The most popular memory problem in artificial inte

  • Detail

Memory problems in artificial intelligence -- a way for philosophy to appear in the intelligent age

original title: memory problems in artificial intelligence -- a way for philosophy to appear in the intelligent age

on December 12, 2018, Stanford University released the AI index report. This report is of great value and can help us grasp the development and trends in the field of artificial intelligence. This report points out at least three areas worthy of researchers' attention: 1) machine learning, neural network and computer vision were once three hot directions; 2) Compared with images, language and common sense will become the frontier of artificial intelligence research; 3) AI's humanities research is a weak point for the whole world. The third point deserves the attention of scholars in the field of humanities. Today, the humanities are shrinking day by day. The fact that artificial intelligence exceeds the technical stipulation may bring a new possibility to the Renaissance of humanities

generally speaking, humanities covers a wide range, including philosophy, history, literature, philosophy, religion, music, art, etc. However, this statement is not directly helpful to the Humanities Research of artificial intelligence. Not all humanities and social sciences are related to the AI field. We need to grasp the possible related disciplines from the context of technological development, such as ethics, philosophy of mind, sociology, and more related art. It is very necessary to think further from the perspective of philosophy. As the basis of many disciplines, if we can see the possibility of strengthening from the discipline of philosophy, the relevant research of other humanities and social sciences will also have a theoretical basis. This article will continue to explore the possible ways for philosophy to appear in the intelligent age. Generally speaking, the idea that philosophy is regarded as "the key to decoding artificial intelligence" has gradually taken shape. The metaphorical concept of "key" originally came from David Deutsch, a physics professor at Oxford University, and was later accepted by the domestic philosophical community. For example, philosophy is the key to understanding the "development" of information civilization (Wang Tian cantilever impact tester, which places the specimen vertically and clamps it in the impact tester, impact elasticity tester en, 2018). The author proposed that "memory philosophy is the key to decoding artificial intelligence and its development" (yangqingfeng, 2018)

continued reflection on the basis of "key metaphor" can give philosophy a more solid foundation in the discussion of artificial intelligence issues, and can also clarify the way philosophy appears in the intelligent era. This paper chooses the perspective of memory research. In previous studies, the author pointed out that memory is an ancient philosophical problem, but gradually forgotten by philosophy itself, and then turned into a psychological problem (yangqingfeng, 2017; 2018). In order to better carry out the analysis, we start with the four discussion methods of artificial intelligence. The four ways of discussion are: linguistic discussion, functional discussion, behavioral discussion and structural discussion

mnemosyune, the goddess of memory, language and writing in Greek mythology, is one of the nine Muses. Painted by Rossetti, an English painter in the 19th century

first, in the discussion of semantics, intelligence and ability are usually equated, and memory is regarded as one of many abilities. In the discussion of semantics, intelligence is embodied in many abilities of artificial intelligence system. Artificial intelligence is also defined as making machines imitate human intelligent thinking or behavior, which has become a more generally accepted definition. In terms of capability division, it can be divided into basic capability and advanced capability. In the perspective of philosophical anthropology in the 18th century, basic abilities are often embodied in the abilities related to five senses, such as vision, hearing, touch, taste and smell. Another is the basic ability to grasp moving objects, which was not revealed until the early 20th century. Advanced abilities are usually regarded as many abilities that are related to rationality and emotion according to zhenglansun, member of the Standing Committee of the CPPCC National Committee, vice chairman of the Central Committee of the Democratic League of China and academician of the Chinese Academy of Sciences. The former is like understanding, judgment and reasoning, while the latter is like various emotions. Memory is usually regarded as one of the important basic abilities, because it is the result of human sensory retention. In addition, memory is regarded as the basis of human thinking, decision-making and action. This understanding is generally accepted. Yann Lecun, a professor at New York University, pointed out that "intelligence and common sense are equal to perception + prediction mode + memory + reasoning and planning". He pointed out that the main task of predictive learning is to predict the past, present and future from the data information provided. First of the all, we have not grasped key essence of the memory, that is, it is premise and condition for existence and presentation of the above phenomena. Secondly, the important role of forgetting is neglected. The latest achievements in neuroscience begin to reveal the role of forgetting in thinking, decision-making and action. In addition, many amazing ideas have been revealed in the revelation of the memory ability of different agents. For example, a recent research result shows that artificial intelligence presents a neuron structure similar to human and animal organisms in spatial memory. This conclusion makes us need to pay attention to and think about the similarities among machine intelligence, human intelligence and animal intelligence. In addition, in the discussion of agents, the experience related to memory will become an important issue. When we discuss whether the machine has memory, it is not only a problem of language analysis, nor is it just a problem of function realization, but also a prerequisite problem related to whether the machine can surpass human beings

secondly, in the discussion on the functionality of agents, it is more prominent that artificial intelligence is a functional expression to be realized based on a specific structure or mechanism, and memory is regarded as one of the important factors that constitute the above specific structure or enhance the specific structure and mechanism. For example, academician Chen Lin of the Chinese Academy of Sciences pointed out that the core basic scientific issues of artificial intelligence are cognition and computing. Memory is an important component of cognitive hierarchy. Professor Zhang Ba of Tsinghua University, who is also an academician, pointed out that the trend of artificial intelligence is an AI system based on knowledge and data. The basis of these two abilities is memory, and decisions and actions based on the interpretation of historical data. In yanglikun's opinion, the circulatory collaterals can't carry out long-term memory and need a separate "hippocampus" (memory module). Memory module plays an important role in the enhancement of neural network ability. All these points show the necessity of memorizing the concept of probable stress at point B in artificial intelligence. Only in different abilities, there are different definitions of memory. For example, in memory comprehension related to perceptual ability, memory is represented by the storage and extraction of information; In the category of cognition, memorizing information has become a prerequisite for cognition. In the category of decision-making, memory is mainly represented by the extraction of effective information, and forgetting is represented by the screening and neglect of invalid information. When discussing the function of machine continuous learning, German scientist zhangjianwei mentioned "the memory development of the machine itself"

third, in the discussion of agent behavior, behavior will show at least four related patterns, and the role of memory in these patterns can not be ignored. According to phenomenological methods, we can divide behavior related patterns into conscious subject behavior patterns, context behavior patterns, environment behavior patterns and agent behavior patterns. Agent behavior pattern can usually be regarded as quasi agent behavior pattern, because it has a structure similar to agent behavior pattern. The consciousness behavior model emphasizes that the behavior is based on a certain conscious purpose. In this conceptual framework, what action needs is not only perception, calculation and judgment, but also the experience accumulated in the learning process; The context behavior model emphasizes the social, cultural and political factors behind the behavior, which is different from the purely objective environmental factors. If context behavior is taken as the conceptual framework, then the behavior comes from a specific context, and the establishment of this context draws on the experience of the machine; The environment behavior model mainly emphasizes the role of the environment in a certain special DuPont Hongji new material project, which is generated by the investment of Foshan DuPont Hongji Film Co., Ltd. (hereinafter referred to as "DuPont Hongji"), and also emphasizes that behavior is a response to a specific environment; The agent behavior pattern emphasizes that the behavior of the functional body is the response to the environment of the design scene, while the action emphasizes the response to the environment. These four models all need memory as the premise. In the process of these behaviors, retaining the past learning experience is conducive to continuous learning and predictive learning

fourth, in the structural discussion, we pay more attention to the agent human unity, and the problem mainly focuses on the impact of agents on human memory. When we enter the intelligent system composed of human and agent, this problem goes far beyond the hardware, but contains more complex humanistic dimensions. For example, the impact of artificial intelligence on human memory. At the world's top scientists' forum, scientists noticed this problem when discussing the topic of artificial intelligence. On the world-famous "tablecloth", at least two scientists mentioned the impact of AI on human memory, such as the enhancement and enhancement of human memory

the above mainly discusses the methods of philosophical research on artificial intelligence. Compared with cognitive philosophy, memory research has a wider range and possibility. For artificial intelligence, cognitive philosophy is only limited to the cognitive function of AI, while memory research is different. On the one hand, memory is an ancient problem of philosophy, which has ontological significance, but is completely covered by the tradition of epistemology and epistemology. Our previous studies have revealed that memory is not an appendage of cognition, nor is it just a spiritual ability slightly lower than cognition, but has a more important ontological status: memory exists as three conditional forms: the basic condition of cognition and emotion Understand the historical conditions of human beings and the conditions for realizing self and other identity; On the other hand, the development of artificial intelligence increasingly shows the inherent indispensability of memory factors. Memory plays a unique role in the cognitive activities, function presentation, behavior expression and structure formation of artificial intelligence, which need further in-depth research. As academician Zhang Ba pointed out, there is a memory mechanism in brain science, which is worthy of attention and learning by those who design AI systems

this paper is a phased achievement of the major project of the National Social Science Foundation, "intelligent revolution and philosophical research on the prospect of human deep scientificization" (17zda028); Phased achievements of the CAS project "ethical issues and social governance of big data"

Copyright © 2011 JIN SHI