Natural language processing and other forms of machine intelligence

Scientists have been working on the creation of artificial intelligence for more than half a century, and in recent years they have made considerable progress. To solve practical problems, systems are formed similar to human neural connections, which are called “neural networks”. The results achieved make it possible to introduce artificial intelligence technology into various areas of human life: jurisprudence, medicine, etc. Although, of course, the need for artificial intelligence in the legal field can be questioned, since legal problems are easily solved using linear programming. But when it comes to medical research and life extension challenges, strong AI can accelerate progress.

At this stage in the development of artificial intelligence technologies, it can be argued that scientists are striving to create a machine that will become equal in mental abilities to the human mind. The learning process takes place in such a way that the machine is given a large number of solutions to the same problem, after analyzing and deriving some average value with many variations and deviations, the machine can later cope with this type of problem. For example, a face recognition system. To find out someone's face, the program refers to its database, where it recorded the coordinates of the location of the eyes, nose, lips and other features of the face of a particular person from different angles of view. In this case, the machine is trained on a certain sample, thereby programming the correct answer.

This way of teaching the machine would presumably lead to a highly educated system that would take the best out of its teachers in terms of mental characteristics. Those. such a machine will have an “intelligence” equal to that of a human, and only by virtue of pragmatism will it surpass an individual person who is subject to emotions and forgetfulness. But the true goal pursued by the creator of strong AI is an artificial system that can perform mental operations, while possessing the mental potential of all mankind. Thus, self-developing automated thinking will be able to perform as many tasks and create as many technologies in a matter of days as all of humanity has generated in its entire history.

To go from neural networks that recognize flowers from photographs to artificial intelligence, which will become a tool for solving the pressing problems of science, it is necessary to delve into the process of thinking itself and recreate its artificial counterpart. Scientists came up with the idea to reproduce the way a person thinks, delving into the analysis of the language aspect. Teach a machine to understand human speech and it will begin to think like a human. In his interview Liu Kang, Chief Scientist of Speech and Language Computing at Huawei's Noah's Ark Lab, reveals the idea of three types of machine intelligence: perceptual, motor, and cognitive. The first refers to the functions that are already mastered by machine thinking - this is the study, processing and generation of audiovisual materials.

In recent years, there has indeed been significant progress in improving speech recognition and face recognition. The second type of intelligence is also widely used in robotics. We've all seen Boston Dynamics' videos of cars surmounting varying degrees of physical barriers. And the third type of intelligence is cognitive, which is the most complex and has not yet been mastered at a sufficient level so that a machine can think and create like a person.

We and the Eon.plus strong AI development team are also inclined towards Mr. Liu Kang's approach, because there is a way of thinking behind speech. The verbal embodiment of a thought shows us how a person's thinking is actually built. The only problem is that there is no need to delve into the details, to raise the structure of thought starting from atoms and neural connections. And it is not necessary to transfer the problems of language to the plane of artificial intelligence, but it is necessary to understand the essence of thinking.

If you clearly define what the process of thinking is, then it is enough to reproduce it at a high level - at the level of ideas and the functions that they perform. The house can be built from small bricks, or it can be built from concrete blocks. In both cases, the main idea is achieved - the construction of the building. But in the first case, it will be necessary to spend more time, if you count, for example, in man-hours.

Exactly the same situation with the construction of artificial thinking. Do not build from small parts that require more time and computing power. It is enough to take two components of thinking: analysis and synthesis and competently embody them in the program code. In practice, the machine will operate directly with a high-level language, i.e. human. That is why it is worth paying attention to linguistics as a way of reflecting a person’s thoughts and the practical application of the principles of analysis and synthesis The authors of The Next Big Breakthrough in AI Will Be Around Language, H. James Wilson and Paul Daugherty of the consulting company Accenture, explain the importance of the emergence of artificial intelligence and give an example of the unprecedented hype around the GPT-3 natural language processing algorithm in various areas of life.

Their examples of already existing analogues of natural language processors once again confirm our assumption about the importance of linguistics in the process of becoming a strong machine intelligence. But the disadvantage of existing systems is that they have to use huge computing power to prove their suitability.

As we said, it's about the way that computing power is used. Without a clear understanding of the thinking process, developers leave this aspect to the judgment of the neural network. In fact, a large amount of information is loaded into a program based on a neural network, which functions thanks to a programmed mathematical processing model and produces a result similar to thinking. This method has the right to exist, but it involves honing this very math model to such a level that the thinking of the machine finally coincides with the human one. This naturally requires huge resources, which developers face with a lack.

If we go the other way and use the two components of thinking: analysis and synthesis at a higher level of constructing sentences that are the embodiment of ideas, then we will be able to reduce the load on the hardware of the machine. In this case, the AI is a functional idea processor that can dissect ready-made ideas and synthesize new ones.

For the integrity of this approach, another component is missing - a software package that will allow checking the relevance of a new idea in such a way that the machine does not generate nonsense, such as the idea of a blue sound. Such artificial intelligence is designed to generate something new. The goal of strong AI is to help solve problems that individual scientists spend more time than they could. If we attach hardware to such an idea-generating machine that can conduct experiments in real conditions, then we will get a universal scientist for solving problems from almost any field of human activity.

P.S. A little about the fears about job cuts due to the development of AI. It is difficult to refute the fact that a machine can do some work better and faster than a person. But no one talks about the number of professions where a person performs only a function, routine and daily. It is worth noting that with the development of technology, new professions appear that contribute to the development of the person himself.

For the same calculations some time ago, several employees with calculators were required, now this work can be done by one PC operator. The rest have become free from routine functions, and in connection with this, more creative professions appear that allow people to realize themselves and become happier.