The authorities of the largest countries are investing billions in artificial intelligence
The brain from the machine
Andrey Konstantinov (with the participation of Alyona Lesnyak), "Russian Reporter" No. 11-2013
The United States and the European Union are accelerating brain research: governments and major corporations are organizing megaprojects for the construction of artificial brains and "neuromorphic computers". We are talking about huge financial investments. In addition to the "Human Genome", there were no projects of this scale in biology. We decided to find out what they hope for and whether we will get artificial intelligence in the foreseeable future.
The human brain is the most complex device in the known part of the universe. By what magic does this pink lump of mucus give birth to consciousness, mind and the entire subjective world? How can chemical reactions determine our emotional impulses? How can the interaction of atoms turn into love and hate? There is no more intriguing mystery.
Since the time of Descartes, this riddle has put philosophers in a stupor. The philosophers were replaced by researchers – optimistic experimenters looking for a technical solution to any problem. The 1990s were declared the "decade of the brain" by the scientific community of the USA and Europe, but the mystery remained unsolved, although even the great Francis Crick, who deciphered the DNA code, enthusiastically fought over it in the last years of his life. The methods of studying brain activity did not even come close to the level of complexity of the brain – it's like studying a computer by its buzzing, turning its details in your hands.
In 2005, Science magazine, celebrating its 125th anniversary, interviewed the world's leading scientists about the main problems that science will have to solve in the next quarter of a century. The first places were taken by two questions: "What is the universe made of?" and "What is the biological basis of consciousness?".
Eight more years have passed, and it's time for a new attempt, much more ambitious than all the previous ones. After a century of experimentation and improvement of methods, brain science is embarking on a decisive assault on this problem.
"The New Space Race"In a February address to fellow citizens, Barack Obama said: "It's time to reach a level of scientific research and development that has not been seen since the peak of the space race."
It turned out that the US government's budget cuts will not affect advanced scientific developments: "Now is not the time to gut investments in science and innovation." On the contrary, the US president proposes to "invest in great ideas" and fork out for a national project similar to the "Human Genome".
"Every dollar we invested in creating a map of the human genome has returned $140 to our economy – every dollar! And today our scientists are creating a brain map." Obama is referring to the "Brain Activity Map" planned by leading US scientific institutions. This is a ten-year research program that requires a budget of about $ 3 billion ($ 3.8 billion was spent on the Human Genome, and according to a government study, 800 billion has already returned to the economy, and this is just the beginning).
Not only Americans are willing to invest billions in brain research, despite the economic crisis. The belt–tightening countries of the European Union have just chipped in a billion euros for the Human Brain project, whose goal is nothing less than to create a functioning model of the human brain. It turned out that progress needs megaprojects like the Large Hadron Collider like air – without them there are no fundamental breakthroughs, and technology development goes into improving gadgets and new packaging for old ideas.
The newspapers write about the beginning of a "new space race". However, we are not participating in it this time, we have other priorities: the construction of the Adler–Krasnaya Polyana Olympic road alone costs more than American and European projects combined. Well, what are they fighting for?
American BAMLast year, the respected Neuron journal published an article by George Church, a professor of genetics from Harvard, who once led the Human Genome project.
Together with five colleagues, he proposed to start the project "Brain Activity Map" (Brain Activity Map, or simply YOU). The goal is to learn how to register the impulses of all neurons in the brain of an animal. "The essence of our proposal is to create technologies that will allow you to track every pulse of every neuron in the brain of a living organism. Maybe then we will understand what consciousness is, or even different levels of consciousness," explains Church.
The task is clearly more daring and risky than even the "Human Genome". There are almost a million times more synapses (neural connections) in the brain than the number of letters in the genome. And we will have to work not with biomaterials like blood or saliva, but with a living brain, the principles of which we still understand very roughly.
A single neuron. This three-dimensional model was built as part of the Blue Brain project. Photo: ©EPFL/Blue Brain Project.If "genome" means the totality of all the genes of an organism, then the complete map of neural connections is called "connectome" (the suffix "-om" is generally in fashion now).
So far, the only organism for which scientists have been able to make a connectome is a nematode worm a millimeter long with three sexes: boys, girls and hermaphrodites. Nematodes of traditional sexual orientation have a thousand neurons, and hermaphrodites for some reason have only 302. These 302 neurons and seven thousand connections between them were mapped back in 1986 after twelve years of hard work.
Then, based on this scheme, they made a virtual nematode – they programmed a virtual environment for it and tried to make the worm model crawl along the environment model. But there was little sense from such a model: it was not possible to reproduce the simple behavior of the nematode, because we do not know many details of the work of its nervous system.
It is necessary to create not just a connectome, but a functional connectome – a live map of the activity of the nervous system in real time. The authors of the BAM project plan to work out the technology on some part of the drosophila brain that does not exceed 15 thousand neurons. This takes five years, in the next five years they expect to visualize the work of all 135 thousand neurons of the drosophila brain, and then move on to more complex objects consisting of a million neurons: the nervous system of the danio fish and the mouse hippocampus.
If everything works out, they will take on the brains of our smaller brothers: starting with the dwarf multitooth, the smallest and simplest mammal, and ending with primates. The final reference point is the map of the working human brain.
Our brain contains about 100 billion neurons and almost ten thousand times more connections – 1 quadrillion. This is an incredibly intricate network of connections between neural ensembles that connect into large "orchestras" depending on the task that the brain is facing. In addition, they differ from one person to another.
– There is no clear localization of functions in the brain at all, – Svyatoslav Medvedev, director of the Institute of the Human Brain of the Russian Academy of Sciences, explained to us. – Rather, it has interfaces of interaction with the world, such as the mechanism of visual perception. And the brain is engaged in higher forms of activity entirely. As a result, clear mapping of the brain like mapping the human genome is impossible. Some fields of the cerebral cortex are involved in some human activity more, others less, but the brain works as a whole.
There is a risk that the American BAM may turn into a long-term construction. The authors of the project hope that they will be helped to create a map of brain activity by a huge fleet of nanomachines that will penetrate into the brain, sit on each neuron and register its activity. All this sounds on the verge of nonsense, but not for the pragmatists from DARPA, Google and Microsoft, who are already meeting and discussing the project together with representatives of government agencies and private foundations. In March, the "Brain Activity Map" is going to be announced at the state level as a national project, then it is planned to start creating a network of state "brain observatories" for it.
Artificial brain will be created in NeuropolisA few weeks before Obama's speech, the European Commission issued a billion-euro Technology of the Future grant for the Human Brain Project (HBP).
The goal of the venture is to create a "silicon brain", that is, a full–fledged model of the human brain on a supercomputer.
Full–fledged, then, thinking? It seems that even the project director, professor of the Swiss Federal Technical Institute of Lausanne, Henry Markram, is not sure about this.
Henry Markram, Professor at the Swiss Federal Technical Institute, Director of the Human Brain Project.
The cost of this project is more than a billion euros.
For comparison: the entire budget of the RFBR is 0.5 billion euros.
Photo: Valentin Flauraud/Reuters– Will we create something reasonable?
We do not exclude the consciousness of an artificial brain, because surely there is something more in the work of neurons than just impulses. But this is already a philosophy," he said in an interview with a correspondent of "RR".
Since 2005, Markram has been modeling the cerebral cortex using the Blue Gene supercomputer, an advanced descendant of the Deep Blue chess champion. The scientist managed to realistically reproduce a column of 10 thousand neurons – a structural unit of the rat brain cortex. 8192 supercomputer processors were used to simulate the column – each processor was modeled by a neuron. Recently, Markram has already connected a hundred columns in the neural network – a million virtual neurons. Soon he expects to create a real-time model of the rat brain (that is, 1 second of brain activity will be simulated by processors in 1 second).
A fragment of a connectome – a complete description of the connections between brain neurons
Photo: UCLA Human Connectome Project/MGH/dapd/APThe cortex of the human brain also consists of similar columns, there are 20 billion neurons in it, and Markram undertakes to build its virtual analogue in less than a decade.
–Perhaps in ten years a hologram with artificial intelligence will make a report on our successes," Markram tells the TED conference participants, while endless intricacies of virtual neurons flash on the screen.
The problem here is not so much the number of neurons, but the quality of the simulation. In the laboratory of Nobel laureate Gerald Edelman, back in 2008, a neural network model was created from a hundred billion neurons, but much simpler and less similar to real ones.
In order to realistically simulate the human brain, a Neuropolis is being built in Switzerland near Lausanne – an analogue of CERN and Silicon Valley for brain research. Neuropolis, whose first laboratories will open this year, will become the center of the HBP megaproject, which already involves 120 teams of scientists from 90 institutes in 22 countries.
Oh, a brave new neuroworld!The space race brought people to the moon, gave us cellular communications and navigators, accelerated the development of computers, gave impetus to a number of high-tech industries and the entire economy.
And what will the race of artificial brains give?
– Two billion people on the planet suffer from mental disorders, and we are still trying to find medicines for them by trial and error, – says Professor Markram.
He expects that the virtual brain will help to understand the causes of mental illness. It will be possible to simulate the effect of drugs and other methods of treatment. The American president also says this: "He will give YOU a detailed picture of any pathology, and we will understand the causes of Alzheimer's disease, schizophrenia, depression, autism and other brain diseases."
The development of nanorobots, which will be needed in order to track the activity of each neuron, can become a locomotive for medicine and the entire economy. Similar nanomachines will be able to repair the body at the cellular level, and in the future, like bacteria, they will fill our world, making it intelligent and manageable.
The creation of neuromorphic computers simulating the functions of the human brain threatens even more ambitious consequences – it is not for nothing that IBM provided Markram with the Blue Gene supercomputer and sponsored several brain reconstruction projects at once.
Computers with classical architecture can do a lot where it comes to sorting through a large number of options (as in a game of chess), but naive hopes that artificial intelligence can be created simply by increasing their performance have long collapsed.
Since the 1970s, the performance of computers has exceeded forecasts by millions of times, but they have not become smarter. It turned out that programs are much more important than hardware, but stagnation and despondency have been reigning in the field of software development for decades. Translation programs translate just as helplessly as they did twenty years ago; we have not made any progress with image recognition; text editors, browsers and all programs for the average consumer have not undergone any fundamental changes.
To make computers smarter, it is necessary to change the basic principles of their work. For example, to create them based on memristors. "Memristor" is another word from the current lexicon of science. We are talking about electronic analogues of synapses – connections between neurons of the brain (each neuron is connected to the other thousands of synapses). To get an analogue of the brain, memristors must be able to form new connections (this is the essence of learning at the neural level), but we do not yet know exactly how this process occurs in a real brain.
The need for neuromorphic, or, as they are also called, cognitive computers is very great and serves as one of the main engines of the "new space race". The fact is that we are getting worse at coping with the exponentially growing avalanche of data that is crashing down on us. It is increasingly difficult to extract useful information from giant arrays of medical records, instrument readings, and press reports. Today, data science is one of the fastest growing fields of knowledge, and specialists in big data, large arrays of data, are the most in demand on the market.
"We don't need cognitive computers to replace people," says John Kelly, IBM's director of research, "but so that we can survive in a world of gigantic amounts of data. If we don't learn how to extract information from them, these data will bury us, we will start making disastrous decisions.
However, intelligent neurobots will also be able to replace people. If you can create an artificial analogue of the human brain, then you can do something more powerful. And then, in order not to become pets of supercomputers, people will have to create amplifiers for their brains.
Renowned futurist Ray Kurzweil believes that the time is approaching when we will create an artificial neocortex that radically expands the capabilities of our intelligence. You won't have to carry it with you: the brains of people and computers will be connected by a neural network, which will be the next step in the development of neural interfaces that already allow you to control external objects with the power of thought. The time will come for the symbiosis of biological and artificial life. The history of humans as a biological species will end, the history of posthumanism will begin – a species that will create itself.
Where does the mind come from?As always, the upcoming breakthroughs in understanding the work of the brain and the creation of artificial intelligence inspire optimists with hope, and pessimists with fear.
In addition to the uprising of machines, our contemporaries are afraid, of course, that deciphering the neural code will give the special services and other forces of evil unlimited opportunities in mind reading, mind control and zombification of the population. In the United States, where the government is traditionally viewed with suspicion, such fears are especially strong.
Optimists and pessimists are united by the belief that mind and consciousness arise by themselves from the work of a neural network, like "god from a machine" (this expression in antiquity meant an unexpected resolution of a confusing situation, not following in any way from the natural course of events).
"I am my connectome," Sebastian Seung, a professor at the Massachusetts Institute of Technology and one of the leading developers of neural connection maps, said at the TED conference. A lot of data has been accumulated in favor of this hypothesis. For example, back in the 90s, the existence of a "Bill Clinton neuron" was experimentally proven, that is, a neuron that is activated if and only if we remember the ex-president, regardless of whether we read his name, see his image or recall events related to him. When activated, this neuron lowers the threshold of excitation of the associated neural ensemble encoding similar concepts: Hillary, Monica, saxophone, "I inhaled, but did not inhale," etc.
Apparently, there is a neuron in the human brain that is responsible for encoding specific objects, including his grandmother (the theory that predicted the existence of such neurons back in the 60s was called "my grandmother's neuron theory").
But there is also data that does not agree with the idea that all memories and aspects of a person's personality are encoded in the connections of neurons. You can start with the fact that the brain consists not only of neurons, it has ten times more so-called glial cells. Until recently, they were considered "second-class cells" created to feed and repair neurons. But a number of studies have shown that they have a lively dialogue with neurons and with each other using chemical signals that regulate the activity of neurons and the formation of new synapses. If this is true, then most of the brain was simply ignored until recently.
And what about the neurons of the peripheral nervous system, and with all other cells of the body? Why are nerve cells able to generate experience, but other cells are not? After all, neurons are fundamentally no different from them. Not only neural networks, but also unicellular organisms, such as amoeba or infusoria, are capable of simple forms of learning and memorizing information. What do they remember?
Perhaps the secret of the mind is inside the cell, not outside. According to one hypothesis, microtubules, which make up the "skeleton" of the cell, serve as a substrate capable of storing information and processing it inside the cell. The protein tubulin contained in these microtubules is able to change its state, moving into one of several stable positions, and thus record information.
Anyway, modeling a "spherical brain in a vacuum", you can't create a mind: even for a small nematode, together with the brain, you had to model the body and its entire simple environment. Our mind is formed in active interaction with the world, and is not an automatic function of a complex brain.
These arguments are given by skeptics who, like the famous mathematician Roger Penrose, believe that consciousness and mind cannot miraculously arise from the exchange of signals in a network, no matter how complex this network may be.
"People think that consciousness comes from some complex aspect of computational activity," Penrose explains. – I look at this problem completely differently. In my opinion, there is a lot of computational activity going on in the brain, but this is the unconscious. And consciousness, in my opinion, is something fundamentally different. Understanding is not calculation. Something else is going on. I believe in science and believe that everything that happens in our head obeys the same laws that the universe around us obeys. But these laws are not yet fully understood by us. I'm trying to find this gap in our knowledge, it's "something else".
It's difficult to model consciousness if we don't know what it is. But if we don't engage in a scientific assault on the mysteries of the brain, we will never know.
Portal "Eternal youth" http://vechnayamolodost.ru21.03.2013