Wise Shepherd

Our little happy meadow.

Let's imagine that we have a sheep. It's a lovely and useful creature that happily grazes on any pasture and, in return, provides us with wool. Its only need is to eat.

Well, it may be happy, but unfortunately, it's not very savvy.

However, upon closer inspection of such an ordinary sheep, it turns out that it is guided only by the need for food. It doesn't care if what it's eating is good or if there's plenty of it. It doesn't care whether it can eat from this place and whether eating from this patch now will leave it with nothing to eat later. Its brain simply isn't adapted for that because it doesn't need to be.

The third end of the stick.

For such considerations, a wise shepherd is needed. This is an honorable person responsible for monitoring many parameters: weather, meadow's richness, the sheep's level of satiety, and even its mood.

Such a shepherd, thanks to their more developed perception apparatus, observes all parameters that simply escape the sheep. Thus, they can satisfy the sheep's needs and their own interests much better. Because a well-fed and content sheep has a much better wool growth rate in relation to the amount of food eaten.

Let's get to the specifics.

But where does this little story about our cheerful sheep and the honorable shepherd lead?

It turns out that the automation of various processes is sometimes done inefficiently due to the use of limited algorithms and the study of basic parameters.

The Woolen Boiler and Its Bimetallic Brain.

As an example, let's take the automation of a boiler. Its controller is based solely on monitoring the temperature from one point. Based on this, it turns the "heating" process on or off. The heating process itself involves maintaining a constant temperature by burning oil and heating water. This water is then transported to a radiator, which heats the air and raises its temperature to the desired level.

The only parameter is the set temperature. There is an option to set the water temperature on the boiler itself, but I assume (I'm not an expert) that it should be adjusted to the specific installation and radiators.

The controller simply says "heat" or "don't heat." Inside the boiler, there is something that only maintains the set water temperature by cyclically turning the "burning" process on and off.

Why Would a Wise Shepherd Help Heat Our Home?

Doesn't this remind you a bit of our sheep? The boiler doesn't consider any parameters other than temperature. But it could consider quite a few.

It could take into account fuel consumption in relation to the need to reach a certain temperature at different water temperature settings. It might even determine which temperature minimizes fuel consumption because it turns out that the one provided by the radiator manufacturer isn't optimal.

The boiler could also know the dynamic characteristics of the building and learn that if it allows the temperature to drop below a certain value, it would be more cost-effective to let it cool down by another degree Celsius rather than maintaining the current temperature (thereby reducing hysteresis).

It could take into account a significant drop in outside temperature as a warning of strong building cooling and proactively heat it because it's aware of how costly it is to heat a cold building.

You could multiply and list many such examples.

But all these functionalities would be fulfilled by our honorable and wise shepherd, meaning a controller that can better optimize the heating process (in the example provided) or any process that is currently managed inefficiently.

The algorithm of its operation would be based on many input data, both those that can affect a given process (e.g., outside temperature) and those that are the result of this process (fuel consumption). This algorithm would have to be adaptive, adapting to how the process changes, simply by modeling it within itself. Initially "learning" from scratch, starting from a black box and ending with a more accurate description of the process.

Can We Do Without Surgery?

The very method of implementing such a controller could be based on two paths. The first, more difficult one, is to remove the original brain of the sheep and transplant the shepherd's brain in its place, which would then take over control completely. It seems to me that this is the more difficult option due to the need to intervene inside the optimized organism.

The second option is the concept of a "wise shepherd" controller, i.e., a higher-level controller whose outputs would be the inputs of the current controller. In this case, there is no need to modify the current "configuration," but it requires the ability to control using available parameters. You simply need to know the sheep like the back of your hand and train it in a way that suits you. Animal trainers will surely say it's doable, but certainly not an easy task.

Dilemmas.

Cost-Related.

In any case, it is necessary to consider the cost-effectiveness of changing the current state. Maybe it's not necessary? Perhaps changing the controller to any other or designing the shepherd algorithm is too expensive?

Implementation.

Perhaps there is no easy, effective way to implement the shepherd, whether through transplantation or oversight.

Controllability.

When choosing the overseer option, will it be easy to control the sheep? Resistance of animals (and people) knows no bounds. Perhaps the use of the overseer function will lead to complete confusion of the functions that the controller was supposed to perform, and there will be no optimization, or even a reduction in process efficiency.

Opportunities.

Simplicity.

If the overseer mode is feasible, it may turn out that an advanced shepherd algorithm can easily optimize the process.

Scalability.

The ideal situation would be to create a universal algorithm where you provide input data - their nature, output data, and the algorithm has a free hand in creating the model during the "test period." Then, it independently optimizes the process step by step and presents the benefits. After creating such a universal algorithm, you could sell it in many places.