|
| |
artificial intelligence |
|
| |
|
|
| |
Emulation of the intelligence of the human mind in a computing system. Since the earliest days of computing parallels have been drawn between the operation of a computer and that of the mind, and efforts have been made to find parallels at the neurophysiological level. Artificial intelligence has been advanced as a means of relieving humanity of many of its less demanding mental tasks; of making the abilities of highly talented individuals more widely available; and of capturing sophisticated human expertise in the form of a set of rules that can be applied routinely by a computer; expert systems are a form of artificial intelligence.
Many attempts have been made to apply methods of artificial intelligence to problems in human geography. Dobson (1983) wrote of \'automated geography\', implying that many forms of routine geographical analysis, such as search for patterns or detection of anomalies, might be turned over to automated experts. Information on incidences of a disease might be analysed continuously with such \'geographical analysis machines\', as soon as the data became available, allowing outbreaks to be detected much more rapidly. Today, the term data mining is often applied to this concept of rapid automated search through the masses of geographic data that are becoming available through remote sensing and other programmes. Stan Openshaw has been a staunch advocate of the use of artificial intelligence in geographical analysis (Openshaw and Openshaw, 1997).
Two recent developments in artificial intelligence have attracted particular attention in human geography. Neural networks are computer applications in which input data are assigned to a layer of simple processors. The output of these processors then becomes input to a second layer; some applications may include many such layers until finally the last layer provides the outputs. Because of the complex network of connections between processors, it is possible for a set of inputs to produce virtually any output. By modifying the internal connections a sufficiently large number of times, a neural network can be \'trained\' to produce a required set of outputs from a given set of inputs. Neural networks have been applied to the automated classification of remotely sensed images, for example in assigning land-use classes to images based on the radiation received by the sensor. This is an extremely complex task even for a human, and there is very little theory available to guide the development of automated procedures. They have been used to build models of complex behavioural systems, such as shopping by consumers. Thus neural networks are appealing because of their ability to learn from a few observed instances, and because of the inherent flexibility in the nature of the connections that can be built between inputs and outputs. Neural networks are popular in other areas where the lack of theory makes this extremely general approach to modelling worthwhile (Fischer and Gopal, 1994; Miller et al., 1995).
Genetic algorithms also provide extremely rapid search capabilities over complex functional spaces, but use an analogy to biological evolution rather than to brain function, since evolution can be conceptualized as the process by which a population optimizes itself for an environment, driven by survival as the measure of success. Possible solutions are represented by strings in a suitable alphabet; at each iteration, pairs of solutions \'breed\' other solutions by exchanging parts of their strings. In each iteration only the best solutions are retained, and over time the set of solutions steadily improves. Genetic algorithms have been applied to the solution of certain difficult problems in spatial optimization models (Hosage and Goodchild, 1986), where more conventional methods of search for optimum solutions prove unable to handle the complexity of the problem. For example, they can be used to search for suitable locations for several retail stores in a city, over possibly thousands of available sites, or for optimal tours through a set of destinations minimizing travel distance. (MG)
References Dobson, J.E. 1983: Automated geography. The Professional Geographer 35 (2): 135-43. Fischer, M.M. and Gopal, S. 1994: Artificial neural networks: a new approach to modeling inter-regional telecommunication flows. Journal of Regional Science 34 (4): 503-2 7. Hosage, C.M. and Goodchild, M.F. 1986: Discrete space location-allocation solutions from genetic algorithms. Annals of Operations Research 6: 35-46. Miller, D.M., Kaminsky, E.J. and Rana, S. 1995: Neural network classification of remote-sensing data. Computers and Geosciences 21 (3): 377-8 6. Openshaw, S. and Openshaw, C. 1997: Artificial intelligence in geography. New York: Wiley. |
|
| |
|
|
| |
Bookmark this page:
|
|
| |
|
|
| |
<< former term |
|
next term >> |
|
|
|
|
|
| |
|
|
|
