Computer Science. The English Language Perspective - Беликова (1176925), страница 39
Текст из файла (страница 39)
Conventional AI is based on machine learning,which is the development of the techniques and algorithms thatallow machines to “learn” or at least simulate learning. Machinelearning attempts to use computer programs to generatepatterns or rules from large data sets. This problem is similar tothe data-mining problem (and data mining is one area where AIhas found commercial success). Machine learning makes heavy231use of symbolic formalism and logic, as well as statistics.
Keyareas in conventional AI include case-based reasoning,behavior-based AI, Bayesian networks, and expert systems.Computational intelligence, in contrast, relies more on cleveralgorithms (heuristics) and computation and less on formallogical systems. Computational intelligence is sometimesreferred to as soft computing.
It often involves iterativemethods using computation to generate intelligent agents.Whereas conventional AI is considered to be a top-downapproach, with the structure of solutions imposed from above,computational intelligence is more bottom-up, where solutionsemerge from an unstructured initial state.
Two areas ofcomputational intelligence will be discussed further: neuralnetworks and fuzzy logic. Hybrid intelligent systems attempt tocombine the two approaches. Some proponents claim that thisis appropriate, because the human mind uses multipletechniques to develop and verify results, and hybrid systemsshow some promise.Another distinction within the artificial intelligence communityis weak AI versus strong AI. Weak AI refers to using software tosolve particular problems or reasoning tasks that do notencompass fully human intelligence. Strong AI implies creatingartificial systems that are fully self-aware, the systems that canreason and independently solve problems.
Current research isnowhere near creating strong AI, and a lively debate is ongoingas to whether this is even possible.Another division in the artificial intelligence community is overthe best way to design an intelligent system (Neats versusScruffies). The Neats maintain that the solution should beelegant, obvious, and based on formal logic. The Scruffies holdthat intelligence is too messy and complicated to be solvedunder the limitations the Neats propose.
Interestingly, somegood results have come from hybrid approaches, such asputting ad hoc rules (Scruffy style) into a formal (Neat) system.Not surprisingly, the Neats are often associated with232conventional artificial intelligence, whereas the Scruffies areusually associated with computational intelligence.Conventional AI has achieved success in several areas. Expertsystems, or knowledge-based systems, attempt to capture thedomain expertise of one or more humans and apply thatknowledge. Most commonly, this is done by developing a set ofrules that analyze information about a problem andrecommend a course of action. Expert systems demonstratebehavior that appears to show reasoning.
Expert systems workbest in organizations with high levels of know-how andexpertise that are difficult to transfer among staff. The simplerexpert systems are all based on binary logic, but moresophisticated systems can include methods such as fuzzy logic.At the heart of an expert system is an inference engine, aprogram that attempts to create answers from the knowledge baseof rules provided by the expert. Knowledge engineers convert ahuman expert’s “rules-of-thumb” into inference rules, whichare if-then statements that provide an action or a suggestion if aparticular statement is true.
The inference engine then usesthese inference rules to reason out a solution. Forward chainingstarts with the available information and tries to use theinference rules to generate more data until a solution is reached.Backward-chaining starts with a list of solutions and worksbackward to see if data exists that will allow it to conclude thatany of the solutions are true. Expert systems are used in manyfields, including finance, medicine, and automatedmanufacturing.Another approach from conventional AI that has achievedsome commercial success is case-based reasoning, or CBR,which attempts to solve new problems based on past solutionsof similar problems.
Proponents argue that case-basedreasoning is a critical element in human problem solving. Asformalized in computer reasoning, CBR is composed of foursteps: retrieve, reuse, revise, retain. First, access the availableinformation about the problem (Retrieve). Second, try to extenda previous solution to the current problem (Reuse).
Next, test233the refactored solution and revise it if necessary (Revise).Finally, store the new experience into the knowledge base(Retain).Behavior-based artificial intelligence (BBAI) attempts todecompose intelligence into a set of distinct, semi-autonomousmodules. BBAI is popular in the robotics field and is the basisfor many Robocup robotic soccer teams, as well as the SonyAibo. A BBAI system is composed of numerous simplebehavior modules, which are organized into layers.
Each layerrepresents a particular goal of the system, and the layers areorganized hierarchically. A low layer might have a goal of“avoid falling,” whereas the layer above it might be “moveforward.” The move forward layer might be one component ofa larger “walk to the store” goal. The layers can access sensordata and send commands to the robot’s motors. The lowerlayers tend to function as reflexes, whereas the higher layerscontrol more complex goal-directed behavior.Bayesian networks are another tool in the conventional AIapproach.
They are heavily based upon probability theory. Theproblem domain is represented as a network. This network is adirected acyclic graph where the nodes represent variables, andthe arcs represent conditional dependences between thevariables. Graphs are easy to work with, so Bayesian networkscan be used to produce models that are simple for humans tounderstand, as well as effective algorithms for inference andlearning. Bayesian networks have been successfully applied tonumerous areas, including medicine, decision support systems,and text analysis, including optical character recognition.There is no widespread agreement yet on exactly whatComputational intelligence (CI) is, but it is agreed that itincludes neural networks and fuzzy computing. A neuralnetwork consists of many nodes that cooperate to produce anoutput.
The system is trained by supplying input on thesolution of known problems, which changes the weightingbetween the nodes. After training has tuned the parametersbetween the connections, neural networks can solve difficult234problems in machine vision and other areas. Also known asneurocomputing, or parallel distributed processing, neuralnetworks loosely model structures in the human brain. Neuralnetwork outputs rely on the cooperation of individual nodes.Data processing in neural networks is typically done in parallel,rather than sequentially as is the standard for nearly all moderncomputers. Neural nets can generalize from their training, andsolve new problems, so they are self-adaptive systems.
Neuralnetworks have been criticized as “bad science” because it isdifficult to explain exactly how they work. Nonetheless, neuralnetworks have been successfully applied in areas as diverse ascredit card fraud detection, machine vision, chess, and vehiclecontrol.Fuzzy logic, fuzzy systems, and fuzzy set theory are all ways torefer to reasoning that is based upon approximate values, ratherthan precise quantities.
Modern computers are built uponbinary, or Boolean, logic that is based on ones and zeros. The bitis zero or one, yes or no, with no middle ground. Fuzzy systemsprovide for a broader range of possible values. Consider thequestion, “Are the books in the study?” Well, yes, there arebooks in the study. There are also books in the office, books inthe bedroom, and a pile of books in the doorway to the study.Fuzzy logic provides for an answer of 72%, meaning that 72%of the books are in the study.
Fuzzy sets are based on vaguedefinitions of sets. They are not random. Fuzzy logic is notimprecise; rather, it is a formal mathematical technique forhandling imprecise data. Like neural networks, fuzzy logic issubject to controversy and criticism. But systems based onfuzzy logic have an excellent track record at certain types ofproblems.
Antilock braking systems are based on fuzzy logic,and many appliances incorporate fuzzy logic.Notes:Bayesian network (Bayesian network, Bayes network, beliefnetwork, Bayes(ian) model or probabilistic directed acyclic235graphical model) is a probabilistic graphical model (a type ofstatistical model) that represents a set of random variables andtheir conditional dependencies via a directed acyclic graph(DAG).Robocup is an international robotics competition that aims todevelop autonomous robots with the intention of developingresearch and education in the field of artificial intelligence. Thebest universities in the world compete in several leagues.AIBO (Artificial Intelligence robot) is a robotic project fromSony.
In Japanese, AIBO means pal or partner. AIBO was one ofseveral types of robotic pets that were designed andmanufactured by Sony. Sony Aibo is basically a robotic dogthat that is able to walk and “see” its environment using the onboard cameras. It is even able to recognize spoken commandsin languages including Spanish and English.Assignments1. Translate the sentences from the texts into Russian inwriting paying attention to the underlined words andphrases:1. Machine learning makes heavy use of symbolicformalism and logic, as well as statistics.2. Whereas conventional AI is considered to be a top-downapproach, with the structure of solutions imposed fromabove, computational intelligence is more bottom-up,where solutions emerge from an unstructured initialstate.3.
Some proponents claim that this is appropriate, becausethe human mind uses multiple techniques to developand verify results, and hybrid systems show somepromise.2364. Weak AI refers to using software to solve particularproblems or reasoning tasks that do not encompass fullyhuman intelligence.5. Current research is nowhere near creating strong AI,and a lively debate is ongoing as to whether this is evenpossible.6. The Neats maintain that the solution should be elegant,obvious, and based on formal logic.
The Scruffies holdthat intelligence is too messy and complicated to besolved under the limitations the Neats propose.7. Expert systems, or knowledge-based systems, attempt tocapture the domain expertise of one or more humansand apply that knowledge.8.
![](https://s.studizba.com/z.php?f=/templates/si/i/noavatar.png&w=100&h=100&t=1"){kind=avatar}
