1.1.2 attributes of intelligent systems
In a non-preferential order will list those attributes or essential dimensions of their systems intelligent.
Adjustment and training.
The ability to adapt to varying conditions is absolutely necessary. It involves not disclosed ability to learn (training), but as the degree of variability of the conditions is greater, training becomes a necessary condition.
It should be noted that training does not appear as a stage or as a level of intelligence, but as a way of increasing intelligence as a result of experience. By training short-term memory is transposed into long-term memory and allows behavior modification system on the basis of what has been memorable.
Training is therefore a mechanism for the storage of knowledge about the outside world and ownership of a way of behavior. Meanwhile, training associated with the adjustment process is a generalizare because the process of training underpins any system multidecizional processing knowledge that is built from abstract models, general.
Becomes an attribute of adjustment, which allows the achievement dezideratului mainstay of intelligent control namely that growth increased functionality without the complexity of the operations calculation.
Autonomy and intelligence.
A system is considered autonomous when it has the capacity to act correctly in environments defined incomplete without external intervention over a period of time.
There are several degrees of autonomy, which we can associate with regulating the functions included in intelligent control:
-
a system of adjusting the parameters are set the minimum degree of autonomy; -adjustment adaptive systems have a high degree of autonomy.
To the extent that a system has a greater degree of autonomy, support that has a higher level of intelligence.
To differentiate degrees of intelligence may be adopted and other criteria such as computing power system, the degree of complexity of algorithms used for the acquisition, processing and evaluation of data obtained from environment;
-
storage capacity in memory of data. In most artificial systems, increased intelligence reflected by the computing power and memory capacity is increasing due to the complexity of hardware structures and therefore may become an obstacle to the application or through a cost too high, either by the impossibility of processing in real time information stored.
For more specific, it may try to define a "catalyst for inteligent ground" like the coefficient of intelligence used in human testing capacity.
The parameters components of this vector intelligence could be: the power of calculation, the number of processors; interprocesor communication, memory size, speed petitioning; mode of representation of knowledge-type maps, symbols, pairs values attributes, variable state ;
Modus operandi with the knowledge such as procedures and answer questions, searching the list, the organization of waiting lists;
functional capacity evaluation and decision;
dynamic range and resolution of related sensor;
how data provided by sensors - signals the transformation of symbols.
The recovery signals with noise, the estimate recurring;
predictive ability of the development parameters;
capacity of evaluating the making costs and the degree of risk; learning capacity by the possibility of recognition of objects and symbols of assimilation or experimental results from an instructor.
Setting a certain degree of intelligence are reflected in other attributes of intelligent system, in particular concerning its capacity to adjust and autonomy.
Since intelligence is an internal property system and not a mode of behaviour, the degree of intelligence can not always be appreciated by the behaviour of the system, but tests found active in science.
A highlight of this degree is to examine how the behavior of the system when changes occur in the symbolic representation of information, which may reveal the extent to which the system "understand" the meaning of symbols they use and determine the difference of an a priori autonomy and an ad hoc, the latter being the only specific situations in which the system can work with any group semantic symbols.
Structuring.
As a structure, an intelligent system must have an appropriate fine architecture, usually modular structured and organized on different levels of abstraction (resolution granularity), or at least have a form of partial order to ensure the hierarchy.Hierarchy refer either to the functions and objectives, either on the resolution and may, but not required, hierarchies and hardware architecture.
We predict that by resolution of a control system understand the size of the area indistinctibilitate to represent an objective, model, plan, law or regulation.
Resolution determines the size of computing power. The resolution control system is higher, the degree of complexity of the increase. Area total interest to be considered, at least in the initial phase, low resolution, then in this area should be chosen subsets of interest for a higher resolution.
This approach avoids a compexity excessive and structure is also a way of operating based on the decomposition in multilevel tasks.
A system with multiple levels of resolution (called the system with representation multi-rezolution) will appeal to generalizare procedure that grouping several subsets of interest and will be replaced with entities towards greater abstraction.
Therefore, several times the levels of resolution to be called in literature and levels of abstraction or generaly levels.
The existence of several levels of abstraction suggests a hierarchical structure and in this respect could even use a measure based on entropy degree of complexity of each level.
Such an approach can highlight least three hierarchical levels, structured themselves as appropriate on more functional sublevels.
The first hierarchical level (lower), level of organization, modeled as a Boltzmann machine used for abstract reasoning, planning and development of decisions tasks.
The second level is the level of coordination usually composed of Petri networks that allow exchange of commands and interfacing level of organization.
The level of education is the implementation containing hardware blocks specializing in data acquisition, processing orders and delivering them to appropriate process.
The definition of intelligent.
Given all these considerations, we will continue to formulate a definition "work" for a system (control) intelligent.
An intelligent control system is a system with high degree of adaptability to changes neanticipate so that training during the operation appears to be essential. The system must have a high degree of autonomy in line with the need for operation in environmental poorly structured and pronounced degree of uncertainty.
To tackle these problems complex system must have a structure, encompassing architectures multifunction or ranked.
We will finish this paragraph stating that the structure of a complex system involving intelligent and complex calculation, which produces serious problems adapting to real-time management processes.
Reducing the volume of calculation for maintaining overall performance is an important requirement for systems with high performance. In this regard the use of models with high degree of abstraction, which have only a minimum of information is essential, as important is the ability to accelerate calculations using dedicated processors, parallel processing and data structures with multiple processors.