Published on

  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide


  1. 1. Artificial Intelligence & Applications A. S. Md. Kamruzzaman Binghamton University Abstract: When an ordinary person thinks formed on the basis of those programs. The about a computer, he would immediately say storing of programs allowed the computer to that computer is a machine with programs change function quickly and easily by running that cannot act like a human brain. If a a new program. This capability implies that a computer is asked to make a decision on computer might be able to think and learn by certain aspects, it gives the result. It should be itself. impossible for a machine to act like a human AI mostly came from the thinking intelligence. But in fact, when “Deep Blue” perspective of a computer. A computer can (International Business Machines [IBM] think of itself and can produce decisions. software) defeated, Gerry Kesparov, World Dealing with a decision accordingly, a Champion Chess Grandmaster for six times computer can exhibit behavior similar to a within ten rounds, it was a big surprise.. The person. The key point is some programs are idea of acting computers like a human being definitely intelligent. The problem is whether came from Artificial Intelligence (AI). The a machine can think or a program running on most common and important areas of AI are a computer can be intelligent. Most programs searching (for solutions), expert systems, do not perform the same tasks in the same natural language processing, pattern way that a person does. An intelligent recognition, Robotics, machine learning logic, program should act like a human being and uncertainty and “fuzzy logic”. AI provides a exhibit behavior when confronted with a wide range of ideas of how software can similar problem. It is not necessary in the simulate human behavior. Artificial same way that the programs actually solve, or Intelligence deals with a specific kind of attempt to solve problems that a person software that relates to human activities. This would. is a branch of Computer Science that is This paper will provide an overview mostly concerned with the study and creation survey of AI while focusing on areas of of computer systems that exhibit some forms research and applications. A history of of intelligence: systems that learn new Artificial Intelligence will be given to readers concepts and tasks. AI has systems that can to overview ideas. This paper will also provide reason and draw useful conclusions about the the visual parts of Artificial Intelligence surroundings, systems that can understand a related to human beings. The paper will natural language or perceive and comprehend conclude on future perspectives developed by a visual scene, and systems that perform other AI. types of feats that require human types of intelligence. AI has some terms that should be understood from the Artificial Concepts such I. Introduction as, intelligence, knowledge, reasoning, thought, cognition, and learning. The Turing machine is considered the first Artificial Intelligence that works on the = = AI stored program computers. During the early MACHINE age of computers, there were actually machines that literally had to be rewired to solve different problems. Turing’s recognition was that those programs could be stored as data in the computer’s memory and could be executed later. All modern computers are
  2. 2. Figure 1. A robot is taking solution from an it can be explained by a reduction to AI machine [1] ordinary physical processes. Philosophers staked out most of the Humankind has given itself the important ideas of AI, but to make the scientific name Homo Sapiens which leap to a formal science required a level means “man the wise”. This is so of mathematical formalization in three because human mental capacities are so main areas: computation, logic and important to everyday life. The field of probability. In computation theorem, AI attempts to understand intelligent intractability, reduction, NP (Non entities. Thus, one reason to study AI is Probabilistic) completeness and decision to learn more about humankind. But theory has a great impact on AI which unlike Philosophy and Psychology, arose from math. Psychology plays which are also concerned with another major role in AI. Behaviorism intelligence, AI strives to build started discovering animals’ brain and intelligent entities as well as understand cognitive psychology started on brain them. Another reason to study AI is that possesses and processing information. these constructed intelligent entities are Kenneth Craik [2] made a connection interesting and useful in their own right. between stimulus and response. The AI has produced many significant and three key steps of a knowledge based on impressive results even at this early agent; (1) the stimulus must be translated stage in its development. Although no into an internal representation; (2) the one can predict the future in detail, it is representation is manipulated by clear that computers with the level of cognitive processes to derive new human intelligence would have a huge internal representations, and (3) these in impact on the future course of turn are retranslated back into action. civilization. In Figure 1, this idea is Computer Engineering catalyzed ideas illustrated visually. of AI. The computer has been In terms of Philosophy, AI has unanimously acclaimed as the artifact inherited many ideas, viewpoints, and with the best chance of achieving techniques from other disciplines. The artificial intelligence. The Turing idea story of AI began around 450BC [2]. changed the vision of AI. The idea of "When Plato reported a dialogue in knowledge representation (the study of which Socrates asks Euthyphro,"I want how to put knowledge into a form that a to know what is characteristics of piety computer can reason with) was tied to which makes all actions pious ... that I language and informed by research in may have it to turn to, and to use as a linguistics that was connected to standard whereby to judge your actions philosophical analysis language. and those of other men."[2] Dualism Figure 2 can make the ideas more which is part of the mind (or soul or clear. It shows how AI could be spirit), is outside of nature exempts from described in terms of the environment physical laws. The mind or brain holds and an agent. An agent is anything that the entire world which operates can be viewed as perceiving the according to physical laws. It is also environment through sensors and acting possible to adopt an intermediate upon that environment through effectors. position in which one accepts that the A human agent has eyes, ears and other mind has a physical basis, but denies that organs for sensors, and hands, legs,
  3. 3. mouth, and other body parts for field of AI. Newell and Simon wrote a effectors. A robotic agent substitutes reasoning program which is capable of cameras and infrared range finders for thinking non-numerically, and solved the the sensors and various motors for the mind-body problem. The name of effectors. Software agent has encoded bit Artificial Intelligence came from the strings as percepts and actions. A generic Darmouth Workshop [2]. agent is diagrammed in Figure 2. Early enthusiasm great expectations (1952 – 1969): Newell and PERCEPTS Simon’s [2] early success was followed SENSORS up with the General Problem Solver or (GPS). This was probably the first ENVIRONMENT program to embody the “thinking AGENT humanly” approach. Starting in 1952 [2], ACTIONS Arthur Samuel wrote a series of programs for checkers (draughts) that eventually learned to play tournament EFFECTORS level checkers. He discovered the idea Figure 2. Agents interect with environments that computers can only do what they are through sensors and effectors[2] told to do. Early work building on the neural networks of McCulloch and Pitts also flourished. The work of Winograd and Cown (1963) showed how a large II. History number of elements could collectively represent an individual concept, with a It is difficult to pinpoint an exact corresponding increase in robustness and starting date for the invention of AI. It parallelism. began to emerge as a separate field of A dose of reality (1966 - 1974): study during 1940 and 1950s when the Weizenbaum’s ELIZA program (1965) computer became a commercial reality. [2] which could apparently engage in Here is the subdivision of some part of serious conversation on any topic, AI history. actually just borrowed and manipulated The gestation of AI (1943 - the sentences typed into it by a human. 1956): Warren McCulloch and Walter The illusion of unlimited computational Pitts (1943)[2] drew on three-source power was not confined to problem- knowledge of the basic physiology and solving programs. Early experiments in function of neurons in the brain. They machine evolution (new called genetic proposed a model of artificial neurons in algorithms) were based on the which each neuron is characterized as undoubtedly correct belief that by being “on” or “off”, with a switch to making an appropriate series of small “on” occurring in response to stimulation mutations to a machine code programs, by a sufficient number of neighboring one can generate a program with good neurons. This work was arguably the performance for any particular simple forerunner of both the logicist tradition task. This idea, then, was to try random in AI and the connectionist tradition. In mutations and then apply a selection the early 1950s, the invention of neural process to preserve mutations that network computer opened another new seemed to improve behavior. During
  4. 4. this time there was certain difficulties. framework. There have been a number The first difficulty was that programs of advances that built upon each other had insufficient knowledge of their rather than starting from scratch each subject matter. Secondly the time. Probabilistic Reasoning in intractability of many of the problems Intelligent Systems marked a new that AI programs worked by representing acceptance of probability and decision the basic facts about a problem and theory in AI, following a resurgence of trying out a series of steps to solve it. interest in formalism was invented to The theory of NP (Non Probabilistic) allow efficient reasoning about the completeness brought the problem. This combination of uncertain evidence. This theory could not able to bring the approach largely overcomes the problem solutions for problems. Third difficulty with probabilistic reasoning systems of came from fundamental limitations on the 1960s and 1970s. It also has come to the basic structures being used to determined AI research on uncertain generate intelligent behavior. reasoning and expert systems. Similar Knowledge based systems revolutions have occurred in robotics, (1969 - 1979): The widespread growth computer vision, machine learning of applications to real-world problems (including neural networks) and caused a concomitant increase in the knowledge representation. A better demands for workable knowledge understanding of the problems and their representation schemes. A large number complexity properties, combined with of different representation languages increased mathematical sophistication, were developed. Some were based on has led to workable research agenda and logic. For example the Prolog language robust methods perhaps encouraged by became popular in Europe, and the the progress in solving the sub problems PLANNER family [2] in the United of AI, researchers have also started to States. look at the “whole agent” problem again. Neural Networks (1986 - present): Some disillusionment was occurring concerning the applicability III. Definition of AI the expert systems technology derived from MYCIN- type systems [2]. Many AI is a branch of Computer corporations and research groups found Science concerned with the study and that building a successful expert system creation of computer systems. AI involved much more than simply buying exhibits some form of intelligence: a reasoning system and filling it with systems that learn new concepts and rules. Some predicted an “AI Winter” in tasks, systems that can reason and draw which AI funding would be squeezed useful conclusions about the world. AI severely. systems can understand a natural Recent events (1987 - present): language or perceive and comprehend a Speech technology and the related field visual scene, and systems that perform of level written character recognition are other types of feats that require human already making the transition to types of intelligence. Intelligence is the widespread industrial and consumer integrated sum of those feats which applications. An elegant synthesis of gives us the ability to remember a face existing planning programs into a simple not seen for thirty or more years, or to
  5. 5. build and send rockets to the moon [3]. to try to construct precise and testable The intelligence requires knowledge. AI theories of the workings of the human is not the study and creation of mind. The rational thought which conventional computer systems. govern the operation of the mind, and From the perspective of initiated the field of logic. Acting intelligence; AI makes machines rationally is another part of the definition "intelligent" -- acting, as we would of AI which means acting so as to expect people to act. The inability to achieve one’s goals, given one’s beliefs. distinguish computer responses from An agent is something that perceives and human responses is called the Turing acts. If we look at AI impressive test. Intelligence requires knowledge achievements, it is still impossible to Expert problem solving - restricting produce the brain abilities of a three- domain to allow including significant year-old child. These include the ability relevant knowledge [4]. to recognize and remember numerous From a research perspective: diverse objects in a scene, to learn new artificial intelligence is the study of how sounds and associate them with objects to make computers do things which, at and concepts and to adopt readily to the moment, people do better. One way many diverse new situations. to measure the success of AI within computers is to interrogate it by a human via a Teletype. The computer passes the IV. AI Performances test if the interrogator cannot tell AI has performed a vital role in whether there is a computer or a human so many fields. Researchers are devoting at the other end. The computer needs to their time and effort to establish a good posses the following capabilities [4]. performance on AI. The features of AI Natural language processing - to enable can illustrate some ideas of AI it to communicate successfully in performance. English (or some other human Knowledge representation: It is a language). design for knowledge–based agent. A simple logical language for expressing 1. Knowledge representation- to store knowledge and showing how it can be information provided before or during used to draw conclusions about the the interrogation. world and to decide what to do. The 2.Automated reasoning – to use the language is capable of expressing a wide stored information to answer questions variety of knowledge about complex and to draw new conclusions. worlds. It could be represented several 3. Machine learning – to adapt to new ways. circumstances and to detect and 1.Knowledge Acquisition- extrapolate patterns. formalizing knowledge and 4. Computer vision - to perceive objects implementing knowledge bases are and robotics to move them about. major tasks in the construction of large A computer program has to think like a AI systems. The hundreds of rules and human. The interdisciplinary field of thousand of facts required by many of cognitive science brings together these systems are generally obtained by computer models from AI and interviewing expert in the domain of experiment techniques from psychology application. Representing expert
  6. 6. knowledge as facts or rules is typically a to confer the ability (or reasons about tidious and time-consuming process. one’s own knowledge). Techniques for automating this Expert systems: these constitute knowledge acquisition process would most of AI’s commercial success. Expert constitute a major advance in AI Systems are programs that mimic the technology. Knowledge acquisition can behavior of a human expert. They use automate in three ways [5]. Firstly, information that the user supplies to special-editing systems might be built sender an opinion on a certain subject. that allow persons who possess expert The expert system asks user questions knowledge about the domain of until it can identify an object that application to interact directly with the matches with the answer from the user. knowledge bases of AI systems. Secondly, advances in natural language For example: processing techniques will allow humans Expert: Is it green? to instruct and teach computer systems Users: No. through ordinary conversations. Thirdly, Expert: Is it red? AI systems might learn important Users: Yes. knowledge from their experiences in Expert: Does it grow on a tree? their problem domains. User: No. Representational Formalisms; Expert: Does it grow on a cane? 2. Commonsense reasoning- Many of User: Yes. the existing ideas about AI techniques Expert: Does the cane have thorns? have been refined on “toy” problems, User: Yes. such as problems in the ‘block worlds’, Expert: It is a raspberry. in which the necessary knowledge is Every expert system has two parts reasonably easy to formalize. [6]: the knowledge base and the Representing Prepositional Attitudes reference engine. The knowledge base is [5]- a database that holds specific information and rules about a certain San knows that Pete is a lawyer. subject. The inference engine is the San does not believe that John is a information that the user supplies to find doctor. an object that matches. It has two Pete wants it to rain. branches; 1.Deterministic, and John fears that Sam believes that the 2.probabilistic. morning star is not Venus. The interface engine can also be The underlined portions of these defined as the forward-chaining method sentences are propositions, and the and the backward–chaining method. relations know, believe etc. refer to The Forward-Chaining Method: attitudes of agents toward these Forward-chaining is sometimes called propositions. A logical formalization for “data-driven”[6] because the inference expressing the appropriate relations engine uses information that the user between agents and attitudes. provides to move through a network of 3. Meta–Knowledge – A good logical AND and OR until it reaches a solution to the problem of reasoning terminal point, which is the object. about the knowledge of others ought also In Figure 3, a fruit knowledge base creates a Forward-chaining interface
  7. 7. engine. The engine would arrive at the Understanding commands written in object apple, when it is given the proper standard human languages. NL processor attributes as shown in Figure 3. A extract information from any given input. The core of any NLP system is the Round Grows on trees parser. The parser is the section of code that reads each sentence, word by word Does not grow in Deep South and decide what is what. The example of a parser. Red or yellow The State Machine Parser: The state- machine parser uses the current state of the sentence to predict what type of word Apple may legally follow. Figure 5 shows the state machine that is a directed graph Figure 3. Forward-chaining to the object that shows the valid transitions from one apple[9] state to another. For example, a noun can be followed by a verb or a preposition. A Forward-chaining system essentially state machine is shown in Figure 5. builds a tree from the leaves down to the root. The Backward-Chaining Method: Noun Backward-chaining is the reverse of forward-chaining. A backward-chaining Preposition inference engine starts with a hypothesis Adjective Verb (an object) and request information to confirm or deny it. In Figure 4, the fruit Adverb Try apple Figure 5. The state-machine of the restricted grammar [9] Grows on trees Grows on vine Is round Red or yellow Vision and Pattern Recognition: Vision systems can be Does not grow in Deep South implemented in two ways. First method Is orange Apple tries to reduce an image to the lines that form the outline of each object. This Figure 4. Backward-chaining to the object apple method uses various filters to remove [9] information from the image, and contrast enhances to make all parts of the image either black or white. They are called Question is an apple, applying binary image because every paint in the backward-chaining inferences to the fruit image is either black or white. Second knowledge base. As the diagram shows, method attempts to give the computer a backward-chaining prunes a tree. more humanlike view of the image. This Natural Language processing: method gives to the computer Natural Language Processing (NLP) information about the brightness of the tries to make the computer capable of part of the image. It allows the computer
  8. 8. to derive two important features from the information in a database. Role does not image that are not possible with a light require any generalization to be derived contrast image because of surfaces and or any high level thinking. shadows. It is easy to interpret an image Cognitive learning – This form of but correctly identifying the objects or learning requires analyzing, organizing, features that make up the image. There and correlating specific pieces of are several ways to do it. Firstly, knowledge. The product of this mental computer can do it by controlled effort is the creation of class hallucination [6]. Recognition of object descriptions. The ability to learn class is also another issue in the pattern description is fundamental to the recognition. creation of a computer that thinks the Robotics: There are two types of way that human does. robots. Industrial assembly robots are Logic and uncertainty: Logic lies used in a controlled environment. It can at the heart of computer programming. A perform only programmed task. There programming language is simply an are two ways to teach a new task to a implementation of a special form of robot: knowledge. One of the most pleasing aspects of logic is that it is certain. using teach pendant, or Things that are represented by logic are 2.programmed by using a Robotic- ‘true’ or ‘false’. Resolving or dealing control language. with uncertainty is critical to machine Teach pendant is a hand-held control intelligence because it is required for box that allows an operator to move the successful interfacing with the real various joints of the robot. It is linked world. Fuzzy logic deals with the through the robot’s main control evaluation of logical expression that computer. Moving each joint can do it contain uncertain values probabilistic and computer records each position [6]. systems utilizes the probability of the For complex jobs Robotic Control occurrence of various events in order to Language s are used. It is a computer arrive at an answer [6]. program used to control a robot. Appearing human: The idea is to Autonomous Robots: It is much more make a computer appear to be like a complex than industrial robots. It sensors person. It is completely integrated that allows to hear and see and program which appears to be human. understand natural language and what The name of this program is ELIZA. If a the language means. It can also solve human is compared with a computer problems. It can be implemented by (program), the human has the emotions parallel processing but it is still under and the personality. In terms of research. computer, a machine can not act like a Machine learning: Two types of human. The program which is created by learning: role learning and cognitive the human being, can do whatever the learning. programmer wants. It can appear to have Role learning is something from emotions and personality because the memorization. In terms of computer, it is programmer built the program with those a set of instruction programmed in a functions in it. This way the computer database, which can easily follow a shows that it has emotions and procedure or store some item of personality.
  9. 9. AI is the field where human brain V. Application of AI and machine talks together. The importance of AI is very wide. Human The distinction between a brain can be transformed into a machine computer “user” and a computer format and all the research is done “programmer” is that the user provides through AI. Cognitive Psychology and new input, or data (words or numbers), AI are very related. Cognitive while the programmer defines new Psychology discusses on human operations, or programs, as well as new behavior and AI deals how to transform types of data [7]. machine close to human. The GPS developed in 1957 by Alan The invention of supercomputers Newell and Hervert Simon, embodied a is one of the great inventions of AI grandiose vision. A single computer which changed the view of AI. With a program that could solve any problem, combined budget of about one billion given a suitable description of the dollars [3], the Japanese are determined problem GPS caused quite a stir when it to realize many of their goals, namely, to was introduced and some people in AI produce systems that can converse in a felt it would sweep in a grand new era of natural language, understand speech and intelligent machines. It is much easier to visual scenes, learn and refine their implement a GPS in steps. There are few knowledge, make decisions and exhibit steps which are [7] other human traits. The Defense Advanced Research Projects Agency a. Describe the problem in vague term (DARPA) has increased it’s funding for b. Specify the problem in algorithmic research in AI. In addition, most of the terms. larger high-tech companies such as IBM, c. Implement the problem in a DEC, AT&T have their own research programming language. programs. d. Test the program on representative examples. VII. Future Perspective e. Debug and analyze the resulting program and Lots of plans have taken to f. Repeat the process. improve the research of AI. At the same The main programming languages used time, funding is increased to improve its in AI are Lisp and Prolog. Both have standing. Researchers are trying to get features which make them suitable for the Autonomous Robots which will AI programming, such as support for list change the entire AI field. processing, pattern matching and : (1) Reducing the time and cost of exploratory programming. Both are also development is a big plan for AI. widely used -Prolog especially in Europe (2) Allowing students to work and Japan, and Lisp in the US. This wide collaboratively is another plan from use within the field is another reason to researchers. choose Lisp or Prolog for AI One important research issue is implementation [8]. reducing the time and cost in order to develop such systems. Current strategies for doing this include the development VI. Importance of AI of authoring tools and creating systems in a modular fashion. Solving this
  10. 10. problem will be an enormous environment but also on the actions of breakthrough in ITS research, since other agents. The standard scenario more systems could be constructed and involves a set of agents who make their thus more research into the effectiveness decisions simultaneously, without of computer based instruction could be knowledge of the decisions of the other performed [9]. agent. AI is trying to discover some desirable property P; There are number VIII. Conclusion of choices for p which are, Perfect rationality: the classical A computer is a game device to a notion of rationality in decision theory. child, but it can be used in different ways A perfectly rational agent acts at every depending on the user’s needs. instant in such a way as to maximize its Programmers build all kinds of programs expected utility, given the information it to satisfy the needs of the growing has acquired from the environment. number of users. It will not be surprising Calculate rationality: the notion of to utilize computers with highly rationality that has used implicitly in developed artificial intelligence designing logical and decision capabilities a few years from now to do theoretical agents. A calculatively thee tasks. Computers will have the rational agent eventually returns what ability to create a program that can would have been the rational choice at create another program and thus simulate the beginning of its deliberation. This is the behavior of the human brain. The an interesting property for a system to best example thus far of these exhibit because it continues an “in- capabilities were documented by Deep principle” capacity to do the right thing. Blue in a chess game that serves as a Bounded optimality: A bounded technological landmark for the future. optimal agent behaves as well as The shear knowledge of a computer possible given its computational adapting and functioning faster than the resources. That is, the expected utility of human mind though programmed to do the agent program for a bounded optimal so by humans is in essence a frightening agent is at least as high as the expected reality that needs to be confronted. utility of any other agent program Though AI just finished with its period running on the same machine. of infancy, it has ramifications that yet Philosophy has also seen a remain unknown. to everyone. The effort gradual evolution in the definition of and research can bring the surprising rationality. There has been a shift from innovations that the majority crave and consideration of act utilitarianism – the desire but there are also results which rationality of individual acts – to rule cannot be forseen when the computer utilitarianism, or the rationality of begins to think for itself. general policies for acting. Another area is game theory, a branch of economics that began References widespread study of decision theory. [1] My own Creative picture. Game theory studies decision problems [2] Norvis, Peter &Russel, Stuart Artificial in which the utility of a given action Intelligence: A modern Approach, Prentice Hall, depends not only on chance events in the NJ, 1995
  11. 11. [3] Patterson, Dan W. Introduction to Artificial Intelligence and Expert Systems, Prentice Hall of India Private Limited New Delhi, 1998 [4] Brown, Carol E. and O'Leary, Daniel E. “INTRODUCTION TO ARTIFICIAL INTELLIGENCE AND EXPERT SYSTEMS” Artificial Intelligence / Expert Systems Section of the American Accounting Association , es_tutor.htm#1-AI 01/03/2000 [5] Nilsson, Nils J. Principles of Artificial Intelligence, Narosa Publishing House New Delhi, 1998 [6] Schildt, Herbert Artificial Intelligence Using C, Osborne McGraw Hill Berkeley, California,, 1987 [7] Norvig, Peter Paradigms of Artificial Intelligence Programming Morgan Kaufmann Publishers, San Mateo, California 1992 [8] Cawsey, Alison Databases and Artificial Intelligence 3 Artificial Intelligence Segment, ml 08/ 19/1994 [9] Beck, Joseph, Stem, Mia and Haugsiaa “Applications of AI in Education” The ACM's First Electronic Publication l 02/13/ 2000 A. S. MD. KAMRUZZAMAN; received an Associate in Science from LaGuardia C. College/CUNY in Computer Science in August 1999 CAREER OBJECTIVE: Webmaster ORIGIN: Bangladesh HONORS from CUNY: National & College Dean's List ‘97 – ‘99 Vice-president-Phi Theta Kappa International Honor Society ‘98 – ‘99. College Senator, Student Govt. Association ‘98 – ‘99. Student Advisory Council - Foreign Student Club, Asian Club, Alpha Theta Phi ‘97 – ‘99. AWARDS: Leadership Award’98 Honors Award for Academia from the Dept. of Student Affairs’98 & ‘99 Student Govt. Association Award’98. My Bio and Web Creations -- http://