At present time main elaborations of “intellectual machines” is carrying out in teaching of neural nets on selection of information. The restriction of this approach is connected with the fact that internal presentation of information in neural net is difficult to formalize and analyze.
From the CSNT point of view it is possible to suppose, that algorithms of neural nets are the method of experimental calculation of definite notions’ location co-ordinates in notion space. That is to say, weight ratios of neural nets are certain co-ordinates in certain area of notion space.
That is to say, neural net is map of notion space of given subject area, but not ordered, without given orthonormalyzed hubs of semantic co-ordinates. This leads to impossibility of neural net’s applying without preliminary teaching on concrete selection. Teaching has to be started again for each new class of tasks, that doesn’t allow moving knowledge, fixed in one neural net, to another.
The semantic space map allows solving this problem, because:
1. All notions are located in one space
2. Co-ordinate hubs in this space are orthonormalized
Then it allows modulating a solving of creative tasks, without formalizing of algorithm of solving from the start.
Let’s consider next example. Let be task of reaching some aim with unknown or non-formalized way of solving (creation of substance with given properties, reaching given position of the company in market, etc). At the it has to be an algorithm (number of steps) to reaching this aim.
With help of co-ordinate space notion theory:
- Determine the position of target notion in semantic space map
- Chose notions the closest to the target
- Consider semantic differences – difference between co-ordinates of target notion and existing notions
- Reflect semantic differences into notion space and search the closest actions
- Apply consecutively chosen actions to notions
- If the result is close enough to the target, we’ll fix the set of notions and actions as optimal.
Then it is possible to create of systems, which are able to solve non-formalized tasks.