Two Classic Schools of Artificial Intelligence

When it comes to intelligent manufacturing, people can easily think of various advanced algorithms, such as machine learning and logical reasoning. In fact, the most important development of artificial intelligence technology in the past ten years is deep learning technology, which is why artificial intelligence has become a hot topic recently.

The original intention of man inventing the computer was to help people perform data calculations. Computers can calculate many problems, but only some algorithms are called artificial intelligence algorithms.

Generally speaking, artificial intelligence algorithms often have two characteristics: one is that ordinary algorithms cannot solve them, and the other is that they are close to human thinking. Therefore, common algorithms such as arithmetic, equation solving, and sorting are generally not counted as artificial intelligence algorithms. They are only called artificial intelligence algorithms when they involve complex logical reasoning and knowledge learning.

When a computer solves a logical reasoning problem, it often turns it into a search problem first. The search problem that artificial intelligence pays attention to often faces combinatorial explosion, and it is difficult for computers to find the optimal solution. Chess is a typical problem of this kind. However, in the face of such combinatorial explosion problems, humans are often able to find relatively better solutions with limited search. This is where intelligence is embodied. Some people describe the characteristics of intelligent algorithms as algorithms that can quickly find better solutions from a huge search space. Therefore, Google was positioned as an "artificial intelligence company" at the beginning of its establishment.

To turn people's ideas into computer code, the premise is that they must be accurately expressed in computer language. However, many people's understandings are just hard to express clearly in words. For example, it is easy for us to recognize an acquaintance, and it is easy to recognize the taste of pears, but these recognitions are not easy to tell. For another example, a chess player has an intuitive understanding of "chess situation", which can help people focus on individual important chess pieces. However, this intuition is also difficult to describe in precise language. These are generally called "tacit knowledge".

What is not clearly expressed in human language often cannot be directly turned into computer code. Machine learning is used to solve this difficulty. The so-called machine learning generally uses mathematical functions to simulate the nervous system of humans or animals, and continuously revises this model through data to form knowledge similar to perceptual knowledge. This avoids the difficulty of tacit knowledge that is difficult to encode.

However, it is not easy for a computer to gain "perceptual knowledge". For example, the error rate of model recognition is often too high. There are many reasons for this kind of problems, including data reasons, model reasons, and training algorithms. With the enhancement of computer computing and storage capabilities, more and more data are accumulated. In this context, deep learning technology has emerged and achieved great success in many fields. As a result, artificial intelligence has become a hot spot in the near future.

People can understand artificial intelligence from many angles, and many schools of thought have emerged from it. Among them, the two classic mainstream schools are the semiotic school that simulates logical reasoning and the connection school that simulates the structure of the nervous system. The methods of these two schools can be combined and applied together. For example, Alpha Dog needs to perform logical reasoning, but in order to solve the problem of combinatorial explosion in search, it needs to simulate the perceptual knowledge of chess players, and this perceptual knowledge is obtained through deep learning.

Cybernetics School of Automation and Artificial Intelligence

In addition to the two classic schools mentioned above, there is another important school of artificial intelligence called the cybernetics school. Cybernetics is the theoretical basis of automation and intelligence. Over the years, the discipline of automation has been more mature than artificial intelligence, with a wider range of applications and greater influence. Therefore, when academia talks about artificial intelligence, they often refer to the above two schools, rather than the cybernetics school. However, the thinking of this school is precisely the main theoretical basis of intelligent manufacturing.

In the 1940s, Norbert Wiener, the father of cybernetics, thought of a question: What is the difference between a machine and an animal (or human)? Wiener believes that machines generally can only operate according to established steps and logic, while animals can perceive changes in the external world through information, and make decisions and take actions based on new information. For example, if a goat grazing suddenly sees a wolf, it will immediately stop eating grass and run for its life. Automation is to unify the three elements of perception, decision-making and execution. These three elements are similar to the functions of animal sensory organs, brain and limbs. This is the essential feature of automation. In fact, automation systems generally consist of sensors, controllers, and control objects, which are used for information acquisition, decision-making, and execution, respectively.

Unlike the two classic schools of artificial intelligence, cybernetics cares about effects and functions, and often does not care whether algorithms and logic are complex. In fact, some of the algorithms and logic used in automation may be quite simple.

In recent decades, the scope of automation applications has become wider and wider, but there are also limitations. Generally speaking, what an automated system can deal with is "foreseeable" changes. When there are unexpected problems such as equipment failures and production abnormalities, people still need to deal with them. This is because computers have plans to deal with problems, and their ability to deal with problems flexibly is far inferior to humans.

The Concept of Smart Manufacturing

Intelligent manufacturing technology is driven by the development of information and communication technology, and it is the extensive and in-depth application of information and communication technology in industry. Industry 4.0 in Germany and the Industrial Internet in the United States can be included in the category of smart manufacturing.

From the overall effect, intelligent manufacturing can strengthen the ability of enterprises to respond quickly to changes. When the market or users have new needs, they can design and manufacture them as soon as possible to supply the market; when the supply chain changes, they can try to avoid adverse effects on production and operation; when there are problems with production equipment or product quality, problems can be found as soon as possible Root causes and solutions to problems.

From a business perspective, the main role of promoting smart manufacturing is to promote multi-party collaboration, resource sharing and knowledge reuse. In layman's terms, collaboration is multi-party collaboration without delaying each other's work; resource sharing is conducive to low-cost access to high-quality resources; knowledge reuse can improve the efficiency of research and development and services, and reduce the cost of acquiring knowledge. When the material, knowledge, and human resources in an enterprise can be described digitally, the Internet can easily promote collaboration, sharing, and reuse.

The computer's computing power is very strong but its ability to deal with problems flexibly is weak. This is an important reason that limits the widespread application of automation technology. In order to solve this kind of problems, advanced manufacturing companies have generally adopted information technology. The information system can collect information for managers, help managers make decisions and manage the production and operation of enterprises. Compared with automated systems, information systems leave the decision-making work to humans.

For common problems, the logic and methods used by experts to deal with the problem can be turned into computer code, allowing the machine to make decisions based on human ideas. This is the digitization of human knowledge. In this way, we can further reduce the burden of humans dealing with problems and increase the level of automation of decision-making-this is actually intelligence.

In a sense, intelligence is the fusion of automation and information. The idea of fusion of automation and information technology has been around for a long time, but when the information and communication technology is not developed enough, it is technically difficult to realize it. As a result, the opportunity remains in the era of intelligence.

Intelligent Manufacturing and Artificial Intelligence

Intelligence is a decision-making revolution, that is, to replace human decision-making, help people make decisions, and "supervise" human decision-making through digital methods. For industrial processes, the knowledge required for decision-making is often the result of years accumulated by industrial people. The logic of these knowledge is often clear and can be accurately expressed. When advancing intelligent manufacturing, it is easy to convert this knowledge into computer code, but intelligent manufacturing may not use typical artificial intelligence algorithms. Therefore, the view that "intelligent manufacturing is equal to artificial intelligence plus manufacturing" is wrong.

However, the classic artificial intelligence technology can indeed promote the development of intelligent manufacturing technology. In some scenarios, the signal collected by the sensor is not easy to transform into clear semantic information. For example, a camera can collect image information on the surface of a product, but it cannot correlate the image information with the types and levels of quality defects. If such problems cannot be solved, the logic of quality management will be difficult to realize automatically, and the process of intelligentization will be hindered.

Typical artificial intelligence technologies such as deep learning are particularly good at solving image recognition problems. In fact, image recognition is the most typical and main application field of artificial intelligence algorithms in the industry. Without artificial intelligence technology, the system of intelligent manufacturing is often incomplete.

In a sense, artificial intelligence is a technical issue as well as an academic issue, which is also an issue that academia likes to study. In reality, automation is often only a technical problem, not an academic problem, because engineers generally like to use the simplest way to solve problems. The process of advancing intelligence not only involves technical issues, but also often involves the reconstruction of corporate organizational processes and the innovation of business models. In this sense, the problem of intelligence can often be seen as a problem of corporate management or even strategy.