Chapter 95 16.4 Agents in the Behavioral Framework

In the 1940s, Europe's famous animal-watching trio—Konrad Lorenz, Carl von Frisch, and Nicol Tinbergen—began to describe the logic behind animal behavior.Lorenz kept a flock of geese at home, von Frisch lived in a house surrounded by beehives, and Tinbergen spent his days with thornbacks and seagulls.Through rigorous and ingenious experiments, three researchers of animal behavior reduced the antics of animals to a respectable discipline - "ethology" (roughly speaking, the science that studies the characteristics of behavior). In 1973, they were jointly awarded the Nobel Prize for this groundbreaking achievement.Later, when cartoonists, engineers, and computer scientists delved into the literature on animal behavior, they were surprised to find that these three behaviorists had already established a very good behavioral framework, which could be used directly. to the computer.

At the core of the behavioral framework is the key concept of "decentralization".As Ting Bogen pointed out in his 1951 book "Insect Studies", animal behavior is a kind of decentralized coordination, which builds together many independent action (drive) centers like building a house.Some behavior modules are composed of reflection phenomena; they can invoke simple functions, such as retracting when heated, or dodging when touched.These reflexes don't know where they are, what's going on in the outside world, or even what the current goals of the body they're attached to are.They are triggered whenever the appropriate stimulus is present.

Male trout instinctively respond to stimuli such as a female trout reaching mating stage, a bug swimming nearby, or a predator approaching from behind.However, when these three stimuli were present at the same time, the predator module always suppressed the mating or feeding instinct and responded first.When conflicts arise between different behavioral modules or between multiple simultaneous stimuli, certain modules are activated to make decisions.For example, you are in the kitchen with dirty hands when the phone rings and at the same time there is a knock on the door.In this case, those conflicting urges — hurry up and answer the phone!No, wipe your hands first!No, have to rush to the door! — and it might throw you off guard, unless another acquired behavioral module is arbitrating at this point, and maybe it's this module that makes you yell, "Wait a minute, please!"

From a more positive point of view, what Ting Bogen called the drive center is equivalent to some kind of "agent".An agent (regardless of its physical form) detects a stimulus and then responds.Its responses, or "outputs" in computer jargon, may appear to be inputs to other modules, driver centers, or agents.An agent's output may activate other modules (pull the firing pin), activate other modules that are active (pull the trigger), or may deactivate adjacent modules (close the firing pin).It's quite difficult to do a tummy rub and a head pat at the same time, because for some unknown reason, one overpowers the other.Often, an output message may activate some centers while inhibiting others.Obviously, this is a network structure, full of a large number of circular causality and strange circles connected from beginning to end.

Outward behavior thus emerges from the tangle of blind reflexes.Due to the distributed nature of the source of behavior, the simplest agent at the bottom layer can also produce unexpectedly complex behavior at the top layer.There is no central module in a cat that decides whether the cat is scratching its ear or licking its paw.Instead, what the cat does is determined by a tangled network of independent "behavioral agents"—reflexes—that cross-activate one another to form an overall behavioral pattern (known as licking or scratching), emerge from this distributed network. It sounds very similar to Brooks' inclusion structure.It is actually an inclusive structure!Animals are functioning robots.The same decentralized, distributed control that governs animals applies to robots and digital beings.

In the eyes of computer scientists, the network diagrams of interconnected behavioral modules in behavioral textbooks are actually the logical flow charts of computers.The conclusion reached: Behavior can be computerized.By arranging subbehaviors, any personality trait can be programmed.In theory, any emotion an animal has, any subtle emotional response, can be generated by a computer.The same bottom-up behavioral management mechanisms used to dominate Robbie the robot can also be used to dominate the creatures on the screen, and this is precisely the mechanism borrowed from the living chaffinches and sticklebacks.Instead of songbirds singing and fish wagging their tails, distributed systems churn data and move thighs on computer screens.In this way, autonomous animated characters on screen can act according to the same general organizational rules as real animals.Despite being synthetic, they behave real (or at least hyper-real).Therefore, it can be said that animated characters are robots without entities.

Much more than actions can be programmed.Personality can also be encapsulated into numbers.Frustration, excitement, and anger can all be added as modules to a creature's operating system.Some software companies sell fear programs better than others.Perhaps, they'll also sell "associative fear"—a fear that's not just manifested in a creature's physical body, but seeps into a cascade of emotional modules that dissipate over time.