Humanization of the robot - mechanization of man

If we choose artificial intelligence from popular myths, it can turn out to be an extremely promising and useful invention. Man and machine - will this combination create an unforgettable tandem?

After being defeated by the Deep Blue supercomputer in 1997, Garry Kasparov rested, thought it over and... returned to the competition in a new format - in cooperation with the machine as the so-called centaur. Even an average player paired with an average computer can defeat the most advanced chess supercomputer - the combination of human and machine thinking has revolutionized the game. So, having been defeated by the machines, Kasparov decided to enter into an alliance with them, which has a symbolic dimension.

Process blurring the boundaries between machine and human continues for years. We see how modern devices can replace some of the functions of our brain, a good example of which are smartphones or tablets that help people with memory defects. While some detractors say they also turn off many brain functions in people who were previously free of defects... In any case, machine-generated content is increasingly infiltrating human perception - be it visuals, such as digital creations or content in augmented reality, or auditory. , as the voice of artificial intelligence-based digital assistants such as Alexa.

Our world is visibly or invisibly cluttered with "alien" forms of intelligence, algorithms that watch us, talk to us, trade with us, or help us choose clothes and even a life partner on our behalf.

No one seriously claims that there is artificial intelligence equal to human, but many will agree that AI systems are ready to integrate more closely with humans and create from “hybrid”, machine-human systems, using the best from both sides.

AI is getting closer to humans

General artificial intelligence

Scientists Mikhail Lebedev, Ioan Opris and Manuel Casanova from Duke University in North Carolina have been studying the topic of increasing the capabilities of our minds for some time, as we have already talked about in MT. According to them, by 2030, a world in which human intelligence will be enhanced by brain implants will become an everyday reality.

Ray Kurzweil and his predictions immediately come to mind. technological singularity. This famous futurist wrote a long time ago that our brains are very slow compared to the speed at which electronic computers can process data. Despite the unique ability of the human mind to analyze huge amounts of information at the same time, Kurzweil believes that soon the growing computational speed of digital computers will far exceed the capabilities of the brain. He suggests that if scientists can understand how the brain performs chaotic and complex actions, and then organize them for understanding, this will lead to a breakthrough in computing and an artificial intelligence revolution in the direction of so-called general AI. Who is she?

Artificial intelligence is usually divided into two main types: narrow Oraz The total (AGI).

The first we can see around us today, primarily in computers, speech recognition systems, virtual assistants such as Siri in the iPhone, environmental recognition systems installed in autonomous cars, in hotel booking algorithms, in x-ray analysis, marking inappropriate content on the Internet. , learning how to write words on your phone keypad and dozens of other uses.

General artificial intelligence is something else, much more reminiscent of the human mind. It is a flexible form capable of learning anything you can learn from cutting hair to building spreadsheets as well reasoning and conclusions based on data. AGI hasn't been built yet (fortunately, some say), and we know more about it from the movies than from reality. Perfect examples of this are HAL 9000 from “2001. Space Odyssey" or Skynet from the "Terminator" series.

A 2012-2013 survey of four expert groups by AI researchers Vincent S. Muller and philosopher Nick Bostrom showed a 50 percent chance that artificial general intelligence (AGI) would be developed between 2040 and 2050, and by 2075 the probability will increase to 90%. . Experts also predict a higher stage, the so-called artificial superintelligencewhich they define as "an intellect far superior to human knowledge in every field". In their opinion, it will appear thirty years after the achievement of the OGI. Other AI experts say these predictions are too bold. Given our very poor understanding of how the human brain works, skeptics are postponing the emergence of AGI by hundreds of years.

Computer eye HAL 1000

No amnesia

One major barrier to true AGI is the tendency for AI systems to forget what they have learned before attempting to move on to new tasks. For example, an AI system for face recognition will analyze thousands of photographs of people's faces in order to effectively detect them, for example, in a social network. But since learning AI systems don't really understand the meaning of what they're doing, so when we want to teach them to do something else based on what they've already learned, even if it's a fairly similar task (say, emotion recognition in faces), they need to be trained from scratch, from scratch. In addition, after learning the algorithm, we can no longer modify it, improve it otherwise than quantitatively.

For years, scientists have been trying to find a way to solve this problem. If they succeeded, AI systems could learn from a new set of training data without overwriting much of the knowledge they already had in the process.

Irina Higgins of Google DeepMind presented methods at a conference in Prague in August that could eventually break this weakness of current AI. Her team has created an “AI agent” — sort of like an algorithm-driven video game character that can think more creatively than a typical algorithm — able to “imagine” what it encounters in one virtual environment would look like in another. In this way, the neural network will be able to separate the objects it has encountered in the simulated environment from the environment itself and understand them in new configurations or locations. An article on arXiv describes the study of a white suitcase or chair recognition algorithm. Once trained, the algorithm is able to "visualize" them in a completely new virtual world and recognize them when it comes to meeting.

In short, this type of algorithm can tell the difference between what it encounters and what it has seen before - like most people do, but unlike most algorithms. The AI ​​system updates what it knows about the world without having to relearn and relearn everything. Basically, the system is able to transfer and apply existing knowledge in a new environment. Of course, Ms. Higgins's model itself is not AGI yet, but it is an important first step towards more flexible algorithms that do not suffer from machine amnesia.

In honor of stupidity

Mikael Trazzi and Roman V. Yampolsky, researchers from the University of Paris, believe that the answer to the question of the convergence of man and machine is the introduction of artificial intelligence into algorithms also "artificial stupidity". This will also make it safer for us. Of course, artificial general intelligence (AGI) can also become safer by limiting processing power and memory. Scientists, however, understand that a superintelligent computer could, for example, order more power through cloud computing, buying equipment and shipping it, or even being manipulated by a dumb person. Therefore, it is necessary to pollute the future of AGI with human prejudices and cognitive errors.

Researchers consider this quite logical. Humans have clear computational limitations (memory, processing, computation, and "clock speed") and are characterized by cognitive biases. General artificial intelligence is not so limited. Therefore, if it is to be closer to the person, it must be limited in this way.

Trazzi and Yampolsky seem to forget a little that this is a double-edged sword, because countless examples show how dangerous both stupidity and prejudice can be.

Emotions and manners

The idea of ​​mechanical characters with lively, human-like features has long stirred the human imagination. Long before the word "robot", fantasies were created about golems, automata, and friendly (or not) machines embodying both the form and spirit of living beings. Despite the omnipresence of computers, we don't quite feel like we've entered the era of robotics known, for example, from a vision in the Jetsons series. Today, robots can vacuum a house, drive a car, and manage a playlist at a party, but they all leave a lot to be desired in terms of personality.

However, this may change soon. Who knows if more characteristic and campy machines like vector Anki. Instead of focusing on how many practical tasks it can perform, the designers sought to endow the mechanical creation with a "soul". Always on, connected to the cloud, the little robot is able to recognize faces and remember names. He dances to music, responds to touch like an animal, and is stimulated by social interactions. Although he can speak, he will most likely communicate using a combination of body language and simple emotional signs on the display.

In addition, he can do a lot - for example, competently answer questions, play games, predict the weather and even take pictures. Through constant updates, he is constantly learning new skills.

Vector was not designed for refrigeration professionals. And perhaps this is a way to bring people closer to machines, more effective than ambitious programs to integrate the human brain with AI. This is far from the only project of its kind. Prototypes were created for several years assistant robots for the elderly and sickwho find it increasingly difficult to provide adequate care at a reasonable cost. Famous robot pepper, who works for the Japanese company SoftBank, must be able to read human emotions and learn how to interact with people. Ultimately, it is helping around the house and caring for children and the elderly.

The old lady interacts with the Pepper robot

Tool, superintelligence or singularity

In conclusion, it may be noted three main streams in reflections on the development of artificial intelligence and its relationship with humans.

• The first assumes that building an artificial general intelligence (AI) that is equal and similar to a human is generally impossible. impossible or very distant in time. From this perspective, machine learning systems and what we call AI will become more and more perfect, more and more capable of performing their specialized tasks, but never exceeding a certain limit - which does not mean that they will only serve the benefit of humanity. Since it will still be a machine, that is, nothing more than a mechanical tool, it can both help in work and support a person (chips in the brain and other parts of the body), and possibly serve to harm or even kill people.
• The second concept is opportunity. early construction of AGIand then, as a result of the very evolution of machines, to rise artificial superintelligence. This vision is potentially dangerous for a person, because the supermind may consider it an enemy or something unnecessary or harmful. Such predictions do not rule out the possibility that the human race may be needed by machines in the future, although not necessarily as a source of energy, as in The Matrix.
• Finally, we also have the concept of Ray Kurzweil's "singularity", that is, a peculiar integration of humanity with machines. This would give us new possibilities, and machines would be given human AGI, that is, flexible universal intelligence. Following this example, in the long run, the world of machines and people will become indistinguishable.

Types of artificial intelligence

• reactive - specialized, responding to specific situations and performing strictly defined tasks (DeepBlue, AlphaGo).
• With limited memory resources - specialized, using the resources of the information received for decision-making (autonomous car systems, chat bots, voice assistants).
• Gifted with an independent mind - general, understanding human thoughts, feelings, motives and expectations, able to interact without restrictions. It is believed that the first copies will be made in the next stage of AI development.
• self-awareness - in addition to a flexible mind, it also has awareness, i.e. concept of oneself. At the moment, this vision is completely under the sign of literature.