In our attempt to formulate a concept or a theory about the-map making 
problem, we have therefore also to rely on artificial intelligence. By way 
of analogy, in order to formulate a concept of a metric camera, we need to 
understand how electromagnetic radiation interacts with lenses and 
photographic emulsions. Hence, we rely on physics and chemistry, and only 
after having derived a theory, a camera can be designed and developed. 
Experiences from experiments and practical use will lead to verification 
and to refinenebt of the theory. In the same vein, analytical geometry and 
statistics let us formulate a theory. about aerotriangulation and the 
implementation on computers offers an opportunity to test the theory and 
to refine it. To implement the theory of aerotriangulation we need tools 
and again we are well advised to closely follow sound principles and 
proven methods, such as those of numerical mathematics to solve large 
equation systems, or methods from computer science to search and sort 
large data sets effíciently. Tools should not be confused with methods 
that helped us to formulate the theory, however. There is a danger that for 
solving photogrammetric and cartographic problems, at which people are 
better, we will try different tools before a sound concept or theory is 
established. This (re)search is driven by tríal and error. 
Computer vision may serve as another example to emphasize further what 
has been outlined above. Vision is our most impressive sense and ever 
since computers became available, computer scientists as well as 
photogrammetrists, attempted to perform stereopsis on computers. Since 
most people are remarkably adept at seeing stereoscopically and do it 
without conscious effort, the complexity of the problem was 
underestimated. Only ten years ago serious attempts were made to 
understand the human visual system. These efforts culminated in a 
comprehensive, yet incomplete theory that treated vision from a 
computational viewpoint and that lets us at least understand the modules 
of the early visual process. Stereopsis is one of those modules (see, e.g., 
Marr(79), Mayh(81)). Only after such a theory was formulated was it 
possible to develop computer programs that perform not only in restricted 
cases. We ought to critically evaluate, on the tool level, what methods we 
use in photogrammetry for stereo matching. As indicated in Sche(86), it 
may well be that inadequate tools are being used. 
Conclusions 
When examining photogrammetric and cartographic tasks that we attempt 
to solve by computers, two extremes can be observed. On the one hand, 
there are tasks for which a detailed theory exists such that a computer 
475