In science fiction novels of the 1960s, artificial intelligence emerged as a hero. In books, computers not only communicated with people in ordinary natural language and made difficult decisions, but also recognized themselves as individuals. Will this remain an eternal dream, or will computers sooner or later be able to catch up with humans?
Will computers be able to think like humans? This is an exciting and very interesting question, and the more we study it, the more we learn about ourselves and the processes of our thinking. Despite the uniqueness of human thinking, computers can greatly outperform humans in certain tasks. Few of us can multiply two decimal numbers in our heads, beat the world chess champion, or even find the best route through a traffic-jammed city. But when it comes to human-computer interaction, things are far from brilliant. Not to mention the issues that require human perception and intuition for their solution - here computers can be completely useless.
Ability to learn
Computers have tremendous computing power, but they have no human feelings and emotions, no human sensibility. This is the main fundamental difference between a computer and a human. The difference lies not at the level of the mind, but at the level of feelings and emotions, which precisely determine how and why we think. And this, in turn, gives us the opportunity to self-learn under the influence of some kind of internal stimuli - in contrast to a computer, whose ability to learn is more or less strictly limited by the framework of software. A computer solves individual problems much more efficiently than a person, but a machine cannot think like a person.
One of the characteristic examples of the reflection of our way of thinking is language. Almost any natural language often ambiguously defines various concepts, therefore, for a computer, recognizing the meaning of even ordinary text is a serious problem. In order for a computer to process such information, one has to resort to "translation" - formalization of speech, text or any other information. But we cannot expect a computer to do this on its own. Of course, with the help of programs, he will be able to form an answer for us that will make sense and seem completely human. But this is actually an imitation, not real human thinking. The computer in this case is a common information processing tool.
Nearly accurate imitation
Promotional video:
Modern software algorithms and computing power allow computers today to imitate human behavior so accurately that many media outlets write seriously about "thinking." Our computer IBM Watson has become widely known, which surpassed man in the game show Jeopardy (Russian analogue - "Own Game"), and both the questions of the game and the answers of the computer were formulated in natural language. Nevertheless, Watson is not a model of the human brain, but a specialized information processing system that parses natural language questions using algorithms and estimates the likelihood of a particular answer from an extensive database based on accumulated statistics. And although Watson is currently the most advanced system capable of "understanding" requests in natural language and responding to them, I assure you thatinside our computer you will not find a person - in any sense of the word.
Mechanistic way
To move from external imitation to real modeling of human thinking requires solving a completely different problem. Creating a computer that will not only act within a given program, but actually think like a human, requires repeating the biological path that nature has already gone through. In fact, you need to build an analogue of the human brain and give the machine all those channels of communication with the outside world that a person possesses. Of course, all this is speculative, since the practical implementation of such a project is still impossible to even imagine. And not so much because of imperfect technology or lack of computing power, but because we still do not understand how the human brain and our perception work.
Human perception is a huge mystery. So far, no one even has a rough idea of how it works, in the scientific study of this issue (psychologists, biologists, and cybernetics are also engaged in this) we are at the beginning of the path. Try to imagine the volumes of data that enter the brain: visual (with a huge resolution), audio data, tactile, temperature, gustatory, olfactory, emotional. All this information affects the emotional state, which affects analysis, data processing and decision making. The brain processes this gigantic amount of information in parallel and in real time. Now we don't even have any ideas how it would be possible to simulate such a scheme entirely in hardware (although, of course, some elements are already used in the development of new architectures).
Do we need a superbrain
An important aspect of modeling is energy efficiency. A human brain weighing about 1.5 kg consumes about 30 watts. Modern supercomputers occupy entire buildings, and the power consumption is calculated in megawatts. This means that if we could build a mechanistic model of the human brain, it would be huge and consume many orders of magnitude more energy than the original, not to mention cooling. However, technologies do not stand still - both IBM and other companies are working on new processor architectures, on new semiconductor materials that will reduce the consumption and size of computers. In addition, parallelization of computational processes will help to increase efficiency. Quantum computers are quite promising in this respect.
When it will be? If we set ourselves such a task today and provide sufficient funding, it may take a hundred years (this is a rather optimistic forecast). But will such a goal be justified? Creating a model of the human brain will not provide anything fundamentally new for solving everyday problems that traditional computers can handle. Moreover, you will have to face not only technological, but also ethical problems. However, they will arise in any case, because ordinary computers are penetrating all new key areas of human activity. For example, there is no longer any doubt that soon computers will control cars, and here we come into the field of ethics - who will be responsible in the event of an accident? But I have no fear of new technologies. After all, a computer is just a toolhelping to make the world more convenient for us people.
David Ferrucci, Artificial Intelligence Specialist, Head of Semantic Analysis and Integration at the IBM Thomas Watson Research Center, IBM Emeritus, creator of the IBM Watson supercomputer.
Interviewed by: Alexey Levin, Oleg Makarov, Dmitry Mamontov