Comments on computer system simulation

From InterSciWiki

Comments on Steve Doubleday's computer system simulation by WikiSysopWikiSysop 14:20, 26 February 2008 (PST)

Steve

Looking again at your Computer_system_simulation

You have four levels

1 Produce known patterns ---> 2 problem of Finding pattern

3 Simulate to create pattern finder ---> 4 Reproduce pattern found (simulating an organization)

3 depends on 5: how to assess whether pattern is found

You are wanting to test whether

1=simple ==> 3 can be found
2=too complex  org cannot find

Problem statement takes three examples:

A. Create system for medical data input and evaluation

(is it flexible enough to handle new formats and problems?)
(when can a bureaucracy succeed?) 

B. Create Internet

(done without bureaucracy)

C. Create software

(done by open source)

Seems like B is an engineering problem: make specs and elements generic enough to evolve to handle all kinds of new inputs traffic uses – its a communication net that’s built and it’s the forms of communication that evolve.

Seems like C is also an engineering problem: make specs and elements generic enough to evolve to handle all kinds of new inputs traffic uses

A is wrapped around a real-world problem with all kinds of specifics. It needs B (a communication system) and C (software generic enough to evolve and adapt to the problem).

Seems like A has the following constraint: every application has to have certification that the system won’t make a mistake. I.e., lots of certification, error checking, limitations on who can use and on how it can be used. Etc. Same for airline systems, ticketing systems, other real world apps.

Seems like if you wanted open source / experimentation or add-on engineering (hardware or software) for a real world system where failure is not allowed, you would have to have a firewall between the free-form experiments and allowable real-world uses.

And while the latter are bureaucratically supervised and controlled the former has to also be firewalled to protect from susceptibility to efforts to subvert the system. This is a bit of a problem in an open source community but not so much because of trust, quarantine, and updating tested systems. I don’t know of bureaucracies that done keep the experimental development in the back rooms because of (a) the problem of trust or subversion from outsiders and (b) the proprietary nature of the software.

So in terms of these contexts, what can your simulation show us? How free-form but bureaucratic agents might evolve to incorporate more experimental approaches? What rules, constraints and roles are best suited to the task? And when the real world task is too complex to be solved at all?

Seems like if you know (1) the nature of the patterns to begin with, you can do (5), to evaluate designs to find the patterns that work and don’t work, and add elements from (1) to do (2) and (3)

The other problem is even if you can (3) “discover” pattern, (4) try to reproduce it, how to do (5) when you don’t know (1)? That’s the causality problem that Hal White and Judea Perl have gotten pretty good at.

Can you boil this problem down to its simplest statement of the objective once again after this initial layout? And update how you will learn the most relevant to the objective from the overall design of a solution space?