Artificial intelligence
Artificial intelligence (English: artificial intelligence, abbreviated as AI), also known as omnibus, machine intelligence, refers to the intelligence displayed by machines made by humans. Usually artificial intelligence refers to the technology that presents human intelligence through ordinary computer programs. The term also points out whether and how to study such an intelligent system can be realized. At the same time, through advances in medicine, neuroscience, robotics, and statistics, normal predictions believe that many human occupations are gradually being replaced by them.
The definition field of artificial intelligence in general textbooks is "the research and design of intelligent agent." The intelligent agent refers to a system that can observe the surrounding environment and take actions to achieve goals. John McCarthy's definition in 1955 was "the science and engineering of making intelligent machines" Andreas Kaplan and Michael Haenlein define artificial intelligence as “the system correctly interprets external data, learns from this data, and uses this knowledge to achieve specific Goal and task capability".
The research of artificial intelligence is highly technical and professional, and each branch field is in-depth and unconnected, so it covers a wide range . The research of artificial intelligence can be divided into several technical issues. Its branch areas are mainly focused on solving specific problems, one of which is how to use various tools to complete specific applications.
The core issues of AI include the ability to construct reasoning, knowledge, planning, learning, communication, moving objects, using tools and manipulating machinery that are similar to or even superior to humans . Artificial intelligence is still a long-term goal in this field . At present, there have been preliminary results of weak artificial intelligence, and even some unilateral capabilities such as image recognition, language analysis, board games, etc. have reached a level that surpasses humans, and the versatility of artificial intelligence represents that the above-mentioned problems can be solved. The same AI program can directly use existing AI to complete tasks without re-developing algorithms. It has the same processing capabilities as humans, but it takes time to study to achieve integrated and strong artificial intelligence with thinking ability. The more popular methods include statistical methods and calculations. Intelligent and traditional AI. There are currently a large number of tools using artificial intelligence, including search and mathematical optimization, and logical deduction. And algorithms based on bionics, cognitive psychology, and probability theory and economics are also being explored gradually.
Deduction, reasoning and problem solving
Early artificial intelligence researchers directly imitated humans for step-by-step reasoning, just like the way humans think when playing a board game or performing logical reasoning. In the 1980s and 1990s, using the concepts of probability and economics, artificial intelligence research also developed very successful methods to deal with uncertain or incomplete information.
For difficult problems, a large amount of computing resources may be required, which means that a "probable combination explosion" occurs: When the problem exceeds a certain scale, the computer will require astronomical orders of memory or computing time. Finding more effective algorithms is a priority artificial intelligence research project.
Humans usually use the quickest and intuitive judgments to solve problems, rather than conscious, step-by-step derivation. Early artificial intelligence research usually uses step-by-step derivation. Artificial intelligence research has made progress in this "sub-representative" problem-solving method: the research of materialized Agent emphasizes the importance of perceptual movement. Neural network research attempts to reproduce this skill by simulating the brain structure of humans and animals.
Knowledge representation
Ontology represents knowledge as a set of concepts in a field and the relationship between these concepts.
Knowledge representation is one of the core research issues in the field of artificial intelligence. Its goal is to allow the machine to store the corresponding knowledge and to be able to deduct new knowledge according to certain rules. There are many problems to be solved that require a large amount of knowledge of the world, which includes prior knowledge stored in advance and knowledge obtained through intelligent reasoning. The prior knowledge stored in advance refers to the knowledge that humans tell the machine in some way. The knowledge obtained through intelligent reasoning refers to the knowledge obtained by combining prior knowledge and a specific reasoning rule (logical reasoning). First of all, prior knowledge can refer to the knowledge that describes the relationship between goals, characteristics, types and objects, and can also describe events, times, states, causes and results, and any knowledge you want the machine to store. For example: if there is no sun today, it will be cloudy without the sun. Then in the language of propositional logic, this knowledge can be expressed as: today --> no sun, no sun --> cloudy. This knowledge is prior knowledge, then new knowledge can be obtained through reasoning: today --> cloudy. From this example, it can be seen that the correctness of prior knowledge is very important. In this example, if there is no sun, it is cloudy. This proposition is not rigorous and general, because it may rain or snow if there is no sun. In addition, if artificial intelligence can see the sun, in addition to the question of how to judge, under this premise, it should also be able to judge the difference between cloudy and sunny days. The representation of logical propositions is very important in knowledge representation, and logical reasoning rules are currently the main reasoning rules. You can use logical symbols to define each logical proposition in the machine, and then let the machine store the corresponding logical reasoning rules, then the machine can naturally reason. At present, there are many dilemmas in knowledge expression that cannot be solved.
For example, it is almost impossible to build a complete knowledge base, so the resources of the knowledge base are limited; the correctness of prior knowledge needs to be tested, and sometimes the prior knowledge is not necessarily only Right or wrong choice.
Planning
Intelligent Agent must be able to formulate goals and achieve these goals. They need a way to build a predictable world model (express the state of the world in a mathematical model and predict how their behavior will change the world) so that they can choose the most effective behavior. In traditional planning problems, the intelligent agent is assumed to be the only influential one in the world, so what behavior it will perform is already determined. However, if this is not the case, it must regularly check whether the state of the world model matches its own predictions. If it does not comply, it must change its plan. Therefore, intelligent agents must have the ability to reason under uncertain results. In multi-agent, multiple agents plan to accomplish certain goals in a cooperative and competitive manner. The use of evolutionary algorithms and group wisdom can achieve the goal of an overall emergent situation.
Learn
Machine learning
The main purpose of machine learning is to allow the machine to obtain knowledge from users and input data, so that the machine can automatically judge and output the corresponding results. This method can help solve more problems, reduce errors, and improve the efficiency of problem solving. For artificial intelligence, machine learning is important from the very beginning.
Natural language processing
Natural language processing explores how to process and use natural language. Natural language cognition refers to making computers "understand" human language. The natural language generation system converts computer data into natural language. The natural language understanding system transforms natural language into a more tractable form of computer programs.
Weak artificial intelligence
The weak artificial intelligence view believes that it is "impossible" to create intelligent machines that can "really" reason and solve problems. These machines only "look" like intelligent, but do not really possess intelligence, nor do they have autonomous consciousness. .
Weak artificial intelligence only appeared in contrast to strong artificial intelligence, because the research on artificial intelligence was in a state of stagnation. It was not until the neural network had powerful computing power to simulate it before it began to change and significantly advanced. However, artificial intelligence researchers do not necessarily agree with weak artificial intelligence, nor do they care about or understand the content and differences between strong artificial intelligence and weak artificial intelligence, and debate on the definition.
