Or, on the contrary, he will solve all the problems of humanity.
In early November this year, Google engineer François Schollet issued a manifesto. In it, he described two key problems: scientists and researchers have not agreed on what they mean by artificial intelligence (AI). In addition, no one has yet come up with a way to compare AI between themselves and with human intelligence. What is the situation with its development today? What is it and where is it applied? Let's try to figure it out.
1. Games of the mind
A young Chinese man in a black suit, white shirt, and square-rimmed glasses sat hunched over with his head in his arms and completely lost in intense thought. The owner of the ninth dan in the ancient eastern intellectual game of go has never been in such a difficult position in his career.
Go is a game with deceptively simple rules: two players, a 19 x 19 line field, sets of black and white stones. The players alternately move their stones, trying to surround the opponent's stones with them. Throughout his life, Ke Jie easily defeated any opponent in this game - until May 23, 2017, when the computer program AlphaGo with powerful artificial intelligence created by Google defeated him, winning three games out of three.
Ke Jie has never lost a game of go in his life. Until I ran into AlphaGo.
For a long time, go remained the unconquered frontier for computers. Checkers, chess, card games and strategies like Starcraft - machines consistently and universally took over humans. And only go remained the prerogative of creativity and intuition, uncomputable for mathematical algorithms, because the number of possible combinations on the go board is greater than that of atoms in the Universe. So far, no supercomputer can calculate and enumerate this number.
Promotional video:
For many hours of games, Ke Jie fought the program with all the means available to him. He used conservative, aggressive, defensive and unpredictable tactics, but none of them worked. AlphaGo surrounded the man over and over again and forced him to surrender. Thus fell the last bastion of the superiority of human intelligence over artificial.
In Steven Spielberg's blockbuster IA, the artificial intelligence in a child's body even learned to love. Photo: Still from the movie "Artificial Intelligence".
2. Limited options
The programs managed to achieve such tremendous success by imitating one or another activity of the human intellect, so now it is generally accepted in Russia to define it as any technology or product that leads to a result similar to human intellectual activity and is used to solve applied problems.
The term "AI" was first introduced back in 1956 at a conference at Dartmouth University by computer scientist John McCarthy. He also owns his witty interpretation: "As soon as any artificial intelligence system starts to work well and give results, no one else considers it artificial intelligence."
John McCarthy. Photo: Wikipedia.
This is largely true. Our perception of AI is influenced by films like The Terminator or Robocop, or for the more sophisticated reader, The Matrix or 2001 A Space Odyssey. Because of them, we often combine robots and artificial intelligence into a single whole. It is not right. The robot is just a shell, while it can be anything and does not have to imitate a human body. And AI is a set of algorithms inside a computer inside a robot. A simple example, the software and data that are responsible for the work of Alice's voice assistant from Yandex is artificial intelligence, and the female voice is his personification, and there are no robots in this system.
The second big misconception is that AI, especially those embodied in the body of a robot, can perform any external operations and mental actions that a person is capable of. In the first parts of the famous "Terminator", the T-800 does everything that people do, only faster, better, more efficiently. In reality, even such cool programs as AlphaGo or Watson are not capable of doing anything that they have not been trained in advance, not trained. Kasparov or Ke Jie could immediately give an interview, understanding human speech and formulating answers, call a taxi for themselves and book a table in a restaurant, listen to music and determine its genre, performer and melody names. None of the programs for playing chess, go, quizzes, or any other special function are capable of learning other skills or relearning them. Based on this, AI is divided into two categories: strong (like in movies) and weak.
You come across artificial intelligence every day, as soon as you pick up your smartphone.
3. Remove the mask
How do we know that we have created a strong AI? Probably the only criterion for this is if artificial intelligence is completely indistinguishable from humans in its behavior and abilities. In 1950, the famous British mathematician Alan Turing created a special test for this.
Turing became famous for three things. He broke the codes of the German Enigma encryption machine during the Second World War. He came up with the concept of the simplest computing machine. And he offered a test to establish - can a computer think? To do this, a person consistently communicates with a machine and another person without seeing them. During the dialogues, he must determine - who is his interlocutor? If the participant in the experiment takes the computer for a human or finds it difficult to answer, then the machine is considered to have passed the Turing test.
Alan Turing.
The Turing test is remarkable in that it eliminates the complex philosophical questions of revealing the "consciousness" of robots, replacing them with the criterion for effectively imitating human behavior. For a long time, it was considered the "gold standard", until in recent years modern chatbots and voice assistants have become so sophisticated that they are already very difficult to distinguish from humans. Therefore, computer scientists have complicated it by inventing the Turing meta-test. With its help, a computer is considered intelligent if it has created something that it wants to test for intelligence itself.
4. He's already here
Strong AI is a matter of the distant future. Many skeptics generally have doubts that it can ever be created. However, right now we are living in an era of weak artificial intelligence, while we hardly notice it. Where and when do we intersect with him?
In your smartphone. When you point the lens at your family, and in the viewfinder, smiling faces are immediately marked with squares - this is the Viola-Jones algorithm. It is also used in CCTV cameras to catch criminals, and even in social networks when they offer you to mark yourself on someone else's uploaded photos.
Next, you watch a show on YouTube, a movie on Netflix or another service for subscriptions to video content, listen to music in Yandex. Music
etc. Analyzing your preferences, the system slips other videos, films or compositions. And this is also a weak AI!
Google Translate or DeepL, which will help you translate text from any of the most common and even dead languages like Latin into Russian, are impossible without AI systems. Recognition and recognition of melodies in Shazam, dictated speech in a dictaphone, automatic answering machine in a bank, etc., etc.
But if you think that AI is only about information systems, then you are wrong! The takeoff, landing and control of modern aircraft are almost completely controlled not by pilots, but by artificial intelligence. Anti-lock braking system and fuel injection in cars are also regulated by it. And on the roads of Skolkovo and American highways, completely unmanned self-driving cars with sophisticated computer vision and navigation systems drive around.
Now artificial intelligence trades on exchanges and dispenses money from ATMs, controls air and ground combat drones, diagnoses and performs operations, draws up contracts and lawsuits, removes data from smart meters and predicts the load on heating and power grids.
In fact, our life is already completely unthinkable without weak artificial intelligence.
CONFRONTATION
History of human defeat
1979 - backgammon. The mathematician Hans Berliner came up with an algorithm for assessing the situation on the game board and looking at the game tree several moves ahead to select the optimal continuation. It formed the basis of the BKG 9.8 program, which beat the reigning world champion Luigi Villa with a score of 7: 1.
1997 - chess. The first battle between IBM's Deep Blue computer and reigning world champion Garry Kasparov took place in 1996. Kasparov won confidently (3 wins, 2 draws, 1 defeat). However, just a year later, for a rematch, IBM engineers significantly increased the power of the computer - now it could estimate 200 million positions per second. This time Deep Blue defeated the Russian grandmaster (2 wins, 3 draws, 1 loss) and became the first computer to beat the world's strongest chess player.
2007 - checkers. In Canada, they created a checkers program that cannot lose. At best, if a person plays an error-free game, he will be able to achieve a draw. The algorithm analyzes all possible options for the development of the game and selects the ideal move in each situation.
2011 - game shows. IBM has put on the popular Jeopardy! (in Russia - "Own Game") Watson artificial intelligence system. She left no chance for two legendary champions: Ken Jennings, who set a record unbeaten streak in 74 episodes of the show, and Brad Rutter, who received the biggest win in the history of the show - $ 4.45 million. True, overseas engineers have not yet dared to challenge Anatoly Wasserman.
2017 - poker. In poker, the players' cards are closed, and besides, players can bluff, which greatly complicates the work of artificial intelligence. Even so, DeepStack outplayed 10 out of 11 professional poker players after three thousand games each. And scientists from Carnegie Mellon University (USA) created the Libratus bot. He challenged the pros and beat four of the best players in today's most popular form of poker, Texas Hold'em.
2019 - Starcraft II. Google's Deepmind AlphaStar bot for the world's most popular real-time strategy game, Starcraft II, easily defeated one of the best human players with a 5: 1 score - Pole Grzegorz Komincz, nicknamed MaNa. Starcraft II requires both lightning-fast response to accurately manage hundreds of units on the map and long-term strategic planning. The game contains a lot of important information: the enemy's resource reserves, the size and position of his army, etc. Each of the AlphaStar versions learned all the intricacies and tricks of the game, training in battles with other programs and playing a total of 200 years.
AI CLASSIFICATION
1. Weak Artificial Intelligence is an AI that specializes in one area. He can beat the best chess players in the world, but this is where his possibilities end. Ask him to play your favorite music - and he will fall into a stupor.
2. Strong artificial intelligence - a universal human-level AI, a machine capable of performing the same tasks as a human. The task of creating a strong AI is incomparably more difficult than the task of creating a weak AI, and we have yet to solve it. Such intelligence will be capable of awareness, planning, problem solving, abstract thinking, understanding complex ideas, fast learning, mastering more and more new skills, learning from one's own experience.
3. Technological singularity, or the emergence of an artificial superintelligence. In 1983, science fiction writer Vernon Vinge wrote an essay in which he suggested that in the future humanity awaits a "technological singularity" - a moment in time when the amount of computing power, their combination and the development of software will become so powerful that it will give rise to superintelligence superior to human. Now this idea is supported by the philosopher and futurist Nick Bostrom. He defines superintelligence as "intelligence that surpasses the best minds of humanity in all areas, including scientific ingenuity, common sense and social skills."
However, while humanity is very far from strong AI, but in the development and use of weak AI systems, we have already quite succeeded.
BTW
In 2018, Google presented an audio recording of a conversation at a conference on artificial intelligence, in which a voice assistant phoned a hairdresser and recorded its owner for a haircut. The chatbot so cleverly asked questions, used pauses and interjections, that the administrator on the other end of the line did not understand that it was not a person who was talking to him, but a virtual interlocutor. The conference participants were amazed by the realistic speech behavior of the robot.
HOW DO WE?
The history of artificial intelligence goes back over 60 years. The first attempts to build learning machines were made within the framework of cybernetics. It was based on mathematical models of control processes in living organisms and society. Among the pioneers of cybernetics, there were also many Soviet researchers, such as Viktor Glushkov, Anatoly Kitov, and Alexei Lyapunov.
Among the achievements of the Russians: the construction of the first robotic machines capable of solving simple problems - driving into the light or avoiding obstacles; automation of production and the use of the first - at that time still giant and slow computers - for the needs of the State Planning Commission. The initiatives of the cybernetics were ahead of their time, but due to excessive bureaucratization within Soviet ministries and distrust of everything new, they did not find the application that the pioneers had hoped for. According to the recollections of many contemporaries of these prominent scientists, the introduction of those early artificial intelligence systems was able to prevent the economic collapse of the 1980s and 1990s and pull the USSR out of the crisis.
Unfortunately, now the success of Russian researchers, engineers and developers lags far behind the leading powers of artificial intelligence: the United States and China. Therefore, for research in the field of this end-to-end technology, a special Competence Center was created on the basis of MIPT and the "State Program for the Development of Artificial Intelligence" was approved, supported personally by President Vladimir Putin. According to many world experts, those who can achieve long-term leadership in the development of weak AI technologies and come close to creating a strong one will secure global leadership in the coming decades. Today it is extremely important for us not to miss our chance.
AGAFON SELITRENNIKOV