Smart machines have learned to bluff, outplay professionals in chess or Go, translate and recognize the human voice. Every week we learn about more and more exploits of computer programs that already know how to make medical diagnoses, draw as well as Rembrandt, sing or generate text. Should a person be afraid of artificial intelligence?
Artificial intelligence (AI) has become a really hot topic in the last few years. Scientists associate this with the rapid development of neural networks (one of the areas of research in the field of artificial intelligence), which, in turn, became possible with the advent of powerful computers.
"Since the early 2010s, influenced by the impressive success achieved by the use of multilayer neural networks (primarily convolutional and recurrent), this area has attracted serious attention from both scientists and engineers and from investors," comments the author one of the Russian chess programs, machine learning specialist Sergey Markov.
The scientific community can argue about the timing of the emergence of smart machines, but they agree on one thing: the development of technology will have an unconditional impact on society, economy and relations between people in the future. Already, there are calls to consider the ethical principles of developing artificial intelligence, making sure that artificial intelligence is developing in a direction that is safe for humans.
Economic effect and impact on the labor market
Science fiction and Hollywood have shaped the concept of "artificial intelligence" as the next form of life on the planet to enslave humanity in the Matrix or organize a nuclear Doomsday for it. The survivors will be finished off by the Terminator.
In fact, despite recent advances in artificial intelligence, intelligent machine beings are still a long way off, scientists and specialists admit. Both those and others, however, advise to pay attention to some aspects right now.
Promotional video:
According to the research organization McKinsey Global Institute, in the next ten years, new technologies will radically change the labor market on the planet, which will save about $ 50 trillion.
The changes will affect hundreds of millions of jobs. More and more people will shift some of their job tasks and many routine tasks to the machine, which will allow them to focus on creative work.
"From a certain point of view, humanity as a whole has an important and interesting task - to develop for each individual individual much faster than humanity develops artificial intelligence systems," said Grigory Bakunov, expert, director of technology distribution at Yandex.
But along with automation, less qualified personnel will inevitably suffer, and now it is necessary to think about how to protect them, retrain and prepare them for a new life.
Practice shows that not only blue collars, but also knowledge workers can suffer. A few days ago, Goldman Sachs replaced a team of 600 traders with two people and automated algorithmic trading programs, for which 200 programmers were hired to service.
Artificial intelligence in itself is not identical with the automation of processes, but the development of AI will lead to the fact that more and more tasks will be within the power of a computer program.
Among the options for solving the problem of human displacement by the machine in the labor market, as pointed out by Allison Dutman, program coordinator of the Foresight Institute, a non-profit organization based in Silicon Valley to promote new technologies, is the introduction of the concept of "universal basic income" that would be received by everyone regardless of income and employment levels. Such income would be funded by the so-called innovative Land Value Tax, the introduction of which is now being actively discussed in Silicon Valley.
Is artificial intelligence a person?
Is a robotic system a person? Can a smart computer vote? What gender is he? MEPs are already discussing the relationship between man and a smart machine, wondering whether future robots should be given the status of an "electronic personality."
As Dutmann points out, people are reluctant to share their rights with those they do not understand, which means they will resist the “humanization” of AI.
“Taking into account how long humanity has gone to give equal rights to all people, regardless of their skin color, race or gender, we can already assume that they do not immediately recognize machines as equal.
Along with ethical ones, legal subtleties also appear: who will take responsibility in the event of an accident of an unmanned vehicle or a breakdown of a smart medical device - and moral questions: is it worth developing an unmanned weapon capable of acting without a person's knowledge?
The third ethical conundrum is discussed more often than others and worries humanity much more: what theoretically can a superintelligence, a real smart machine, do to humanity?
Teach AI ethics
AI development experts admit: even if not in the next 20-30 years, humanity will still live to see the emergence of real artificial intelligence, which will be smarter than its creator.
"The last bastion will be taken when the so-called" strong AI "(Strong AI, Artificial General Intelligence) is created, that is, an AI that will be able to solve an indefinitely wide range of intellectual tasks," says Sergey Markov.
And most importantly, such an AI will be able to think independently.
Many institutions, including the Future of Life Institute, Foresight Institute, Future of Humanity Institute, OpenAI, and others, are engaged in researching the threats posed by AI, as well as the ethical issues associated with new technologies.
The decision of Allison Dutman of the Foresight Institute is to let the computer read all scientific literature, scientific articles that raise the issue of ethics, and make this information the basis for his future decisions.
What are Neural Networks and what is their future?
Most experts associate progress in the development of AI with the development of neural networks.
Neural networks are one of the lines of research in the field of artificial intelligence, based on modeling the biological processes that occur in the human brain.
It is to them that we owe the appearance of impressive results in speech and image recognition, medical diagnoses, text translation and image creation, speech generation and musical composition.
Today, as experts admit, neural networks are recognized as one of the best machine learning algorithms, and solutions based on them show the most outstanding results at the moment.
This is despite the fact that modern neural networks are one and a half thousand times simpler than the rat brain.
“So far, the neural networks we create are relatively small compared to, say, the neural network of the human brain, and, in addition, they are a very simplified analogue of natural neural networks. Therefore, for now, with the help of neural networks, we solve mainly purely applied problems,”says Sergey Markov.
Now specialized processors are being created for training such networks (the so-called neuromorphic processors), which will increase the speed of computations by several orders of magnitude.
The developers are not only busy now with increasing the number of neurons in the network, but also changing the design of the networks. “Complex systems of network configuration are the ones with which the largest number of experiments are being carried out,” says Grigory Bakunov.
And the fact that such systems have become relatively accessible to a large number of ordinary developers has led to the emergence of startups experimenting with neural networks, for example Prisma (the application allows you to process photographs, turning them into stylizations of paintings by famous artists) and Mubert (online composer of electronic music) …
What awaits us in the near future
Nick Lane, professor at University College London and Principal Scientist Nokia Bell Labs, predicts that even more “smart things” will surround a person. They will become smaller and more efficient.
The professor gives an example: if earlier the sensor built into the wall could only realize that someone passed by, then in the future he will not only know who exactly passed, but also how the person behaves, does not need anything, does not represent whether he is a threat to himself or others.
A button-sized sensor can alert a person in case of danger.
Grigory Bakunov from Yandex also agrees with the professor: "In the near future, we will see a boom in the development of narrow artificial intelligences that help solve one rather simple problem, but which they will solve even better than humans."
A voice recognition system, for example, already recognizes some voice commands and addresses better than a human.
“The most likely path for the development of our civilization is the path of synthesis of man and machine: a stick, clothing, car, mobile phone, pacemaker or cochlear implant - as we evolve, our instruments more and more resemble an extension of our bodies. Tomorrow machines will be able to receive mental commands from a person, perceive mentally formed visual images, transmit information directly to the brain - such projects already exist outside the walls of the most advanced technological laboratories,”sums up Sergey Markov.
Do you still need journalists?
The Financial Times ran a rather risky experiment last year, simultaneously challenging to write a text to its profile correspondent and a smart program named Emma. The editor of the Financial Times then had to read both articles and guess which of the two notes the journalist was behind and which computer.
Before this peculiar "crash test" the Financial Times correspondent admitted: “I think the program will definitely cope with the task faster than me. But I hope I can do it better."
And so it happened: Emma really turned out to be faster - the program generated a text based on statistics on the unemployment rate in Britain in 12 minutes. It took the journalist 35 minutes. And, as she herself later admitted, Emma exceeded her expectations. The program not only skillfully handled the facts, but also put the news in context, suggesting how a possible "Brexit" (it was in May 2016, before the referendum on Britain's exit from the EU) could change the situation.
But Emma did something worse than a journalist. “Emma's article was written in a slightly more clumsy language. But most importantly, it had a lot of numbers, - the FT editor admitted. "And, perhaps, the main thing that we are trying to do here is to choose only really important numbers."
Emma is a product of a startup called Stealth. The company says that Emma has a team of live helpers, but they say that everything she writes or does is a product of her “brains”.
And yet - should we be afraid of AI?
Many, if not all, of the people involved in the development of artificial intelligence systems today readily admit that AI will not soon make humans unnecessary. Precisely because artificial intelligence is not yet that smart. The main thing that he lacks today is the autonomous ability to think.
“Now you shouldn't be afraid of AI in any form. You can wait 30-40 years for some radical changes to take place,”Bakunov said.
But something is already happening: the line is slowly blurring between the work or task performed by a person and the task performed by a machine. As experts explain, even now it is sometimes difficult to understand who sits inside the system - a person or a machine.
“There are no criteria when we will be able to understand that consciousness has arisen inside the machine,” Bakunov asks.
Man will become a paper clip?
A well-known adherent of the alarmist point of view, in the framework of which the horrors described in the Terminator can become reality, philosopher Nick Bostrom, holds the opinion that AI that has reached the intellectual level of a person will be able to destroy humanity.
Bostrom explains with paper clips: you challenge artificial intelligence to make paper clips as big and better as possible. AI at some point realizes that a person is a threat, because it can turn off the computer, which would be contrary to the task of making as many paper clips as possible. In case the human is not a threat, the AI decides that human bodies are made of atoms that can be used to make great paper clips. The result is that the computer will drive humanity to paper clips.
This scenario seems like an exaggeration to many. According to Sergei Markov, for example, "the high efficiency of achieving an absurd goal is incompatible with the absurdity of this goal - roughly speaking, an AI capable of transforming the whole world into paper clips will inevitably be smart enough to abandon such a goal."
Artificial intelligence is like a goldfish
Margaret Boden, a British specialist in artificial intelligence and professor of cognitive science at the University of Sussex, is skeptical about the imminent advent of smart machines.
The professor gives the example of a "goldfish", when in exchange for freedom a fisherman makes three wishes. One of the desires is to return his son from the war, the second is 50 thousand dollars, and the third is the opportunity to make another wish the next morning.
That same evening there was a knock on the fisherman's house. The son was returned from the war - in a coffin. The fisherman received insurance for 50 thousand dollars.
“Replace the fish with AI in this parable, and everything will become clear,” explains Boden. "Oh, yes, the next day the fisherman took advantage of the third wish and - canceled the previous two."
Is it possible to transfer consciousness into a car?
Sergey Markov:
“If we talk about the possibility of full transfer of consciousness, then the modern predecessors of such a technology of the future are projects like Blue Brain, aimed at creating functional electronic analogs of the brain, as well as projects aimed at creating brain-machine interfaces (BCI) - devices for prosthetics of lost vision, hearing, replacement of lost limbs, even parts of the brain.
Optogenetics is a very interesting and promising area (in principle, to interface the brain and machine, one can change not only machines, but also the nervous tissue itself, creating artificial photoreceptors in it).
When a wide range of engineering problems is solved within the framework of such private projects, I think that the task of transferring consciousness will become quite solvable. Already, dreamers are proposing hypothetical schemes for implementing such a project.
For example, Jan Korchmaryuk, who at one time proposed the name “Setleretics” for the direction of research related to the transfer of consciousness, believes that the most promising scheme is the use of specialized nanorobots implanted in the neurons of the human brain. However, to successfully implement such a scheme, it is necessary to solve a number of complex engineering problems."
Ksenia Gogitidze