The phrase "artificial intelligence" was coined by John McCarthy three decades ago. The current (fifth) generation computers can process natural languages, play games, recognize images of objects and prove mathematical theorems. A better definition of AI calls for formalization of the term "intelligence"
The phrase "artificial intelligence" was coined by John McCarthy three decades ago. The current (fifth) generation computers can process natural languages, play games, recognize images of objects and prove mathematical theorems. A better definition of AI calls for formalization of the term "intelligence"
Direitos autorais:
Attribution Non-Commercial (BY-NC)
Formatos disponíveis
Baixe no formato DOC, PDF, TXT ou leia online no Scribd
The phrase "artificial intelligence" was coined by John McCarthy three decades ago. The current (fifth) generation computers can process natural languages, play games, recognize images of objects and prove mathematical theorems. A better definition of AI calls for formalization of the term "intelligence"
Direitos autorais:
Attribution Non-Commercial (BY-NC)
Formatos disponíveis
Baixe no formato DOC, PDF, TXT ou leia online no Scribd
ARTIFICIAL started to produce the fifth generation
computing machines that, besides having INTELLIGENCE all the capabilities of the fourth generation machines, could also be able to process Abstract: intelligence. The computers of the current At the beginning of the Stone Age, when (fifth) generation can process natural people started taking shelters in caves, languages, play games, recognize images they made attempts to immortalize of objects and prove mathematical themselves by painting their images on theorems, all of which lie in the domain of rocks. With the gradual progress in Artificial Intelligence (AI). But what exactly civilization, they felt interested to see is AI? themselves in different forms. So, they Defining AI: started constructing The phrase AI, which was coined models of human being with sand, clay by John McCarthy three decades and stones. The size, shape,constituents ago,evades a concise and formal definition and style of the model humans continued to date. One representative definition is evolving but the man was not happy with pivoted around the comparison of the models that only looked like him. He intelligence of computing machines with had a strong desire to make the model human beings]. Another definition is ‘intelligent’, so that it could act and think concerned with the performance of as he did. Babbage’s analytical engine machines which “historically have been was thefirst significant success in the judged to lie within the domain of modern era of computing. Computers of intelligence” [17], [35]. None of these the first generation, which were realized definitions or the like have been following this revolutionary success,were universally accepted, perhaps because of made of thermo-ionic valves. They could their references to the word “intelligence”, perform the so-called ‘number crunching’ which at present is an abstract and operations. The second-generation immeasurable quantity. A better definition computers came up shortly after the of AI, therefore, calls for formalization of invention of transistors and were more the term “intelligence”. Psychologist and miniaturized in size. During the period of Cognitive theorists are of the opinion that intelligence helps in identifying the right starting state and measure the total cost piece of knowledge at the appropriate for reaching the goal from the given instances of decision making.The phrase starting state. Let this cost be f. While “AI” thus can be defined as the simulation f = a predefined utility value and the goal of human intelligence on a machine, is not reached, new nodes are generated The above line of thinking has been as children of the current node. However, realized in many intelligent search in case all the neighborhood nodes problems of AI. Some of these well-known (states) yield an identical value of f and search algorithms are: the goal is not included in the set of these a) Generate and Test nodes, the search algorithm is trapped at b) Hill Climbing a hillock or local extrema. One way to c) Heuristic Search overcome this problem is to select d) Means and Ends analysis randomly a new starting state and then (a) Generate and Test Approach: This continue the above search process. While approach concerns the generation of the proving trigonometric identities, we often state-space from a known starting state use Hill (root) of the problem and continues Climbing, perhaps unknowingly. expanding the reasoning space until the (c) Heuristic Search: Classically goal node or the terminal state is reached. heuristics means rule of thumb. In In fact after generation of each and every heuristic search, we generally use one or state, the generated node is compared more heuristic functions to determine the with the known goal state. When the goal better candidate states among a set of is found, the algorithm terminates. In case legal states that could be generated from there exist multiple paths leading to the a known state. The heuristic function, in goal, then the path having the smallest other words, measures the fitness of the distance from the root is preferred. The candidate states. The better the selection basic strategy used in this search is only of the states, the fewer will be the number generation of states and their testing for of intermediate states for reaching the goals but it does not allow filtering of goal. One has to select them intuitively, states. so that in most cases hopefully it would be (b) Hill Climbing Approach: Under this able to prune the search space correctly. approach, one has to first generate a (d) Means and Ends Analysis: This knowledge from Philosophy, Psychology, method of search attempts to reduce the Cognitive Science, Computer Science, gap between the current state and the Mathematics and Engineering. Thus in fig. goal state. One simple way to explore this 1.2, they have been referred to as the method is to measure the distance parent disciplines of AI. An at-a-glance between the current state and the goal, look at fig. 1.2 also reveals the subject and then apply an operator to the current area of AI and its application areas. state, so that the distance between the resulting state and the goal is reduced. In many mathematical theorem- proving processes, we use Means and Ends Analysis. Besides the above methods of intelligent search, there exist a good number of general problem solving techniques in AI. Among these, the most common are: Problem Decomposition and Constraint Satisfaction. The Disciplines of AI Fig. 1.2: AI, its parent disciplines and The subject of AI spans a wide horizon. It application areas. deals with the various kinds of knowledge The Subject of AI representation schemes, different The subject of AI was originated with techniques of intelligent search, various game-playing and theorem-proving methods for resolving uncertainty of data programs and was gradually enriched with and knowledge, different schemes for theories from a number of parent automated machine learning and many disciplines. As a young discipline of others. Among science, the significance of the topics the application areas of AI, we have Expert covered under the subject changes systems, Game-playing, and Theorem- considerably with time. At present, the proving, Natural language processing, topics which we find significant and Image recognition, Robotics and many worthwhile to understand the subject are others. The subject of AI has been outlined below: enriched with a wide discipline of applications in supervised (parametric)learning problems. Besides this type, the other common learning methods,which we do unknowingly, are inductive and analogy-based learning. In inductive learning, the learner makes generalizations from examples. For instance, noting that “cuckoo flies”, “parrot flies” and “sparrow flies”, the learner generalizes that “birds fly”. On the Fig. 1.3: Pronunciation learning of a child other hand, in analogy-basedlearning, the from his mother. learner, for example, learns the motion of Learning Systems: Among the subject electrons in an atom analogously from his areas covered under AI, learning systems knowledge of planetary motion in solar needs special mention. The concept of systems. learning is illustrated here with reference Knowledge Representation and to a natural problem of learning of Reasoning: In a reasoning problem, one pronunciation by a child from his mother has to reach a pre-defined goal state from (vide fig. 1.3). The hearing system of the one or more given initial states. So, the child receives the lesser the number of transitions for pronunciation of the character “A” and the reaching the goal state, the higher the voice system attempts to imitate it. The efficiency of the reasoning system. difference of the mother’s and the child’s Increasing the efficiency of a reasoning pronunciation, hereafter called the error system thus requires minimization of signal, is received by the child’s learning intermediate states, which indirectly calls system through auditory nerve, and an for an organized and complete knowledge actuation signal is generated by the base.. A variety of knowledge learning system through a motor nerve for representation techniques are in use in adjustment of the pronunciation of the Artificial Intelligence. Production rules, child. The artificial neural networks, which semantic nets, frames, filler and slots, and represent the electrical analogue of the predicate logic are only a few to mention. biological nervous systems, are gaining importance for their increasing Example: A semantic net represents knowledge from given knowledge base, knowledge by a structured approach. For setting dynamic data structures for instance, consider the following existing knowledge, learning knowledge knowledge base: from the environment Knowledge Base: A bird can fly with and refinement of knowledge. Automated wings. A bird has wings. A bird has legs. A acquisition of knowledge by machine bird can walk with legs.The bird and its learning approach is an active area of attributes here have been represented in current research in Artificial Intelligence. figure 1.4 using a graph, where the nodes Intelligent Search: Search problems, denote the events and the arcs denote the which we generally encounter in Computer relationship between the nodes. Science, are of a deterministic nature, i.e., the order of visiting the elements of the search space is known. Soft Computing: Soft computing, according to Prof. Zadeh, is “an emerging approach to computing, which parallels the remarkable ability of the human mind Fig. to reason and learn in an environment of 1.4: A semantic net representation of "birds". uncertainty and imprecision” . It, in Planning: Another significant area of AI is general, is a collection of computing tools planning. The problems of reasoning and and techniques, shared by closely related planning share many common issues, but disciplines that include fuzzy logic, have a basic difference that originates artificial neural nets, genetic algorithms, from their definitions. The planning belief calculus, and some aspects of problem, on the other hand, deals with the machine learning like inductive logic determination of the methodology by programming. These tools are used which a successful goal can be achieved independently as well as jointly depending from the known initial states. on the type of the domain of applications. Knowledge Acquisition: Acquisition The scope of the first three tools in the (Elicitation) of knowledge is equally hard broad spectrum of AI is for machines as it is for human beings. It outlined below. includes generation of new pieces of . Fuzzy Logic: Fuzzy logic deals with unsupervised types of machine learning. fuzzy sets and logical connectives for The supervised learning algorithms modeling the human-like reasoning realized with ANN have been successfully problems of the real world It may be noted applied in control, automation, robotics that a conventional set contains its and computer vision. The unsupervised members with a value of membership learning algorithms built with ANN, on the equal to one and disregards other other hand, have been applied in elements of the universal set, for they scheduling , knowledge acquisition , have zero membership. The most common planning and analog to digital conversion operators applied to fuzzy sets are AND of data . (minimum), OR (maximum) and negation Applications of AI Techniques (complementation), where AND and OR Almost every branch of science and have binary arguments, while negation engineering currently shares the tools and has unary argument. The logic of fuzzy techniques available in the domain of AI. sets was proposed by Zadeh, who However, for the sake of the convenience introduced the concept in systems theory, of the readers, we mention here a few and later extended it for approximate typical applications, where AI plays a reasoning in expert systems . significant and decisive role in engineering Artificial Neural Nets: Artificial neural automation. nets (ANN) are electrical analogues of the Expert Systems: An expert system biological neural nets. Biological nerve consists of a knowledge base, database cells, called neurons, receive signals from and an inference engine for interpreting neighboring neurons or receptors through the database using the knowledge dendrites, process the received electrical supplied in the knowledge base. pulses at the cell body and transmit Image Understanding and Computer signals through a large and thick nerve Vision: A digital image can be regarded fiber, called an axon. An artificial neural as a two-dimensional array of pixels net is a collection of such electrical containing gray levels corresponding to neurons connected in different topology. the intensity of the reflected illumination The most common application of an received by a video camera . For artificial neural net is in achine learning. interpretation of a scene, its image should The ANN support both supervised and be passed through three basic processes: low, medium and high level vision speeds, lower than the maximum speed of (fig.1.5). The importance of low level the robot. In the near future, mobile vision is to pre-process the image by robots will find extensive applications in filtering from noise. The medium level fire-fighting, mine clearing and factory vision system deals with enhancement of automation. In accident prone industrial details and segmentation (i.e., partitioning environment, mobile robots may be the image into objects of interest). The exploited for automatic diagnosis and high level vision system includes three replacement of defective parts of steps: recognition of the objects from the instruments.. segmented image, labeling of the image Speech and Natural Language and interpretation of the scene. Understanding: Understanding of Recognition of objects from its image can speech and natural languages is basically be carried out through a process of two classical problems. In speech analysis, pattern classification, which at present is the main problem is to separate the realized by supervised learning syllables of a spoken word and determine algorithms. The interpretation process, on features like amplitude, and fundamental the other hand, requires knowledge-based and harmonic frequencies of each syllable. computation. The words then could be identified from Navigational Planning for Mobile the extracted features by pattern Robots: Mobile robots, sometimes called classification techniques. Recently, automated guided vehicles (AGV), are a artificial neural networks have been challenging area of research, where AI employed to classify words from their finds extensive applications. A mobile features. The problem of understanding robot generally has one or more camera natural languages like English, on the or ultrasonic sensors, which help in other hand, includes syntactic and identifying the obstacles on its trajectory. semantic interpretation of the words in a The navigational planning problem sentence, and sentences in a paragraph. persists in both static and dynamic The syntactic steps are required to environments. In a static environment, the analyze the sentences by its grammar and position of obstacles is fixed, while in a are similar with the steps of compilation. dynamic environment the obstacles may The semantic analysis, which is performed move at arbitrary directions with varying following the syntactic analysis, determines the meaning of the sentences The fuzzy and artificial neural network- from the association of the words and that based learning techniques have recently of a paragraph from the closeness of the been identified as new tools for plant sentences. A robot capable of estimation . understanding speech in a natural Conclusion: language will be of immense importance, The subject of AI deals more with symbolic for it could execute any task verbally processing than numeric computation. communicated to it. The phonetic Knowledge representation, reasoning, typewriter, which prints the words planning, learning, intelligent search and pronounced by a person, is another recent uncertainty management of data and invention where speech understanding is knowledge are thecommon areas covered employed in a commercial application. under AI. Some of the applications areas Scheduling: In a scheduling problem, one of AI are speech and image has to plan the time schedule of a set of understanding, expert systems, pattern events to improve the time efficiency of classification and navigational planning of the solution mobile robots. LISP and PROLOG are the Intelligent Control: In process control, usual languages for programming AI the controller is designed from the known problems. Because of severe non- models of the process and the required determinism. control objective. When the dynamics of References the plant is not completely known, the [1] Bender, Edward A., Mathematical existing techniques for controller design Methods in Artificial Intelligence,IEEE no longer remain valid. Fuzzy control has Computer Society Press been successfully used in many industrial [2] Besnard, P., An Introduction to Default plants. One typical application is the Logic, Springer-Verlag, Berlin,pp.27-35, power control in a nuclear reactor. Besides 1989. design of the controller, the other issue in [3] Bezdek. J. C., Ed., Pattern Recognition process control is to design a plant with Fuzzy Objective FunctionAlgorithms, (process) estimator, which attempts to Kluwer Academic Press, 1991. follow the response of the actual plant, [4] Buchanan, B. G. and Shortliffe, E. H., when both the plant and the estimator are Eds., Rule-Based ExpertSystems, Addison- jointly excited by a common input signal. Wesley, Reading, MA, 1984. [5] Buchanan, B.G. and Wilkins, D.C., Eds., Readings in KnowledgeAcquisition and Learning: Automating the Construction andImprovement of Expert Systems, Morgan Kaufmann, San Mateo, CA,1993.