Kill Decision(53)
She turned to the team. “Mind if I erase this?”
Singleton pursed his lips. “Hmm. I’m still working through some things there.”
McKinney put the eraser back on the tray. “Fine. I guess I can—”
Foxy was already wheeling a portable whiteboard in from the edge of the room. He slid it in front of the first one.
“Thanks.” She grabbed a marker and faced out to the assembled experts sitting around the room. One of the stranger speaking engagements she’d had. She looked to the Japanese scientist, the AI specialist. “Five, quite a bit of this will be elementary for you. I apologize in advance.”
“Not at all. I’m interested to hear it.”
“Well . . .” She took a moment to gather her thoughts. “Ant colony optimization—or ACO—models have been around since the early nineties. Mathematical representations of ant behavior are widely used in private enterprise to optimize complex logistics problems, like delivery truck routing, computer network routing, and market analysis. Antlike swarming intelligence is best illustrated by a classic combinatorial optimization challenge known as the Traveling Salesman Problem. . . .”
McKinney drew a series of dots on the board. “Given a list of cities”—she started connecting the dots with a single traveling line—“how do you find the shortest possible route that visits each city only once?” Her on-board solution quickly failed to do so, and she looked up. “Sounds simple, but it’s not; it’s what’s known as a nondeterministic polynomial-time hard problem—meaning it’s very difficult for humans to achieve. Ants solve this problem routinely. They will always find the shortest possible route to a food source, and as experiments using the Towers of Hanoi Problem set show, if that path is obstructed, they can adapt and find the next shortest route. And so on. They do all this without centralized control and without conscious intent.
“In many ways, individual ants are similar to individual neurons in the human brain. The fact that individual ants—let’s call them agents—follow fairly predictable behaviors, means that metaheuristics can simulate their actions with considerable accuracy.”
Snowcap held up her hand. “A metaheuristic is . . . ?”
“It’s an iterative computation method designed to improve a candidate solution. It’s a form of genetic or evolutionary programming. For example, here’s a basic ant algorithm for detecting the edges of pheromone trails. It was developed way back in 1992 by Marco Dorigo. . . .” She started scrawling on the board.
McKinney pointed at the formula. “An ant is a simple computational agent that iteratively constructs a solution for the problem at hand. At each iteration, each individual ant moves from a state x to state y, which represents a more complete intermediate solution. Thus, for each ant”—she pointed at the formula—“k, the probability of moving from state x to state y depends on the combination of two values—namely the attractiveness ηxy of the move, as computed by some heuristic indicating a priori desirability of that move, and the trail level τxy of the move, indicating how beneficial it has been in the past to make that particular move.”
Odin grimaced. “I think we might be getting too deep in the weeds here, Professor. How does your model function?”
McKinney nodded and erased the algorithm. “Right. Sorry. Just wanted to lay a foundation.”
“You can put the gory details up on the wiki.”
“Now, my work in particular . . .” McKinney thought for a moment, and then wrote two Latin names on the board. “Oecophylla longinoda and Oecophylla smaragdina—two closely related arboreal ant species that dominate the tropical forests of Africa, Asia, and Australia—otherwise known as the weaver ant due to their practice of weaving leaf nests with larval silk. They’re of the order”—she wrote on the board again with her clear, Arialesque print—“Hymenoptera, which includes bees and wasps. Weaver ants are what’s known as a eusocial insect, meaning they exhibit the highest level of social organization in nature.
“I developed Myrmidon, my weaver computer model, based on years of direct field observations.” McKinney paced before the board. “Unlike most ant species, weaver ants are fiercely territorial. They attack any intruders into their domain—no matter what the odds. Climb into a weaver tree, and you will be attacked. They swarm enemies with suicidal disregard. That strategy is not evolutionarily problematic because, as with many colony insects, weaver workers don’t reproduce—only the queens pass on their genetic material. Thus, workers always fight to the death—the colony is their legacy.