In the 60s of the 20th century, science fiction dawned. Writers took on the role of futurists. It seemed to everyone that soon robots will replace people in all simple jobs, cars will fly through the air, and on the planet Mars will grow gardens. But no one could retell the appearance of the Internet.
About a computer rat with intelligence
Today, industrial philosophers are trying to guess the direction of industrial development. Following the promulgation by Germany of the national industrial doctrine Industry 4.0, all the world's experts began talking about industrial drivers such as composite materials, Internet of Things (IoT), large data (BIG DATA), etc.
Again, it all comes down to building a flexible, robotic, uninhabited production, which has been talked about for half a century, instead of traditional conveyors. I am sure that development will go this time in a different direction. Without work, they will soon not be working at routine operations, but again "white-collar workers" in offices.
We are already seeing progress, for example, in medicine, primarily medical diagnostics. In order to correctly diagnose data from a computer tomograph, doctors of the highest qualification are required. Such specialists are taught for a long time, but too few are able to become professionals. More recently, researchers have demonstrated that software complexes based on neural networks not only find tumor images better than physicians, but also successfully diagnose on the basis of global knowledge bases. It turned out that the activity of radiologists is rather a predictive identification of interrelations, rather than logical analysis. They can not name the cause of cancer - they only know what it is.
Approximately on the same principle, the latest algorithms for machine translation of tests from different languages, created by Google. We are standing on the threshold when artificial intelligence (AI) will manage the shops, factories, provinces and whole countries more qualitatively and pragmatically.
Google was founded by Larry Page and Sergey Brin. It was registered as a private company on January 4, 1996, and on August 19, 2004, began selling its shares on the stock market. The mission of the company stated from the very beginning was "the organization of world information, ensuring its accessibility and benefits for all", and the company's unofficial slogan: "Do not be evil".
Google manages more than a million servers in data centers around the world and processes more than one billion search queries. The average human brain consists of approximately 100 billion neurons. Each neuron has up to 10 thousand connections with other neurons, that is, synapses of only 100 to 1,000 trillion. We are still far from creating such large-scale neural networks, but the achievements of Google Brain allowed scientists to create artificial neural networks, comparable already with the brain of a rat.
Suppose a computer rats have to guess the cat. The computer intelligence of a rat is not in itself able to do it. However, if you have a million different people who can assemble into billions of variant groups, you can learn to classify the data with incredible accuracy. Your even non-trained electorate is able to look at the unmarked image of a cat and identify it. But the AI unifying them will surpass all expectations for the accuracy of solving this problem with a cat.
When did it all start?
The term "smart production" was first introduced in 2007 by the owner of the engineering company Solver (Solver) from Russia Radislav Birbraer. He, perhaps, was the first in the world to apply FEM (finite element method) in the practice of calculating machines. He began applying FEM to calculate heavy presses. These presses were manufactured in the USSR at the Voronezh plant "Tyazhmehpress" and were supplied to automobile plants of all developed countries.
The finite element method (MCE) is a numerical method for solving partial differential equations, as well as integral equations arising in the solution of problems of applied physics. The method is used to solve the problems of mechanics of a deformed solid body, heat transfer, hydrodynamics and electrodynamics.
MKE, the long-known theoretical method of engineering calculations, Radislav Birbraer was supplemented by models of contact interaction of solids, which created the conditions for the appearance of CAD-modern CAD systems. A few years later, in the US, which began to implement a policy of localization on its territory of the IT industry, graphic displays and visualization programs for engineering calculations were created. Throughout the world, the process of replacing culmination with computers began.
The technical revolution did not start in the production halls, as everyone had expected, but in the most intellectual knot of industrial enterprises - design bureaus and design studios
A new technological revolution
The same thing will happen now. As soon as a sufficiently reliable algorithm appears for revealing non-obvious relationships in one sphere of life, it can be applied almost immediately to another field. A neural network created by Google to translate text can, for example, learn very quickly millions of pages of legal documents, in order to lack the lives of a million of the most qualified lawyers, it can analyze millions of court decisions and business histories. And why after this will be needed millions of officials, managers, economists, financial consultants, real estate agents?
And who changed the world? It did not take millions of programmers. At first, the Google team engaged in the creation of AI on the basis of neural networks were one or two researchers, then three or four, and towards the end of them there were no more than a hundred. And yet, now in the IT industry, fundamental changes are taking place, which neither governments nor ordinary people have yet realized. A small group of scientists from the company Brain demonstrated that there is a desire, it is always possible to change the paradigm of human development in just 9 months. But the wave of transformations will affect colossal institutions like Google, Facebook or Yandex. As soon as computers can learn themselves, the profession of the programmer will also be under threat.
The project is implemented under the management of Google CEO Sundar Pichai, who was once born in Chennai, (India). He received a diploma in metallurgy at the Indian Institute of Technology in the city of Haragpur. He received his Master's degree at Stanford University with a degree in Material Science and Engineering, and his MBA at the Wharton Business School. He moved to Google in 2004. In 2014, he became the head of Google products, and soon after the formation of a new holding company, Google Alphabet Inc was appointed to the post of general director of the corporation.
"Google Brain" began in 2011 as a third-party research project of Google's research fellow Jeff Dean, Google researcher Greg Corradoren and Stanford University professor Andrew Eun. They wanted to fulfill the mission of creating: "artificial neural networks that learn the world through trial and error as babies, as a result, can develop the flexibility inherent in people." Everyone knows that Google's "Translator" service appeared in 2006 and has since become one of Google's most popular services; a month it was used by more than 500 million people, who every day translated for 140 billion words into different languages. And recently the "Translator" has changed: now most of its traffic is handled by AI. According to user estimates, it has improved qualitatively.
Google's decision to switch the "Translator" service to the principles of a neural network was a clear signal to make a breakthrough in machine learning, such a passionate desire has recently covered the entire IT industry. Over the past four years, large companies - Google, Facebook, Apple, Amazon, Microsoft and Chinese firm Baidu, among others - have entered into a battle for talented experts in the field of AI.
With some delay about the incident, it became known to Russian President Vladimir Putin, who in 2017 "fell ill" with the digital economy. Russian Yandex, too, even late for a full circle, entered the race. The race in the field of AI began to resemble a race for the possession of an atomic bomb. Even Mark Zuckerberg, CEO of Facebook, personally participates in his company's attempts to lure the best specialists.
The meaning of the phrase "artificial intelligence" seems obvious, but everyone has always perceived it in different ways. As soon as we automate some function, we devalue the necessary human skill for it before mechanical work. However, artificial intelligence is also an intellect that blindly does not follow instructions. He is endowed with the ability to recognize subtext, interpret meanings. It is a tool designed to solve a wide variety of problems, achieve in the general context of the most incredible set of goals. And it is almost impossible to deceive.
Who else was involved in this project? Jeff Dean, only a senior research fellow, but de facto is the head of Google Brain. Dean was born into a family of medical anthropologist and epidemiologist, and his childhood was everywhere - in Minnesota, Hawaii, Arkansas, Uganda, Somalia and Atlanta. He has been with Google since 1999. Andrew Eun, a young Stanford professor of computer science who worked as a consultant in the company. Greg Corrado and Kuoka Le disciples Andrew Ynoma. It was with their arrival that Google engineers began to call Project Marvin differently: Google Brain (brains).
About robots with eyes ...
In a real brain, the neurons themselves are not important, but the variety of connections between them. The brain can work even in those conditions when there is little or no information. Google Brain was the first major commercial institute to explore the possibilities of modeling the brain. "That moment in evolution, when the animals had eyes, became a serious phase transition, a breakthrough to a higher level of consciousness," Dean once said. "Now computers have eyes. We can create machines based on these computer eyes that can analyze photos. Robots will change dramatically. They will be able to work in an unfamiliar environment over very diverse problems. " Practical applications for them very much.
In the first year of Brain's existence, the experiments on creating a car with the abilities of a one-year-old child - as Dean put it - went fine. Their speech recognition team has changed part of its old system to a neural network, and as a result, the quality of the work has grown in more than 20 years. For a year in the laboratory Brain have generated a mountain of revolutionary ideas. The thing is that Google has finally allocated resources - computer and human - to fill the gaps that have been empty for a long time.
How to find a cat?
To solve the problem of "cat recognition" the Google Brain team developed fast processors. They will be called TPUs (tensor processing units), and the whole point is that they are less accurate than conventional processors (although this is not always obvious). The Google Brain team understood that it was only necessary to create a program capable of distinguishing between basic linguistic sequences, then many different applications could be built on its foundation-for example, tools for composing an automatic response to an e-mail or for reasonably maintaining a conversation. This does not require great accuracy of calculations, but a large number of operations are necessary.
When the first results appeared, the future of millions of secretaries and ordinary journalists day by day creating mountains of information garbage became vague. Moreover, some realized that on the basis of this it is possible to create a machine that was capable of something similar to thinking.
Quoc Le had many experiments. With each iteration the result turned out better. However, the logic of the neural network was unclear. She found interrelations, but for a person they did not always understand the meaning. Sometimes, when the same neural network, which had recently been beautifully found in the photo of cats, suddenly started enthusiastically mixing together pieces of clouds and cars together with foot puffs. Kuoka Le did not consider himself a linguist, but despite the setbacks, he felt the desire to combine his current work with his early work on the chatbot. He suspected that if a neural network can be taught to discover an image, then it can probably also be taught to find meanings in phrases. The scientific work of Quoc Le showed that the translation of the text
Nobody understood why it worked
Kuoka Le relay took Mike Schuster, who was a staff member in Brain. He grew up in the industrial area of West Germany Duisburg, then studied as an electrical engineer, then took up neural networks. In the fall of 2015, Dean sent two more engineers to Shuster, Yonghui Wu and Zhifeng Chen. It took them two months to reproduce the results of Le in the new system. Le occasionally looked to them, but even he did not always understand what they were doing. As Shuster said, "Some fragments just worked, and no one knew why."
However, soon appeared on one of the scientific sites article competing team of Chinese Baidu. It became clear that they are going in the same direction and are breathing in the back of the American international team Brain. The team Brain also released its scientific work. There were a lot of authors - 31 people. The next day, the employees of Brain and the old "translator" team staged a party. By the time the party began, the new Google Translate has already processed 18 million queries.
The translation system with the help of neural networks has finally worked.
It must be admitted that after such a spasmodic growth of computer intelligence, it became somewhat more difficult for people to determine their exceptional place in this world. But this process does not depend on the will of specific people
The new reality arises by itself and produces expansion.
And we must remember that due to the probabilistic nature of the neural networks they are not suitable for every task. There's nothing wrong with having the system make a mistake of 0.5% and sending you to the wrong flight to the airport, but when it comes to nuclear reactors or civil aircraft, for example, we do not want to risk it. Therefore AI created on the basis of a neural network is not yet a real "artificial intelligence". This is only an intermediate result.
What will happen next?
Most likely we will soon witness a new series called "race for artificial intelligence." Note that progress in the project was achieved when an international team began to work on it, the basis of which was not linguists, as it was before, but engineers: metallurgists, electricians, cybernetics, and anthropologists
Pichai Sundararajan once said that the "Translator" exists in part because not everyone can study the Sanskrit as a physicist Robert Oppenheimer to read the Bhagavad-gita in the original. Ancient Sanskrit underlies all languages of the Indo-European group.
Today we observe that the role of the ancient "Sanskrit" is played in the modern world by the "Translator", which makes the world one world.
Lying in the cold desert, Russia from time to time gives the world those who open the way to a higher level of being. Russia, for example, gave the world Tsiolkovsky, who was the father of flights into space. It is possible that the same thing will happen now. Recall that the history of the creation of AI began many years ago, when Radislav Birbraer, a researcher from Russia, applied new FEM algorithms for calculating machines.
If the concept of "smart production" was first introduced by Radislav Birbraer, then the concept of "smart society" was introduced into the scientific practice by a researcher from Russia Gennady Klimov. He studied the plasma welding processes at the Polytechnic Institute in Voronezh, where he got acquainted with Radislav Birbraer. In the 80 years of the twentieth century, they were engaged in the creation of CAD-systems for computer-aided design.
Birbraer became the ideologist of engineering consulting for the creation of smart enterprises. He established the engineering company Solver, which carried out many real projects in various countries for the construction of smart factories. Gennady Klimov at this time implemented several successful media projects in Russia, and for the last 12 years he has been engaged in anthropological studies of the evolution of consciousness. Klimov refers his research to semiotic anthropology, which studies the evolution of codes and signs that our brain operates
Gennady Klimov noted that the ancient books of all four world religions are similar in structure and have a single "crypt" code - as if broken into four parts. Yes, and the design of the Indian Vedas, Torahs, the Bible, the Koran (through the digitization of all chapters) was associated with the technical task for programmers
Investigating the properties of signs and sign systems used in the process of communication, Gennady Klimov came to the conclusion that there is a common cultural code for all mankind, formed by world culture. The possibility of communication between people is possible only within the range of overlapping paradigms of common cultural codes.
In connection with this discovery, it became possible to move to a higher level of organization of neural networks, which can be assembled into a hierarchy of interconnected platforms. The very platforms that formed during the time when our ancestors wandered through the forests in animal packs, survived in the cave age - and up to the present. In the course of interdisciplinary anthropological research, Gennady Klimov for 12 years managed to restore the chronology of human civilization in conditions of changing climate, geography and social norms of behavior.
In the books of Gennady Klimov, the ideas of the Russian researcher Vladimir Vernadsky on the biosphere and the noosphere developed.
Klimov expressed the idea that the device of the planet Earth correlates with the device of the brain. Some global discoveries that give new states to mankind do not happen in an accidental way, but according to quite predictable geographical laws, in certain places on the planet.
This allowed us to come to the understanding that intellect is not only the ability to recognize the images and meanings of texts, as it does in the right hemisphere of the brain. Intelligence is not only an algorithm, which, in fact, has already been implemented by the international team Brain and the Chinese team Baidu. This is only part of the intellect. To create AI, you need to create two more missing parts, except for those created by the teams Brain and Baidu.
Soon a new team of researchers should appear, which will create an algorithm of the left hemisphere of the brain, which works on completely different principles, operating with minimal logical bits (cultural codes) and morphemes - the smallest units of language that have some meaning.
The mind is the ability to reconcile these two sub-realities into a single picture of reality: female imagery (built on neural connections) and male logic (a system of cultural and linguistic codes). But this is the third stage. Wait for the continuation.
author Dmitry Sokolov