Fifth in a series on the edges of science
Observability is an important issue in control systems because it is essential to the closed-loop feedback on which control relies. We control a system—whether mechanical, electrical, or even social—by applying external influences to it that change its operation, while observing the results. In control theory, the influences are called “controls” and the operation is described in “states.” In many complex systems, there are internal states that are not observable, such as the details of heat flow through an aircraft jet engine. We control the airplane without the ability to directly observe those states; instead, we observe the temperatures at turbine inlet and turbine outlet, using those to approximate our knowledge of the details.
Yet there are times when those unobservable states make a critical difference. Given just the right set of unusual circumstances, the detailed heat flow in that engine can result in a flameout. The engine stops working, and the airplane goes into emergency conditions.
In quantum physics, the problem becomes even more severe. Quantum “observables” are often not directly observable at all and must be inferred from measurements. The entire realm is described in mathematics rather than physics, with mathematical transformations required to convert the measurements to the observables. This might be straightforward in classical mechanics, but the Heisenberg Uncertainty Principle guarantees that the measurement itself changes the system. Usually, measurement of quantum values is an irreversible action.
Yet we don’t even have to bore down to the quantum level to see this same effect. Measurement of electronic devices, software effects, social systems and more all suffer from the same issue: measurement changes the system. In fact, the effect sometimes becomes irreversible simply from the intent to perform the measurement, as in Schrödinger’s cat.
When science performs any experiment, it must rely on the accuracy of the measurements. What if the measurement is correct at the instant of capture, but the system changes immediately after? What if a key parameter is not observable? How reliable is the scientific conclusion?
As an example of such unobservable conditions, let’s take the study of dowsing. What, dowsing? Crazy, you say. You can find all sorts of vituperative web articles denigrating the practice. Dowsing has been debunked by scientific experiments repeatedly over the last two hundred years. In a rather famous test in Kassel, Germany in 1991, thirty dowsers were asked to identify which underground pipe held flowing water. The experiment was double-blinded, set up very carefully to eliminate possible outside sources of information. The dowsers were confident of their ability, many expecting 100% success. They were even astonished at the naiveté of the organizers who made winning the cash prize so easy. The experiment asked them only to show 83% success.
None of the dowsers could do so. The best of the crew achieved 20 out of 30 runs, only 67% success. The statistical mean of success across all dowsers was only 53%, not significantly different than chance. So science clearly said the practice of dowsing was ineffective.
And yet. All over the world, people believe in dowsing. Practitioners frequently earn money with their art. Even this year, with near-drought conditions in California, dowsers are being called into service to successfully locate water. Dowsers claim their practice to rely on spiritual forces of the earth, “ley lines,” and other poorly understood, unobservable, possibly non-existent phenomena.
The problem in applying science to such claims is that we have no reliable way to observe these phenomena. Some of the dowsers in the Kassel test claimed, before the tests, there were interfering anomalies. It has even been suggested that the very nature of the experiment, being so carefully controlled, may have damaged the spiritual preconditions necessary for dowsing—that the controls themselves caused the failure.
If this sounds ridiculous, imagine what would happen with a study on flooding if we were unable to measure the amount of rainfall, or a software speed test if we were unable to measure the input rate. In fact, we often face exactly this problem in trying to debug Internet problems, where measurement is as ephemeral as the quantum physics problems. In a further extension, statistical studies of the economy have failed spectacularly over and over through unobservability; in this case, not only can we not measure the variables, we often don’t even know what variables are present!
This is one of the largest areas in which science is often wrong. In a complex world, many situations are not amenable to study in a scientific way simply due to our inability to observe the preconditions, the states of the experiment, and the results.
Doc Honour
October 2023
Doc Honour Books 