courses/complexity/information-collection.html
date = 2004.02.19
status = short draft
Data needs to be collected and organized into information in order to understand complex behaviors. The same five principles for learning about complex systems are also guides to the information needed. Each of the five principles requires either a different type of information or different forms of some of the information. Reviewing the five principles and their information requirement:
1. metaphor - The initial metaphor provides descriptive information about the type of system, what features are important and the general type of behavior. In addition, these metaphors also provide a heuristic for comparing the system's similarities and differences.
2. experiences - The rich and thick immersion in complex systems contains types of information that can't be compressed and extracted in subsequent analysis.
3. simulations - Simulations require context, boundary conditions, parameter value ranges and starting conditions. In addition, running simulations requires judgment on which output values will be tracked.
4. data collection - Collection of data from real environments requires information about techniques and methodologies (which are linked to the types of data obtained). The data collection process should also grow out of the experience and in this sense is a judgment about what to leave as much as it is what features to keep.
5. analysis - Although the analysis component may seem to be about using just the collected data, it is really the culmination of all of the four principles. The analysis step is really about selecting a model to describe the data at the level of abstraction and simplification that you judged would be useful.
There is always some aspect of learning about a system that has to be efficient. That is just the reality of bounded rationality. However the search for a good strategy to understand complex systems shouldn't jump to hastily to looking for efficient strategies; bad choices and mistakes may be valuable. Forcing a bad metaphor, dwelling too long in the experiential phase, or picking inappropriate data analysis software could be good learning experiences for future searches.
The five principle components help learn about complex systems. But, if you are going to understand the system, understand in the sense of Perkins (1998) which requires involvement and action, then you must also determine what types of actions you can take. The types and ranges of actions you can take represents your decision choices. Often the types of actions that can realistically be take are so limited that they help constrain the amount that you need to know about the system. It is important to run through this loop as part of process of learning about and understanding the complex system. Before launching into a project to collect as much information about a system as possible, a crucial part of your strategy to should be to use possible constraints to eventual actions to limit and define the scope of your inquiry.
There are theoretical constraints that limit the eventual list of choices and thus reduce the amount of information and study that will useful. Some decisions are just not acceptable to the public. Graedel and Allenby 2003 suggest that the first step in a life cycle analysis process is to screen the product or service with an "involates" list. This treats some activities or side effects as obviously incorrect and thus eliminates some choices from consideration and study. Another consideration is that if the problem is going to be solved within a particular scale (size, time, energy density) that the energy applied to the solution must be low enough not to fundamentally change the problem. These issues have been explored for technological and social problems by Adams (1988).
Several practical constraints on the amount of information that can actually be collected and processed will eliminate some studies. A good metaphor or set of metaphor should serve as a heuristic device that will focus on key similarities and differences. The ecological rationality, i.e. the structure of the information related to problem, should be considered (Gigerenzer and Selten 2001, Gigerenzer et al. 1999). It may be that the complex system has a particular structure because of the processes that formed it, and that this structure can be used in making efficient decisions. For example, if you are studying a system that looks like a self-organizing network in which preferential attachment to nodes is important, the distribution of the connections will probably have a fractal distribution and thus you can guess at the structure of the system by only studying it at one scale. The metaphors and simulations should be examined for the structure of the information that results from the processes being studied. This meta-information can be very helpful in describing the information that is needed to address problems.
There are also constraints that help by eliminating potential solutions that are too simple. In many cases that I am aware of, a solution that is too simple in its description is balanced by the use of high levels of power or relies on dissipation, and thus is covered by those considerations. According to Ashby's Law (ref***) any management strategy should have a similar level of complexity as the problem. Using this law doesn't help find a simple solution, but it eliminates overly simple approaches that will probably not work.
Just as simulation models of complex systems can help understand the system, simulations of decision strategies can be used to explore the potential outcomes (Carpenter ****). These simulations can be based on values or if those criteria are met, more detailed decision theories. For most interesting environmental problems, the results of the simulation will probably place you right in the middle of a "wicked problem" in which the challenge will probably be to restructure the social interactions. If it is the case that your real challenge will be to work with the social dynamics and that the science/engineering has to meet conflicting value frameworks, it is at least useful to know that up front before launching into an extended study of a refinement of the science or technology. Fortunately, the social systems in which these problems arise are also complex systems and the same understanding approach that might be applied to science and technology can be used to help frame the question. For example, the adoption of particular anti-pollution technologies for solving a soil pollution problem may also be percolation problems in themselves. In this case, critical levels of acceptance (of the technology) through a the society (a intermediately connected porous media) can have positive feedback effects. There is no reason however to expect a coherence between the scientific/technological problem and its social context. The skills and types of information you might need to address the social dimension are probably very different than you might have expected at the beginning of the problem (see Smart Mobs - Rheingold 2002).
The five principle components for learning about complex systems help understand the range and types of information that are required to make a decision. Theoretical and practical constraints to the types of solutions that are allowed or feasible help narrow the strategy for information collection. The eventual types of decisions may themselves be framed in a complex social or economic system.