at about ten million pixels per second. (That is about a third of the resolution achieved by a standard TV screen per second.) The goal is to have an imaging machine that can process ten billion pixels per second, which would be a world record.
The problem of how to store the data pouring in from the microscope is also staggering. Once his project gets up to speed, Rubin expects to scan about a million gigabytes of data per day for just a single fruit fly, so he envisions filling up huge warehouses full of hard drives. On top of that, since every fruit fly brain is slightly different, he has to scan hundreds of fruit fly brains in order to get an accurate approximation of one.
Based on working with the fruit fly brain, how long will it take to eventually slice up the human brain? “In a hundred years, I’d like to know how human consciousness works. The ten- or twenty-year goal is to understand the fruit fly brain,” he says.
This method can be speeded up with several technical advances. One possibility is to use an automated device, so that the tedious process of slicing the brain and analyzing each slide is done by machine. This could rapidly reduce the time for the project. Automation, for example, vastly reduced the cost of the Human Genome Project (although it was budgeted at $3 billion, it was accomplished ahead of time and under budget, which is unheard of in Washington). Another method is to use a large variety of dyes that will tag different neurons and pathways, making them easier to see. An alternative approach would be to create an automated super microscope that can scan neurons one by one with unparalleled detail.
Given that a complete mapping of the brain and all its senses will take up to a hundred years, these scientists feel somewhat like the medieval architects who designed the cathedrals of Europe, knowing that their grandchildren would finally complete the project.
In addition to constructing an anatomical map of the brain, neuron by neuron, there is a parallel effort called the “Human Connectome Project,” which uses brain scans to reconstruct the pathways connecting various regions of the brain.
THE HUMAN CONNECTOME PROJECT
In 2010, the National Institutes of Health announced that it was allocating $30 million, spread out over five years, to a consortium of universities (including Washington University in St. Louis and the University of Minnesota), and a $8.5 million grant over three years to a consortium led by Harvard University, Massachusetts General Hospital, and UCLA. With this level of short-term funding, of course, researchers cannot fully sequence the entire brain, but the funding was meant to jump-start the effort.
Most likely, this effort will be folded into the BRAIN project, which will vastly accelerate this work. The goal is to produce a neuronal map of the human brain’s pathways that will elucidate brain disorders such as autism and schizophrenia. One of the leaders of the Connectome Project
, Dr. Sebastian Seung, says, “Researchers have conjectured that the neurons themselves are healthy, but maybe they are just wired together in an abnormal way. But we’ve never had the technology to test that hypothesis until now.” If these diseases are actually caused by the miswiring of the brain, then the Human Connectome Project may give us an invaluable clue as to how to treat these conditions.
When considering the ultimate goal of imaging the entire human brain, sometimes Dr. Seung despairs of ever finishing this project. He says, “In the seventeenth century, the mathematician and philosopher Blaise Pascal wrote of his dread of the infinite, his feeling of insignificance at contemplating the vast reaches of outer space. And as a scientist, I’m not supposed to talk about my feelings.… I feel curiosity, and I feel wonder, but at times I have also felt despair.” But he and others like him persist, even if their project will take multiple generations to finish. They have reason to hope, since one day automated microscopes will tirelessly take the photographs and artificially intelligent machines will analyze them twenty-four hours a day. But right now, just imaging the human brain with ordinary electron microscopes would consume about one zettabyte of data, which is equivalent to all the data compiled in the world today on the web.
Dr. Seung even invites the public to participate in this great project by visiting a website called EyeWire. There, the average “citizen scientist” can view a mass of neural pathways and is asked to color them in (staying within their boundaries). It’s like a virtual coloring book, except images are of the actual neurons in the retina of an eye, taken by an electron microscope.