My temperature detector and thermometer in the same cup of water, I looked at the temperature readings. 56.7 degrees Fahrenheit on the thermometer, 56.8 on the detector. Super accurate! I skimmed over my data--I had several readings at this temperature. And each time, it recorded 56.8 degrees. But at temperatures that I had read fewer times, my detector was much more imprecise. Why was the thermistor so precise at 56.7 degrees and not at other temperatures?
You must have noticed that rubber bands do not return to their original unstretched length after being repeatedly stretched. The bands exhibit hysteresis. Here is an excellent experiment demonstrating this:
Hysteresis is "the dependence of the state of a system on its history." It occurs in ferromagnetic, ferroelectric, rubber, and other shape-memory alloys.
In magnetic systems, when a magnetic force is applied, and then removed, the magnetic domains in the material do not return to their original configuration, thus giving the material "memory" of the forces applied to it. Over time, as you repeatedly apply and release force onto an object or cycle a field up and down, the material will lag behind and begin to "remember" how it reacted to previous iterations. Eventually, the lag cycle will always look the same.
In the case of my thermistor, after many trials, the atoms within the semiconducting material reconfigured to their old positions whenever the temperature reached 56.7 degrees. This reconfiguration corresponds with the same resistance value as before, and thus the calculated temperature also remains the same.
I first experienced hysteresis in Science Olympiad when I built and flew lightweight, rubber-band powered helicopters and planes. Every day we would cut strands of rubber, wind them up until they had 100+ winds, and load them onto our planes. Based on the rubber band's length, thickness, brand, and usage frequency, I could mentally calculate the number of winds it could take before snapping. As I got closer to my calculated number, I could feel the tension in the band and knew to stop winding. Additionally, after using the same piece of rubber for dozens of trials, the number of winds the rubber could take would stay relatively the same. It was like the band had a memory of its past uses. And that was exactly the case. The rubber had already been stretched to a certain point before, so the elastic grains within it remembered their previous configurations and fell back into them, thus retaining the same number of winds.
Applying these experiences to cognition and AI models, I wonder about memories stored in our brains and how hysteresis may affect the brain's decision-making process. The neurons in our brains process events by firing signals and making connections with neighboring neurons. This leads to a network of connections that creates a memory of the event. Repeated stimuli of the same kind must strengthen these connections, but what happens when there is a conflicting event? Does it make these connections a little weaker, or is this conflicting event stored and interpreted separately? Would the AI models we design also exhibit hysteresis? These are questions I have to dig deeper into now!
Comments