Disclaimer

The following information is not based on scientific evidence. It reflects my way of describing operant conditioning in terms that I learned when studying and practicing electrical and computer engineering.

It is possible to go overboard when trying to apply engineering principles to behavior modification. Any analogy will break down if overused. My hope is that I have not overdone it and that I may provide engineers with some insight as to how training works.

Four particular engineering topics appear to apply to using the clicker for dog (or any animal) training.

• The use of a strobe signal for data acquisition;
• The use of principles of signal averaging;
• The concept of hysteresis;
• Polled vs interrupt mode for handling computer peripherals.

Data Acquisition

Engineers and Scientists study data acquisition from the standpoint of transforming changes in a physical parameter into electronic signals. The electronic signal may be used to drive a chart recorder or perhaps be digitized so that the data can be stored in a computer.

The sensors used fo data acquisition are very specific. A temperature sensor may be capable of detecting a wide range of temperatures but might not be very accurate as to the particular temperature it has detected. Another temperature sensor may have a very narrow range of temperature which it will measure, but the measurement will be very accurate.

When training a dog, for example, we are dealing with a data acquisition system that has a wide variety of sensors and each of these sensors can be wide ranging or narrow and accurate depending on the benefits it provides to the dog. During training we would like to control which signals (information) is being acquired.

When digitizing electronic signals for a computer, a strobe signal is used to activate the conversion for entry into the computer. Strobe lights can perform a similar function by appearing to provide a visual snapshot of external events. If the scene changes too quickly or the strobe lights do not flash briefly enough, the scene becomes blurry and it is not clear what image was important.

A similar difficulty exists when training a dog. The dog's senses are aware of a continually changing situation--how can we give them a signal to provide them with a snapshot of what we liked? The click sound from a clicker device can provide that strobe signal.

Unlike computer circuitry, though, the dog has no reason to interpret the click as a strobe signal. The click only becomes a strobe signal if the click is followed by something the dog really values.

Hysteresis

Hysteresis is not a term that you will hear a dog trainer use. However, since you are an engineer it is important for you to realized that all animals have hysteresis. Once an animal has learned something it is impossible for that animal to return to the state it was in before the learning occurred. (O.K., perhaps that is not the real definition of hysteresis.)

So when a trainer says "the behavior was extinquished through lack of reinforcement," remember about hysteresis. The behavior may have been extinguished to external viewers, but the animal is still capable of exhibiting the behavior or of building on what it had learned.

Interrupt vs Polled-Mode I/O

Computer engineers have traditionally had two ways of allowing a computer to interact with the outside world: polled mode or interrupt mode. In polled mode, a computer program regularly checks the state of the outside world and decides if it is appropriate to send or receive information. With interrupt mode, an external event causes the computer to execute a particular piece of a program that will send or receive information.

Polled and interrupt mode have been used for the computer to collect keystrokes from a human who is pressing keys on a keyboard; or to send characters to a printer.

For a fast computer, polled mode is very inefficient. Most of the time the keyboard does not have a new keystroke to give to the computer. The computer sits in a loop-- check for a keystroke; wait a specified amount of time; then check again for a keystroke. With interrupt mode the computer can be busy doing something else until the keyboard interrupts the computer with a new keystroke.

How about the human computer? How does it work with a new puppy around the house? Most humans work in interrupt mode--they are busy with their own work until they hear a crash and run to see what happened. Interrupt mode allows the humans to keep busy with some activity, other than watching the puppy, until it receives a puppy-generated interrupt. Unfortunately, the interrupt is usually an indication that something terrible has happened. The puppy has learned a new, probably undesirable, behavior that the human will have to deal with for the life of the dog.

Preventing undesirable behavior is preferable to trying to eliminate it after it has been learned. For this reason I would recommend that owners of puppies or of recently acquired dogs use the polled mode for human-dog interaction. In polled mode the human checks the state of the puppy at predetermined intervals. This allows the human to catch the puppy being good and to reinforce this desirable behavior. Frequent polling can help prevent undesirable behavior.

Tethering a puppy to a human does permit the human to use interrupt mode to a certain extent. An interrupt will occur when the tether is taute. This can allow the human to redirect the puppy just before it is about to learn an undesirable behavior.

However, tethering has its disadvantages. Depending solely on tether interrupts will not allow the human to reinforce good behavior. That can only be done with polled-mode training. Another disadvantage is that humans are known for turning off interrupts when they are busy running high-priority programs. With the tether interrupt turned off by the human, the puppy will be free to chew through the tether and then go to generate an interrupt that is of higher priority than the currently executing human program.

Signal Averaging

Our strobe signal is causing the dog to acquire a variety of information from its senses: smell, muscle tension all over the body, auditory, visual, tactile and taste. In order to help the dog sort out which of the data which it acquired is important and which can be ignored we need to understand the principle of signal averaging.

Signal averaging uses the strobe signal to add the acquired data to the data acquired previously. Acquired data generally has noise in it. If the noise is random then it will sometimes add to the information and it will sometimes subtract from the information that is hidden in the data. As more data is acquired the noise will average to zero whereas the informational part of the data will continually increase.

How does this work in training a dog? Perhaps the information we want our dog to acquire is to relax its hind leg muscles when it hears the word "sit." Our click is strobing in not only the hind leg muscle tension but also other environmental and internal data. Some of the data might be how you look, whether the dog is hungry, what sounds it hears or what smells it smells.

A a trainer it is your responsibility to vary the data that is non-essential to the behavior you are training in such a way that it is random. Each successful execution of the behavior will strengthen the information part of the data and reduce the noise part of the data. If only one training location is used, the location will, by default, become information. Signal averaging requires noise to be random, and in the above situation the training location was not varied and so it did not averaage out.

So here I have utilized the concept of the strobe and signal averaging to explain the use of the clicker and the need for varying the environment during training.