In August this year, Lei feng's network in Shenzhen will hold an unprecedented global influence of artificial intelligence and robotics Innovation Conference. When Lei feng's network (search for "Lei feng's network", public interest) will be published "artificial intelligence and robot Top25 list of innovative enterprises". At present, we are visiting related companies in the area of artificial intelligence, robots, sift through them eventually elected to the list list of companies. If you want to join our list, please contact: 2020@leiphone.com.
Lei feng's network: the author should be benevolent people from bupt and cognition Lab (male: man machine and cognition lab). This article is about the exploration of the thinking part of the machine, part can be seen here: can machines think? (I)
Early research on the mind, basic is the feeling of introspection, this sentiment to build beyond psychology, independent of the physical world, characterized by dualism of matter and mind. By the 20th century, Behaviorism started becoming popular, not about subjective feelings objectively, observe the external behavior and tendencies, not the so-called psychological explanations. After the invention of the computer, especially the work of Turing, replacing Behaviorism functionalism, became the mainstream of mind and intelligence. Functions under the banner of computationalism to achieve the calculation to spy on cognitive function, remains the standard paradigm of artificial intelligence and cognitive science.
17th century United Kingdom philosopher Thomas Hobbes said: "thinking is computation".
Computationalism (computationalism) cognitive process understood as a calculating process, the so-called mental status, mental activity and mental processes is just calculation state, activities of the intelligent systems and calculation process in cognition is calculated. The cognitive computing called "computational theory of the mind" (the Computational Theory of Mind). It is generalized to compute the initial field, also known as special calculations.
Computationalism went on to enter the neighborhood to life. Mathematicians, computer construction of von Neumann, says: living systems can be seen as an automata, automata containing programs and data in two parts, according to programmed instructions for data processing. He proposed the concept of cellular automata, cellular automata depending on their surrounding relationships (input data) according to simple rules (programs), and decide the fate of its next (output data). Program algorithm for cellular survival and reproduction of individual life and cellular description of the ethnic characteristics of distribution data to life, as cells to organisms. People studying artificial life, understanding of natural life with a new perspective, that the essence of life, does not constitute a material life, but calculated. Gene science developments further enriching the life of computationalism. Gene is a program of biological cells, as well as cellular automata program about the variety, shape and fate. Information, algorithms and calculations become an important concept in understanding the nature of life.
Physics, believing that the world is a priori according to certain rules of movement, using mathematical formulas to express the relationship.
So calculation can reveal various forms and nature of change in the physical world. But things are not all like clockwork with a delicate structure, it is difficult for natural system centralized serial like a computer, and how do they work? S. Wolfram systematic study of cellular automata state changes, finding many complex patterns of the natural world, snow, rocks, and leaves, will have no difficulty using cellular automata mechanism to generate, they follow very simple rules for parallel computing, phenomena on Earth but different rules to automatically run. He would study the cellular automaton has written a thick book called a new kind of science, I believe this will be physical, chemical, biological and scientific mainstream have a revolutionary impact. Some calculations in physics, space-time is discrete, this world is not so much running in continuous time and space, rather than a myriad of individuals according to simple rules, spatio-temporal evolution of lattice calculations. Some people believe that the universe is a computing system, information is the source of all things, the history of the universe is continuous quantum computing process, the material is the universe of computer hardware, the mysteries of the universe program password.
Philosophical speculation is meditation different in the past, modern computer upgrades the thinker in the hands of paper and pencil, computationalism provides a versatile tool available for building a variety of functional model study status change in the calculation. It allows us to see and understand system status and change processes based on the material. Many of the past to explore the phenomenon, perhaps just system under the simple rules of evolution, emergence of macroscopic properties. In the modern era, computing has become with experimental and theoretical influences supports various scientific theories and research methods in philosophy became a look at a new conception of the world.
Biological and physical cellular automata and artificial neural network, which belongs to the Connectionist model of computation. They were the first to be studied is now a mainstream research in artificial intelligence. But in the last century, for a long time, known as "classical computational theory of the mind (CCTM)" symbol in the world. It constructed in the bright and modern mathematical logic Research Foundation, has a clear mechanism for context. It mimics the human reason, and heuristic search in predicate logic operation, by symbolic computation machine intelligence. This intelligent research based on logical reasoning and knowledge systems, making expert system in 70 and 80 a byword for artificial intelligence, and can read minds. At that time people believed that, in theory, as long as there is proper procedure, the computer can rational thinking like a person. Realization of machines with human intelligence, but an engineering problem.
80 when the study smart, look to the future with full confidence, philosopher John Searle used "Chinese room" thought experiment, challenge. Searle image simulation expert system for symbolic computation, said he did not understand Chinese, hiding in the room equipped with many words in the film, received Chinese outside note rules he has booked corresponding to the selected text in the piece, in response, will give people a feeling of thinking in Chinese. Computers, too, but according to the rule about characters, no matter how magical performance, in fact, not thinking.
According to symbolism thought AI was then built, that cognition and thinking of the basic unit is the symbol, symbols for knowledge according to the syntactic form strings expressed as propositions, cognition is based on the process of dealing with a string of formal logic, research how best to search the knowledge base for its solution. Searle questioned the soul through the mechanical calculations to describe.
He believes that thinking is not a simple logical operations, must be determined by intrinsic causal mechanisms and the power of intention to drive. When asked, when meat and steamed corn bread that tastes better, floating Chinese mind showing both images, aroma and flavor, driven by the preference intentions, organized into the language. Expert system and the Chinese room Searle, meat packages and Wo TOU are not really feeling, only syntax, in search of the best word answer. Although vivid, but not understand Chinese, no idea.
Expert system of researchers, human rational thinking is not according to Searle manner. When people ask what 1+2 How could think of an Apple and two apples, may also be recited answers. But the need for more rational problem, such as when asked 7x8, you are 99 table to answer. Scientific concepts, such as atomic, electromagnetic waves, gene, transcendental numbers, people are not direct emotional experience, all of this in the answer to question, is based on the book of memory and reasoning. Does this kind of work is also far from understanding and thinking?
However, Searle was the philosopher, his concern is not constructed according to scientific theory and logic reasoning, that is considered to be an objective scientific understanding of the world. He is concerned about is consciousness and thought, the interaction between the human heart and the external environment, which relates to the semantic content of perception, thinking about causal mechanisms and intentions in this process driven.
Searle said: the program is entirely in the form of work (words, syntactic); people's thinking is mental content (semantic, semantics); statement forms are neither by the semantic content, is not enough to determine the semantics of the content. Operation do not touch under the computer program language meaning does not constitute thought, also does not adequately reflect the idea.
Even if the knowledge base of the expert system, like the dictionary with sentence explanation of symbols, one integrated parsing and interpretation of their experience can be semantic. But this is only for human readers, so, the computer itself is just a formal reasoning tool, for a variety of strings according to the rules of syntax and logic symbols under operation, explained the sentence is true even for the knowledge base, the machine does not need not understand, is form of string processing, machine operation has nothing to do with the semantics of these strings.
Back said, abstract formal system does not, but the realization system of the machine is associated with the contents of the string. When answering reasonable questions, both humans and machines in the sense of an abstract machine is a form of reasoning, but to achieve the function of the system is different. In procedural terms, it is only logical analysis of syntax, search for the best answer. But run program of computer systems is by many door circuit composition, each string are has it physical state of meaning, this is a meet sentences expression of semantic, despite this semantic and understanding this statement of human interpretation different, but cannot denied computer systems of answers is in it of semantic interpretation Xia, according to best reply of "intention" drive, according to electronic circuit and computer systems Xia meaning of physical causal mechanism to operation of. Humans and computers, rational problems in analog form on such reasoning systems, although with different semantic interpretation, but the output should be no different. Intelligent judge should be neutral and not on human chauvinism, discrimination of different mechanisms and intentions. Michael Kors usa
But for emotional problems, answers to many questions is not logical, this need is the intelligent pattern recognition, for associate judge. System is clumsy and inflexible, and human performance difference is huge. In reply to reason based on the symbolism of the Chinese room could pass the Turing test, perceptual problems do not. Analyzing and processing for the input statement, associated semantic content causing the differences. This waited until the Connectionist machines appear to be cut.
Apart from Serre in philosophical questions, another challenge from mathematics of computationalism. Since the equivalent to the Turing machine of modern computers is implementing a formal system, Gödel's incompleteness theorem says, a self-consistent system contains arithmetic, some proposition cannot be proven, which means that the human mind can "see" a proposition true or false but cannot be proved by reasoning in the form. Therefore the human mind and cannot be owned without contradiction between knowledge systems in the form described. This does indicate limitations based on the logic of intelligent systems, AI Expert System stuck.
These "classical computational theory of the mind (CCTM)" challenge the background, is the era of computers can only play backgammon in the lab, and now more than 30 years, model regression of connectionism in mind, mainstream AI from symbolism, go to study in depth the connectionism on the mechanism of the machine. Its leading representatives are artificial neural networks, a neuron has input and multiple neuron connections, to their cumulative weight compared with the threshold, its function is to enter values in multivariate space, draw a straight line for a fuzzy judgement. Change connection input weights can be changed with this tendency. Neural networks are connected by the large amount of neurons, and linked the weights can be adjusted through the study of samples in order to network to maximize knowledge with these samples. Neural network learning was obtained as well as human, be able to extrapolate, from here, Lenovo speculated that "sensibility" cognitive abilities. In recent years, with the core of Connectionist machine intelligence has not lose human recognition sounds and images, beat humans in Board intellectual confrontation.
2014 United Kingdom University of reading held a "Turing test", Russia team developed the "Youjin·guziman" software, successfully passed the test, let 33% the referee believed it was a Ukraine 13-year old boy.
Some doubted, the artificial neural network is still not equivalent to the Turing machine implemented on a computer, please? In theory, it is still a form of inference system, how does it not limited by Gödel's incompleteness theorem? In theory, less semantic language forms of operation, will be as described in the sale, should have no essential breakthrough, right? The answer is "simulated". Here is the bit (bits) as a unit of operation system, and in the lower support for simulation of neural network system connections parallel operation. As the human brain at the lower level support logic simulation. Hand calculations and computer work in a finite world, but infinite worlds can be simulated. Cannot achieve unattainable to infinite in the finite world of things in the world, but with a limited number of counts, as well as the computer with limited circuit State, any function that requires precision analog expression of irrational numbers. Computer simulation of neural network system as well. Form action overlook semantics, but neural network system to answer questions, not just forms of operation, the system not only simulate the logical reasoning, also in lower-level simulation of semantic associations.
Searle laughed, as I said, I'm back in 1980 with "Chinese room" shows that the intelligent machines is just a mock up of the play. Copying and simulation are different concepts, "understanding" and "understanding analog" must not be confused!
This is plausible, but it is difficult to read between the lines. Metal artificial heart is simulated, and copied the heart function? Walking is usually implemented with legs of biological phenomena, the prosthetic legs are really going, walking or in simulation? Metal legs of the robot? Scientific classification of any concept is an abstract, grasp the nature of some of its common properties, and to ignore differences in detail. Some philosophers believe that if all properties are a concept depends on the functional structure, and not other details then copy with simulation and implementation of these same properties are the same thing. It would appear that the Turing test is already hitting intelligent nature. Calculation of no error in principle. As for Searle's Chinese room and symbolism AI unable to read minds, and it's only the limitations of formal reasoning system, able to pass the Turing test universal machine (Oracle machine) is not the case.
From an evolutionary point of view, the said function is not sufficient to establish the soul is also untenable. Because bio-dependent functionality to face competition on natural elimination, if the mind has nothing to do with the feature, how can it be in evolution, from low simple substances and be preserved?
Classical computational theory of the mind (CCTM) program in the form of symbolism is clear reasoning, people along the path of serial programs run, details the process of analysis to understand the performance of functions. Sharp makes it a research tool. Due to its limitations, and now mainstream turn to Connectionist models.
Connectionism achievement, micro-mechanism is extremely simple and clear, but a huge amount of units, complex connection, parallel computing, but let the people lost on the track and integrated. People know but don't know how, and also lost to development theory, can only be found in simulating natural crawl. Achievements on this path, technology breakthrough progress according to the trend, is not difficult to build a foot to stay human intelligent machines, but we lost the tool of analysis, in-depth understanding of heart problems is still at a loss.
"References"
1、Stanford Encyclopedia ofPhilosophy:The Computational Theory of Mind
2、Internet Encyclopedia ofPhilosophy:The Computational Theory of Mind
Lei Feng network Note: reprints please contact the authorized and retain complete information may not delete, modify articles.
2460 votes
Microsoft Surface 3 Michael Kors iPhone 5 Case
New atom performance ARM processors have and top closer, but are worse and the mainstream core of the series by a mile, but taking into account the thermal power of 14W and 2W gap, which is understandable. Use it for routine minor Office enough and finally surface series can also run X86 program! Said performance flaw, is that it does not use the SSD, and broken eggs eMMC 4.5, which even with the eMMc 5.0 modern Android flagship.
View details of the voting >>
No comments:
Post a Comment