As opposed to optimal control.

The tendency to employ panaceas may well be a reaction to the extremely difficult task of making decisions under conditions of pervasive uncertainty. Robust control, a methodology that seeks to develop control policies that maintain performance within a given set of uncertainty bounds, can be used to help avoid this tendency. Robust control can help analysts think in terms of robustness– vulnerability trade-offs to inform a management and learning process rather than finding the optimal solution given a particular characterization of uncertainty.

Statistical decision theory and dynamic optimization techniques assign probabilities to uncertain events. This is often impractical or impossible. Worse, I think, such assigned probablilities often reflect biases that may or may not be consciously recognized. The world is not just uncertain, it is ambiguous. It sometimes seems that stuff is just made up, SWAGs, that can tell us something about those who made the assignments, but not much about the systems they were attempting to control.

There's a saying that "the best fertilizer is the farmer's footprints". As a grazier that means for me that I have to walk the pastures if I expect them to produce well. It's not good enough to drive around, even on a quad or trail bike. I have to be on foot. This takes time and effort so it can seem to be an unrealistic expectation when there is so much work to do, and it seems backward in an era that seeks to reduce labor costs with automation and mechanization.

However, it works. When all of the costs are summed, even assigning a value to my time, there are net benefits. What boots on the ground gives you is a fire hose of timely information about the state of the sward, which can be used to determine subsequent processes and interventions. The human body is a pretty good mobile sensor network when trained. Sight, feel, smell and even taste provide detailed data that may be difficult to articulate but combine to influence decisions.

A robust control system, as I understand it, is an analog of this process. My larger, longer term goals and policies are informed and adjusted by my immediate observations and actions. In model-speak it's hierarchically nested feedback loops. When I wear my business man hat I make plans, but those plans are implemented by the fellow in the sweat stained gimme cap with dung on his boots, and that fellow has control, within broad bounds, of the system.

By now you might be wondering when the insightful part begins. This is just bog standard stuff for anyone who actually does anything. A line manager in a cubicle farm might call this management by walking around, and most every other application of system control to achieve plans likely has analogous methods. But it is insightful for those who don't actually do anything, and that includes most of those who develop governance policies. They are like aliens from another galaxy trying to understand the goofy primates in the wildenesses that they have conquered. For them this is mental acrobatics. They are mechanical rather than biological, so this is all muck and mystery to them.

I wish them luck and hope that they do improve and learn since it is unlikely that they will give up and go home. But I doubt that they fully grasp the conclusions that flow from these novel (for them) insights. In some cases it is apparent that things would be better with no policy at all. Attempts to control the system degrade it. My suspicion is that the more we come to understand such systems the more it will become obvious that this is always the case. We may be able to optimize some parameters, but at the cost of minimizing others, many of which are not directly or immediately apparent. By the time we understand the mistake great harm has been done. They are not entirely unaware of this.

These results simply highlight the importance of continual learning. The robust control framework makes it clear that there are fundamental limitations to what can be achieved for any given level of understanding of the system dynamics. Thus, there is no policy that, once perfected, will work from that point on. Policies will not only be limited in their effectiveness by uncertainty, but will also always generate significant new vulnerabilities. Policies should thus always be viewed as experiments and should be embedded in a continual learning and adaptive management process (24).

Robust control can help structure this process by exposing how policies distribute robustness and vulnerability across a given system. If possible, policies can be chosen to maximize robustness to those parameters about which least is known at the expense of being sensitive to parameters about which most is known.Associated with such a policy would be a learning process to first reduce ambiguity concerning parameters to which the policy is most sensitive.

I'm again reminded of an old Wendell Berry quote I've used numerous times. See Mouse-based Monitoring for an almost 4 year old example on this blog, though that's not the first time I've used it.

"If in order to protect our forest land we designate it a commons or commonwealth separate from private ownership, then who will care for it? The absentee timber companies who see no reason to care about local consequences? The same government agencies and agents who are failing at present to take good care of our public forests? Is it credible that people inadequately skilled and inadequately motivated to care well for the land can be made to care well for it by public insistence that they do so?

The answer is obvious: you cannot get good care in the use of the land by demanding it from public officials. That you have the legal right to demand it does not at all improve the case. If one out of every two of us should become a public official, we would be no nearer to good land stewardship than we are now. The idea that a displaced people might take appropriate care of places is merely absurd: there is no sense in it and no hope. Our present ideas of conservation and of public stewardship are not enough. Duty is not enough. Sentiment is not enough. No mere law, divine or human, could conceivably be enough to protect the land while we are using it.

If we want the land to be cared for, then we must have people living on and from the land who are able and willing to care for it. If-as the idea of commonwealth clearly implies-landowners and land users are accountable to their fellow citizens for their work, their products, and their stewardship, then these landowners and land users must be granted an equitable membership in the economy."

This is from an old essay Private Property and the Common Wealth. It doesn't simply advocate any of the panaceas - government ownership, privatization, community property - it points out that robust control systems must include properly motivated humans, those whose very lives depend on good management. The idea of any external planner, no matter how sophisticated the tools and models, being able to do good management is absurd. To properly model such systems would require replicating them. In effect, the system is the model running some experiment we do not and cannot grasp from inside it. The best we can do is to be intimately engaged and sufficiently empowered to deal with the consequences of our unavoidable acts as we pursue our necessary ends.