Automatic control provides a structure for making good decisions. It chooses a desired setpoint; provides techniques to ensure that data is valid; understands cause-and-effect mechanisms; understands response dynamics; understands disturbances and intermediate variables that indicate change is in progress; tempers response to noise; and makes corrective action within the context of many desires. These include constraint avoidance, stability, speed of response, noise, values of the controlled variables (CVs) and manipulated variables (MVs). Without such principles, controllers overreact causing instability, or underreact causing persistent deviations.
How can we use control principles to set workflow procedures, policy, law and regulations? Although control can be mathematically implemented by a computer, desirable outcomes can also be achieved when the humans-in-charge subjectively apply the principles as they make decisions. With today's Internet access to data, big-data techniques to generate knowledge, and social science models of behavior, automatic control can be applied to stabilize and regulate socioeconomic processes of human importance. To make rational decisions, humans-in-charge need to understand the principles. The automation and control community needs to enable such understanding.
Point 1 – Intuitive control is valid
Although control can be mathematically implemented by a computer, desirably equivalent outcomes can be achieved when the humans-in-charge subjectively apply the principles of automatic control. While driving, humans keep their cars at speed, in the lane, avoid obstacles in the future, manage equipment health and expenses, etc. Humans adjust the hot- and cold-water valves to achieve desired total flow and temperature of a shower. Even children can do that. Humans choose meals to provide a balance of nourishment, cost, and preparation time and effort without exact models of what's in the food.
Humans can move into place to catch a fly ball in the outfield, return a tennis volley, and catch a flying disk when the path was just lifted by an air current. This takes a blend of immediate response, incremental fine tuning, and anticipation of what will be happening to the objects. We continually take control action in shifting work schedule, talking to our kids, talking with customers, negotiating business contracts, addressing issues with the county tax office, relating to our spouse, and substituting players in a basketball game. All these human endeavors are intuitively executed, and with success nearly 100% of the time.
In all of those, I think that understanding the impact of too strong or too weak a push on future consequences and on time to success; monitoring the response to see if incremental additions are needed to prevent offset; monitoring the response to see if its development is on target and doesn't need more push to get there, are each actions of automatic control applied to human processes.
If the process is recalcitrant, it needs a big push. If it's seeking to please, a gentle suggestion is all that's needed. If it seems to be moving, but not to the desired spot or desired rate, a bit of extra reminder is needed. If it's becoming on track, continued harping isn't needed.
Point 2 – We have adequate models
We regularly use models to forecast the deer and oyster populations to adjust harvesting rates by hunters and fishers. Let’s go the next step.
Useful mathematical models of many aspects of human response are now available. Or, empirical dynamic first- or second-order-plus-deadtime (FOPDT or SOPDT) models can be generated from available data. Such models can be used by the human-in-charge to evaluate the control action needed to achieve a desired response magnitude and rate.
Of course, models of human behavior are imperfect. However, experience in automatic control reveals that just a mediocre model is fully adequate. There are many aspects of a dynamic model such as gain, order, time constant(s), delay, nonlinearity and disturbances, so there are many ways a model can be imperfect. Plus, it's nearly impossible to quantify the goodness of a model. However, my view is, if the model were mediocre, if it would only earn a grade of “C” in school, if it was only 70% correct, then that's still fully adequate as the basis for automatic control because a model-based controller might take 30 actions in a settling time.
If the first action isn't perfect, if it might leave a 30% error, then the second action corrects it by an additional 70%. After 30 control actions, this concept indicates any remaining error is undetectable relative to the digital discrimination of any process signal. Although it's not that simple an analysis, experience shows we can get good control from an imperfect model. I believe many of our models of collective human behavior are fully adequate to use for control. We can take political polls with a 1% error. We can take marketing surveys and structure test communities to adequately project the sampling response to the entire nation. We take action when statistical confidence in the economics is only 95%.
Models of human behavior could include rates of relapse to unsafe behavior after safety training, the impact of a professional sports team or of public parks on the growth of a community, the shift in values and perspective after becoming a parent, or the national economy response to the prime lending rate.
I think, with rational application of the principles of automatic control, even if the application is intuitive, we can improve effectiveness of the direction of human behaviors. Principles from automatic control can be applied to the decisions that parents, bosses and legislators make in imposing an environment to shape child, employee and populace behavior.
Yes, this suggests the application of automatic control for a national social agenda to shape human behaviors. This is already done to shape national conscientiousness, allegiances, the “way” with art and film and music, with laws and regulations, with marketing of products, etc. Some have the greater good of society in mind, some have a selfish motive, and some arises from the creativity of youth. In any case, I’d like to see the control for the common good to be grounded in some automatic control principles.
Point 3 – Humans-in-charge often ignore best practices
Here is a hypothetical example: the owners of a university (legislature, regents, officers) are very conscious of enrollment, and want growth. It's part of their bragging rights. Growth demonstrates their success as managers, which enables them to move to the next level. However, the statistical vagaries of enrollment have a one-time impact that's greater than the growth trend. If enrollment drops, even within statistical noise, the humans-in-charge see disaster, and implement all sorts of programs to fix it. The next year’s rise affirms the value of their tampering, and cements all of their distraction as essential policy and programs. The next year’s drop means that they have to implement all sorts of new tampering. Control experts know better, we temper vagaries with statistical process control, filtering, and outlier removal.
Of course, that was just hypothetical. You may have your own hypothetical stories of human management that did not employ good control practices.
Point 4 – Control technologists can help decision makers
To make rational decisions, humans-in-charge need to understand the principles of automatic control. The automatic control community understands the concepts, but the control technologist will probably not be given high-level managerial or law-making decision prerogative. To help decision makers, the automatic control community needs to promote and enable the use of control practice to such applications to educate the humans-in-charge.
The humans-in-charge did not achieve their managerial positions by mathematics or science skill, so don’t show them Laplace transforms or calibration techniques. It's not about the glorious and satisfying math of high-stature intellectuals. Instead, show humans-in-charge how to: find simple (but adequate) dynamic models from existing data, determine CV-MV pairing, relate feedback control concepts of immediate push, and observe the response to anticipate where the process might settle to see if more or less pressure is needed.
How does one decide on the right target value for a process? We in process control choose setpoints all the time. Some are grounded in economic optimization of the process, others are selected for intermediate variables to ensure subsequent processes remain on the desired target. Some are selected to ensure safe operation relative to uncertainty or disturbances. Partly, intuition leads us to the basis for choices, but the grounding is in the process behavior within context. Occasionally, models are used to determine the right CV-MV pairing. Share the basis of choices, so that humans-in-charge can see how to apply the methodology to their systems.
It is not about the control topics in engineering textbooks. Humans manage multiple-input-multiple-output (MIMO), nonlinear, constrained processes with uncertainty. We choose the structure associated with control (including cascade, ratio and feedforward for early detection and prevention) of the veracity of data, of frequency of control, of tempering response to noise, and of deciding the desired state. Be a voice to promote the applicability of automatic control within the human context. Share the methodology that would view human behavior as a responsive process that is influenced by external disturbances and random events as well as the manipulated variables. Share a systems approach to design a control structure, rise above mathematics to embrace properly designed intuitive control.
