courses/complexity/decision.html
date = 2004.02.18
status = short draft
Decisions in the contexts of deterministic or complex systems are different to two important ways. First, the behavior of complex systems will have to be based more on empirical data than on deductive laws or generalities. Second, the scale of the system is crucial. If the system is small enough such that the action of the decision maker dominates the outcome or large enough such that the system is insensitive to the decision maker's action, then the decision matrix will be essentially the same as for deterministic instances. However if the system is medium in size and connectivity such that the decision maker is an actor in the outcome, then the decision process can be very different than for the deterministic case. Medium size and connectivity systems will be address more below.
Making a decision using the game approach is fairly straightforward. The process is explained in more detail in the "games viewer". For small and large complex systems, the process will be essentially the same. Information about the possible outcomes and their probability distribution can be put into the matrix. There is no need, or no advantage even, to knowing the underlying mechanism. With complex systems, the information that the decision maker has will be incomplete. This makes this a problem of bounded rationality. In bounded rationality the problems is to either make a decision on limited information, to get more information helps you decide, and to know when to stop trying to get more information. In the "games" approach to decisions there are several algorithms for making a decision based on the outcome matrix. These include the maximin approach that was explained in the games view of the precautionary principle, but it can also include other choice options that focus on regrets or maximum possible gain. There is no reason why these games would be different in these situations.
If the criteria for making decisions on the games matrix is cost/benefits or values there should be no difference between deterministic or complex systems. As long as the outcomes could be assigned a probability, there would be no be no difference between assigning an expected value to cost/benefit situations or predicting the probability of broaching a specific value boundary.
There would be a substantial difference between causal and complex systems if the decision were to be justified scientifically. Deterministic systems lend themselves to analysis that can determine a confidence interval around whether the data shows a pattern or whether it's just random. If the decision depends on some proof that the underlying process will lead to outcomes that lie within a certain range, then it will be much easier to achieve this proof with the underlying deterministic processes. In particular if the underlying process is complex, the decision might be based on a pattern that is detected in the data but not meeting the criteria for scientific proof. Patterns, such as those that might indicate global climate change, will be very difficult to prove to a 95% confidence level but yet still be "real" and have the potential to cause massive damage. In complex systems that have strong positive feedback or thresholds leading to dramatic regime changes, the cost of prevention or amelioration may be small in the stages where the pattern is not scientifically significant and enormous after the point of acceleration or regime change.
Another way that complex systems are substantially different than causal systems is when the system is of intermediate size and connectivity such that the action of the decision maker could have an effect on the system. In some situations individual actions may be crucial. For example, the first action taken to solve an environmental problem may be based on an ethical values such as with committed environmentalist world view. These first actions can lead to cascades of events that may be caused by positive feedback cycles or symmetry breaking events. These examples represent the potential of emergent systems to have a fundamentally different outcome at the next level up. One of the most powerful examples of the power of emergent systems is with networks and self-forming groups within networks. There are examples of just a few people making just the right decisions at just the right time leading to a change in the overall social structure. The possibility of such outcomes make decisions strategies in complex systems potentially very important to understand more fully.
Decisions in any system, deterministic or complex, should be done in a manner to solve the problem in the pattern (Berry 1981). Of course to do that, we need to be able to detect the pattern and not change the whole context of the problem in the process. It is not just a matter of bounded rationality, but also the relationship between available knowledge and the application of power. Adams (1988) makes a good case that even if we were to obtain a limitless supply of energy, "the real problem would be precisely that it" (focusing energy on the problem) "would lead to ever greater complexity and indeterminacy, producing nonlinearities beyond the coping ability of human intelligence."