Computers have learned to think. Where it leads?
THE SAVIOR HAS COME
Did you know that AI has already saved the life of a cancer patient in Japan? Specialists from the Tokyo Institute of Medical Research treated the patient for acute leukemia. Only the therapy did not help. What to do?
And the doctors took a chance - they asked for help from the IBM Watson supercomputer. The results of the examinations were loaded into the soulless machine and the "start" button was pressed. The machine analyzed the medical histories of 20 million cancer patients, compared the diagnoses and returned the result: the doctors made the wrong diagnosis. This means that the woman was not treated as it should. Things were getting better.
“For the first time in Japan, artificial intelligence came in handy in order to save a patient's life,” admitted Arinobu Tojo, a specialist at the Tokyo Institute.
THE MACHINE THINKS LIKE A MAN
Promotional video:
The vague term "artificial intelligence" is now increasingly understood as an artificial neural network (ANN). It simulates the work of our brain neurons.
Such machines do not need to be manually configured by entering millions of parameters. The neural network will learn by itself! How? Thousands of examples of the correct solution to the problem are loaded into it - the more, the better. Plus, you still need to set the network structure.
If necessary, INS will learn to play chess and will smash the best grandmasters. Do you want to start creating no worse than Pushkin. Just throw a couple of volumes of Alexander Sergeevich into it, and now "I remember a wonderful moment …"
We have already fallen into these networks. A speech recognition system in a smartphone, applications for photo processing, even a weather forecast - all this, wherever you look, is the result of a neural network.
It is neural networks that help to recognize a car from camera records on highways, to identify a person from a photograph. And if necessary, identify the terrorist. And neural networks will also drive a car and replace the surgeon at the operating table …
WAITING FOR NEURON ARMAGEDDON?
The volume of digital information doubles every 18 months. According to IT specialists, by 2020 it will reach 40 trillion gigabytes.
And only neural networks are able to grind such volumes of data that even supercomputers never dreamed of. Will such smart machines enslave lazy humanity?
The greatest minds - physicist Stephen Hawking and Tesla founder Elon Musk - offer scientists and programmers, before it's too late, to define the border beyond which neural networks should not poke their noses and prevent machines from climbing into these areas. Otherwise, sooner or later, neural networks may decide that people are superfluous on this planet.
HOW DO WE?
Putting chaos in order …
“We, too, are now actively using and developing neural networks,” says Vladislav Belyaev, deputy head of the laboratory of neural systems and deep learning at MIPT. - This is done not only by large companies - Yandex and Mail. Ru - but also by small ones. For example, DeepHackLab is engaged in conversational systems and intelligent behavior modeling. Fiztech, Skoltech, Moscow State University can boast of successful projects …
- But this requires powerful supercomputers like IBM Watson …
- We have them. Moscow State University owns the country's most powerful supercomputer, Lomonosov-2. He is one of the thirty best computers in the world. MIPT and the Advanced Research Foundation are planning a project to create artificial neural networks.
- Will neural networks help humanity not to drown in this huge flow of information?
- Sure! The more data for training a neural network, the better the result it gives. At the same time, neural networks are able to work with both structured information and that which is in a chaotic form. The main thing is to learn how to set the correct tasks, select data and build neural network architectures.
- What does your laboratory do?
- The laboratory is headed by the candidate of physical and mathematical sciences Mikhail Burtsev. Our goal is to develop algorithms for recurrent neural networks, that is, in which there is a feedback. We use the research results for the analysis of texts and the construction of dialogue systems.
VIEW FROM 6TH FLOOR
Only half a step left
Alexander MILKUS, editor of the department of education and science
Voice assistants - Siri from Apple, Cortana from Microsoft, Ok Google, you know from whom, and many others (there are a lot of them now) are a classic example of a neural network. The more often you communicate with your computer assistant, the better he understands your voice, the faster he finds information and the wider the options for his answers.
Yes, he does not always understand you exactly. And he does not always respond adequately. But the solution to this most important problem of the 21st century is clearly not far off. As soon as the computer learns not only to respond to requests in monosyllables, but to understand the nuances of human speech, our world will turn upside down.