Sarjakoski: 
Förstner (FRG): 
  
Sarjakoski: 
Elassal: 
  
Of course, it could be written. But somehow I don't 
see why it should be done. Because procedural 
languages, such as Fortran, Pascal, Ada, and so on, 
are really designed to make conventional programming 
to be efficient. Why should we use LISP there? 
I want to make three comments. The first thing is 
that you don't have to be ashamed of writing the 
expert system in Fortran. There are big expert 
systems in medicine which are written in Fortran, 
because of the efficiency of these systems. With 
respect to LISP, I only want to mention that at 
Stanford Research Institute all programs are written 
in LISP and they really do all number crunching for 
matching on the LISP machines. It may be that for 
research this is very meaningful because everybody 
knows that 95% of a program is management and organ- 
isation and not number crunching from the program- 
mer's point of view. So, it might be meaningful 
that the logic which is behind the problem is easier 
to show in a language like LISP. And I think this 
whole talk, and this is my last comment, is very 
valuable because it shows us how to look at problems 
from a totally different point of view. That we do 
not try to look at specialized, very small problems 
and efficiently solve only those. We could also 
look at the whole problem and try to structure it 
and find the right solutions for the whole system. 
Yes, mostly I agree. But I could make one more 
comment. You can write a LISP interpreter by using 
Fortran. This only shows that we just have to use 
such tools that are efficient for our purposes. An 
expert system shell can be written by using Fortran 
but once we are using the shell, we don't think in 
terms of Fortran anymore. However, we still want to 
access good subroutine packages like IMSL and so on. 
The interfacing problem has to be solved in the 
shell we are using. 
I would like to make a final remark myself. I guess, 
to me personally, it is very satisfying to see that 
in a photogrammetric meeting, we are talking about 
things which are purely in the domain of computer 
sciences. It took us so many years to come around 
to realize that there is a real need for doing more 
work in the area of software engineering, to trans- 
fer some of the good ideas that we have been talking 
about for years and putting them into practice. If 
we go back to the previous Session, where the sub- 
ject of automatic correlation was mentioned, there 
is no doubt in my mind that techniques exist to ef- 
fectively do it today. However, to transfer these 
existing techniques into operational systems, with 
adequate process controls, is going to be a real 
challenge that we will have to meet. 
Thank you very much. 
60