Artificial intelligence (AI) is a branch of machine intelligence and computer science. It is a new, comprehensive, and dynamic interdisciplinary field that studies machine intelligence and intelligent machines.
Artificial intelligence has been considered one of the world's three cutting-edge technologies since the 1970s (along with space technology and energy technology). It is also considered one of the three cutting-edge technologies of the 21st century (along with genetic engineering and nanoscience).
This is because it has developed rapidly over the past thirty years, has been widely applied in many disciplines, and has achieved fruitful results. Artificial intelligence has gradually become an independent branch, forming its own system in both theory and practice.
Artificial intelligence (AI) is the study of enabling computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It primarily includes understanding the principles of computer intelligence, creating computers that resemble human brain intelligence, and enabling computers to achieve higher-level applications. AI involves disciplines such as computer science, psychology, philosophy, and linguistics.
It can be said that it encompasses almost all disciplines in the natural and social sciences, and its scope has far exceeded the scope of computer science. The relationship between artificial intelligence and cognitive science is one of practice and theory. Artificial intelligence is at the level of technological application of cognitive science, and is one of its applied branches.
From a cognitive perspective, artificial intelligence is not limited to logical thinking. It is necessary to consider figurative thinking and intuitive thinking in order to promote the breakthrough development of artificial intelligence. Mathematics is often regarded as the basic science of many disciplines. Mathematics has also entered the fields of language and thinking. The discipline of artificial intelligence must also use mathematical tools. Mathematics not only plays a role in standard logic and fuzzy mathematics, but also enters the field of artificial intelligence. They will promote each other and develop more rapidly.
Main characteristics of artificial intelligence
Intelligence is the sum of knowledge and intellect. Knowledge is the foundation of intelligent behavior; intellect is the ability to acquire knowledge and apply it to solve problems. Intelligence has the following characteristics:
1. Having perceptual ability: refers to people's ability to perceive the external world through sensory organs such as sight, hearing, touch, taste, and smell;
2. Possessing the ability to remember and think: This is the most important function of the human brain and the fundamental reason why humans have intelligence;
3. Possesses learning and adaptive abilities;
4. Possess the capacity for action.
The origin and development of artificial intelligence
In 1936, 24-year-old British mathematician Alan Turing proposed the theory of automata, greatly advancing the research on thinking machines and computers, and earning him the title of "father of artificial intelligence." Research into artificial intelligence officially began in 1956, the year the term "Artificial Intelligence" (AI) was formally used at a conference held at Dartmouth College.
Phase 1: The rise and fall of artificial intelligence in the 1950s.
After the concept of artificial intelligence was first proposed, a number of significant achievements emerged, such as machine theorem proving, checkers programs, general problem solvers, and the LISP table processing language. However, due to the limited reasoning capabilities of solution-based methods and the failures of machine translation, artificial intelligence entered a period of stagnation. This stage was characterized by an emphasis on problem-solving methods while neglecting the importance of knowledge.
The second stage: from the late 1960s to the 1970s, the emergence of expert systems brought about a new peak in artificial intelligence research.
The research and development of expert systems such as the DENDRAL chemical mass spectrometry analysis system, the MYCIN disease diagnosis and treatment system, the PROSPECTIOR mineral exploration system, and the Hearsay-II speech understanding system have brought artificial intelligence towards practical application. Furthermore, the International Joint Conferences on Artificial Intelligence (IJCAI) was established in 1969.
The third stage: In the 1980s, with the development of the fifth-generation computer, artificial intelligence made great progress.
In 1982, Japan launched the "Fifth Generation Computer Development Project," also known as the "Knowledge Information Processing Computer System (KIPS)," with the aim of making logical reasoning as fast as numerical computation. Although the project ultimately failed, its implementation sparked a surge of research into artificial intelligence.
The fourth stage: In the late 1980s, neural networks developed rapidly.
In 1987, the United States hosted the first international conference on neural networks, marking the birth of this new discipline. Since then, investment in neural networks has gradually increased in various countries, and neural networks have developed rapidly.
Fifth stage: In the 1990s, artificial intelligence experienced a new research boom.
Due to the development of network technology, especially the Internet, artificial intelligence research has begun to shift from research on single intelligent agents to research on distributed artificial intelligence in a network environment. This research not only studies distributed problem-solving based on the same objective, but also multi-objective problem-solving involving multiple intelligent agents, making artificial intelligence more practical.
Furthermore, the introduction of the Hopfield multilayer neural network model has led to a flourishing of research and applications in artificial neural networks. Artificial intelligence has penetrated into all aspects of social life.
Mathematical Foundations of Artificial Intelligence
Artificial intelligence is the simulation of human intelligence on a computer. The core of intelligence is thinking. Therefore, how to formalize and symbolize people's thinking activities so that they can be realized on a computer has become an important topic in artificial intelligence research.
In this regard, the relevant theories, methods, and techniques of logic play a very important role. They not only provide powerful tools for artificial intelligence but also lay the theoretical foundation for knowledge reasoning. Furthermore, the relevant concepts and theories of probability theory and fuzzy theory also occupy an important position in the representation and processing of uncertain knowledge.
Therefore, it is essential to acquire some knowledge of logic, probability theory, and fuzzy theory before systematically learning the theories and technologies of artificial intelligence.
The logic used in artificial intelligence can be broadly divided into two categories. One category is classical propositional logic and first-order propositional logic, which is characterized by the fact that the truth value of any proposition is either "true" or "false", and one of the two must be true.
Because it has only two truth values, it is also called binary logic. Another category refers to logic other than classical logic, mainly including three-valued logic, multi-valued logic, fuzzy logic, modal logic, and temporal logic, collectively known as non-classical logic.
Non-classical logic can be further divided into two categories. One is logic parallel to classical logic, such as multivalued logic and fuzzy logic. These use essentially the same language as classical logic, but the main difference is that some theorems in classical logic no longer hold true in these non-classical logics, and new concepts and theorems are added. The other category is extensions of classical logic, such as modal logic and temporal logic.
They generally acknowledge theorems of classical logic, but expand upon them in two ways: first, by expanding the language of classical logic; and second, by supplementing classical logic theorems. For example, modal logic adds two new operators, L (…is certain) and A4 (…is possible), thus expanding the vocabulary of classical logic. The application of probability theory in artificial intelligence is mainly reflected in concepts related to probability and conditional probability, as well as Bayes' theorem, and has been the theoretical foundation for handling uncertainty in artificial intelligence for many years.
In scientific research and daily life, people have always sought to describe phenomena or solve problems using definite mathematical models. With the rapid development of communication, computer, and network technologies, and the widespread application of basic software, middleware, and application software, computers have greatly enhanced their capabilities in numerical computation, data processing, information retrieval, industrial control, symbolic reasoning, and even knowledge engineering. However, the problems that computers solve in these fields are often "well-defined problems," meaning that the preconditions for solving the problem are clear, the mathematical model is accurate, and it can be described using a computer programming language.
Artificial intelligence has always been developed on the basis of mathematics, and mathematical support is also needed to solve the various uncertainties in artificial intelligence.
Forms and research areas of artificial intelligence
Game Theory
Game theory, also known as strategic decision-making, is a theory that uses rigorous mathematical models to study optimal decision-making problems under conflict and adversarial conditions. The earliest application of game theory in artificial intelligence was in computer games, specifically chess. To design programs that could compete with and even defeat humans, researchers began to study how to enable computers to learn human thought processes and possess the same game-playing abilities as humans.
The process of game theory involves four important issues: problem representation, decomposition, search, and induction. Computer chess games are essentially dynamic games with complete information. This means that both players not only know the current situation and their opponent's past moves, but also what moves their opponent might make next. Although the number of possible moves can be in the tens or hundreds, it is still finite. Computers can describe this game process by constructing a vast game tree with the root at the top and the leaves at the bottom. Then, by utilizing their powerful time and space capabilities, they can perform ingenious searches to find feasible solutions and near-optimal solutions, thus providing the current move.
Clearly, a computer's search capability is a significant indicator of its intelligence. Search algorithms are the core of machine "thinking," encompassing move generation, game tree expansion, various pruning search methods, and heuristic search. It's evident that the design and implementation of search algorithms embody the principles of artificial intelligence. Machine game theory serves as a simple, convenient, economical, and practical medium for studying logical thinking, offering rich content and endless variations.
A game of chess can be played in an hour or two, allowing for the testing of a computer's "intelligence." Furthermore, it allows for undoing moves, retrying, and reviewing the game, gradually revealing the gap between computer and human brain functions and thus continuously improving the computer's intelligence level. Undoubtedly, research into machine game playing can significantly promote the development of artificial intelligence.
Expert System
An expert system is an intelligent program system with a wealth of specialized knowledge and experience. It utilizes the accumulated experience and expertise of domain experts to simulate their thought processes and solve complex problems in the field that typically require expert knowledge. Expert systems represent one of the most active and fruitful research areas in artificial intelligence. As a knowledge-based system, it acquires knowledge from human experts and uses it to solve difficult problems that only experts can address, thus supporting teaching systems.
Artificial intelligence expert systems typically consist of a knowledge base and an inference engine. The inference engine primarily determines which rules satisfy the facts or objectives, assigns priority to the rules, and then executes the highest priority rule for logical reasoning. The knowledge acquisition engine establishes a method for automatically inputting knowledge for the user. The matching module is the core of this artificial intelligence expert system; the implementation of the matching function is crucial to the entire program's implementation. The interpretation module and result processing all depend on its execution results. The process is illustrated in the following diagram:
There are many expert system models that have been studied. Among them, the following are some of the more popular ones:
Rule-based expert systems
Rule-based reasoning (RBR) methods rely on past expert diagnostic experience, which is summarized into rules, and then used heuristic experiential knowledge for reasoning. Early expert systems mostly used rule-based reasoning, such as the DENDRAL expert system, MYCIN expert system, and PROSPECTOR expert system.
Case-based expert systems
Case-based reasoning (CBR) is a method that searches for similar problems that have been successfully solved in the past, compares the differences in characteristics and background between new and old problems, and reuses or refers to previous knowledge and information to ultimately solve the new problem. The first truly case-based expert system was the CYRUS system, developed in 1983 under the leadership of Professor Janet Kolodner of Yale University. Based on Schank's dynamic storage model and the MOP (Memory Organized Packet) theory for problem solving, it was used for travel-related consulting work.
Frame-based expert systems
A frame is a general data structure that organizes all the knowledge of a certain type of object together, and interconnected frames form a frame system.
The most prominent feature of frame notation is its ability to express structured knowledge, and it also possesses good inheritance and naturalness. Therefore, frame-based expert systems are suitable for things, actions, or events with fixed formats.
Fuzzy logic-based expert systems
Unlike binary Boolean logic, fuzzy logic is multivalued. It deals with the degree of belonging and the degree of confidence. Fuzzy logic uses a continuous interval of logical values between 0 (completely false) and 1 (completely true). Unlike black-and-white logic, it uses a color spectrum and can accept things that are both partially true and partially false at the same time.
The advantages of fuzzy logic-based expert systems are: ① They possess expert-level specialized knowledge, demonstrating expert skills and high levels of expertise, as well as sufficient robustness; ② They can perform effective reasoning, are heuristic, and can utilize the experience and knowledge of human experts for heuristic searches and exploratory reasoning; ③ They possess flexibility and transparency. However, fuzzy reasoning knowledge acquisition is difficult, especially the determination of fuzzy relationships of symptoms, and the system's reasoning ability relies on a fuzzy knowledge base, exhibiting poor learning ability and a tendency to make errors. Since fuzzy linguistic variables are represented by membership functions, achieving the conversion between linguistic variables and membership functions is a challenge.
Expert systems based on DS evidence theory
Evidence theory, first proposed by Dempster in 1967 and further developed by his student Shafer in 1976, is an imprecise reasoning theory, also known as Dempster/Shafer evidence theory (DS evidence theory). It falls under the category of artificial intelligence and was initially applied to expert systems, possessing the ability to handle uncertain information. As a method of uncertain reasoning, the main characteristics of evidence theory are: it satisfies weaker conditions than Bayesian probability theory; and it has the ability to directly express "uncertainty" and "not knowing." When the constraints are limited to strict probabilities, it becomes probability theory.
Web-based expert systems
Web-based expert systems are advanced expert systems resulting from the integration of Web data exchange technology with traditional expert systems. They utilize web browsers for human-computer interaction, allowing various users to access the expert system through a browser. Structurally, they consist of three layers: a browser, an application server, and a database server, including a web interface, an inference engine, a knowledge base, a database, and an interpreter.
Pattern recognition
Broadly speaking, any observable thing existing in time and space, if we can distinguish whether they are the same or similar, can be called a pattern. However, it's important to note that a pattern doesn't refer to the thing itself, but rather to the information we derive from it. Therefore, patterns often manifest as information with temporal or spatial distribution. To understand objective things, people categorize them according to their degree of similarity. The role and purpose of pattern recognition is to correctly classify a specific thing into a particular category.
Pattern recognition systems consist of two processes: design and implementation. Design refers to designing a classifier using a certain number of samples (training or learning set). Implementation refers to using the classifier to make classification decisions on the samples to be recognized. Statistical pattern recognition systems mainly consist of four parts: data acquisition, preprocessing, feature extraction and selection, and classification decision, as shown in the figure below:
In summary, the most fundamental problem in pattern recognition is solving the classification of patterns. More comprehensively, it involves the description, analysis, classification, understanding, and synthesis of patterns. Higher-level pattern recognition should also include learning, judging, adapting, optimizing, and automatically discovering patterns.
Therefore, pattern recognition is similar in some ways to "learning" and "concept formation" in artificial intelligence. The combination of pattern recognition and its functions will open up broad application prospects.
Artificial Neural Networks
As we all know, the human brain has a wonderful organizational structure and operating mechanism. From the perspective of imitating human brain intelligence, exploring new ways of information representation, storage and processing, designing a brand-new computer processing structure model, and building an information processing system that is closer to human intelligence to solve problems that are difficult to solve in practical engineering and scientific research fields will surely greatly promote scientific research progress. These factors have led to the emergence of artificial neural networks (ANN).
In simple terms, an artificial neural network (ANN) is a machine designed to mimic the workings of the human brain. It can be implemented using electronic or optoelectronic components, or simulated on a computer using software. Recent research even shows that humans have created the first artificial neural network in a test tube using DNA (this circuit composed of interacting molecules can, like the human brain, regress based on incomplete patterns). Artificial neural networks have the ability to learn and adapt. They can analyze and grasp the potential patterns between a set of pre-provided corresponding input-output data, and finally, based on these patterns, use new input data to predict the output results. This learning and analysis process is called "training".
Neural networks are developing rapidly and have been widely applied in all aspects of the market.
Natural Language Understanding
Natural language understanding has always been one of the important research topics in the field of artificial intelligence, because natural language itself has a unique charm. First, if computers can understand natural language, then human-computer interaction will become smoother than ever before, which will be a major breakthrough in computer technology. Second, the creation and use of natural language is the crystallization of human wisdom over thousands of years, and studying natural language will help to unravel the mysteries of artificial intelligence.
There are four principles for understanding natural language: question answering, text summarization, paraphrasing, and translation. Correspondingly, the processing steps for natural language understanding are: formal description of the language, algorithm design, algorithm implementation, and evaluation. Formal description involves studying the inherent laws of natural language and then using mathematical methods to describe it for computer processing; this can also be considered as mathematical modeling of natural language. Algorithm design transforms the formal mathematical description into an object that the computer can operate and control. Algorithm implementation and evaluation involve implementing the algorithm using a programming language (such as C) and evaluating its performance and functionality.
The intelligent applications of natural language understanding are mainly reflected in translation.
The Present and Future of Artificial Intelligence
Artificial intelligence is no longer the exclusive domain of a few scientists. Almost every computer science department in universities around the world has people researching this subject, and undergraduate students studying computer science must also take such a course. Thanks to everyone's unremitting efforts, computers now seem to have become quite intelligent.
Many people may not realize that in some places, computers are helping people perform tasks that were originally solely human. Computers, with their speed and accuracy, are playing a vital role for humanity. Artificial intelligence remains at the forefront of computer science, and computer programming languages and other computer software exist because of advancements in artificial intelligence.
Currently, research and applications have expanded from laboratories to industrial sites, and from home appliances to rocket guidance. These applications are now widely used in weapon control, robot planning and control, intelligent control of automated processing systems (in manufacturing, mining, etc.), fault detection and diagnosis, intelligent control of aircraft, medical intelligent control, and intelligent instruments.
In the natural sciences, AI intersects, permeates, and promotes other disciplines. AI provides other disciplines with tools and methods, such as knowledge representation and reasoning mechanisms, problem-solving and search algorithms, fuzzy logic reasoning and non-monotonic reasoning techniques, and computational intelligence technologies, enabling the solution of previously intractable problems. Meanwhile, important concepts from other disciplines are also being developed in AI research, such as time-sharing systems, cataloging systems, and interactive debugging systems for computer systems.
The same applies to the social sciences. In disciplines that require mathematical-computer tools to solve problems (such as economics), the benefits of AI are self-evident.
More importantly, AI, in turn, can help humanity ultimately understand the formation of its own intelligence. In reinterpreting the history of knowledge, AI has the potential to resolve the ambiguity and inconsistencies of knowledge. This will lead to improvements in logic and philosophy, impacting core theories in psychology and cognitive science, and bringing about a radical transformation in philosophical and sociological theories.
Furthermore, by comprehensively applying grammatical, semantic, and AI-based formal knowledge representation methods, it is possible to improve the natural language expression of knowledge. Simultaneously, latent knowledge, intuition, and inspiration can also be expressed in applicable AI forms, thereby expanding the domain of knowledge and refining existing knowledge.
If biological computers, quantum computers, and photonic computers represent the future direction of computer hardware systems, then achieving artificial intelligence will be the future goal of computer software. However, in a sense, the goal of artificial intelligence development is to break away from computers and cease to exist as an independent subsystem. It will permeate every aspect of our society, subtly and imperceptibly.
It is foreseeable that as artificial intelligence improves, it will have a huge impact on human civilization as a whole. In fact, this impact has already occurred, but its gradual advancement is not enough to produce an explosive effect. Therefore, it is not being noticed by most people.
The impact of artificial intelligence on the economy
Successful expert systems can bring significant economic benefits to their builders, owners, and users. In the information-exploding knowledge economy, excellent information processing is wealth, contributing greatly to the economic well-being of some. Meanwhile, although the goal of artificial intelligence development is to become an independent application, it will still rely on computers for a long time to come. Increasingly sophisticated AI will place new demands on both computer hardware and software, which will be a driving force for the computer industry.
The impact of artificial intelligence on society
Artificial intelligence and robotics are almost inextricably linked. In Europe and America, industrial process control systems, intelligent robot systems, and intelligent production systems are beginning to emerge. my country has also developed from scratch, with manufacturers of robotic arms appearing, and the rudiments of a robotics industry have taken shape. It is expected to reach a significant scale in 10-20 years, becoming an independent industry separate from automation. However, this has brought about labor and employment issues. Because AI's applications in science and engineering can replace humans in various technical and mental tasks, it will force significant changes in people's work methods, even causing unemployment. A large number of unemployed people will become a factor of social instability.
The impact of artificial intelligence on human thinking
As machines become increasingly "intelligent," people place greater trust in their judgments and decisions. This, to some extent, can lead to a loss of responsibility and sensitivity towards problems and their solutions, resulting in a decline in cognitive abilities and a sluggish mindset. In simpler terms, it means becoming less intelligent. Humans have taken two million years to evolve into the intelligent beings we are today; with the "help" of artificial intelligence, this reverse process may not take that long.
As for the so-called "artificial intelligence going out of control" and "intelligent robots turning against humanity," Hollywood has made too many speculations. But it must be said, can the so-called "Three Laws of Asimov" really bind robots (artificial intelligence in the narrow sense) forever? It's hard to say. Nature is unpredictable. The accidental collision of two atoms sparked the first flame of life. Don't countless combinations of 0s and 1s also have those moments of inspiration? Chaos mechanisms have always been the domain of God. The probability of the birth of digital life is infinitesimally small and can be considered zero in mathematics, but it is possible in reality.
However, we cannot throw the baby out with the bathwater. Artificial intelligence has already—or is—or will soon—prove its enormous role in human society, and we should be optimistic about its future development. We believe that artificial intelligence has a brighter future; although the arrival of that day will require hard work and a high price, and the continuous efforts of several generations. Generations of scientists have provided us with the shoulders of giants, precisely so that we can stand on them and carry on the legacy.
