“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?xy of the move, as computed by some heuristic indicating a priori desirability of that move, and the trail level? xy 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.
“A single weaver colony might span dozens of trees and include hundreds of nests built throughout their territory in an integrated network. From here they launch attacks, raise young, and care for livestock, other insects that they raise for nectar.”
The team members looked surprised at this last part.
McKinney drew another series of points similar to the Traveling Salesman Problem and started connecting them. “Weavers maintain a flexible network of routes between their population centers. And unlike most ants, they have excellent vision. They also have better memories than regimented species such as army ants. Individual weaver ants can accrue ‘experience’ which informs later actions.”
Expert Five piped in. “So they’re like a neural network.”
McKinney nodded. “Precisely. Weavers process experience via mushroom bodies…” She drew the outline of an ant’s head, inside of which she drew several large blobs. The largest, occupying the bottom center, she shaded in. “These are brain structures found in almost all insects, and they manage context-dependent learning and memory processes. Their size correlates with the degree of a species’ level of social organization. The larger the mushroom body in the brain, the more socially organized an insect society is. As we’d expect, weaver ants have an unusually large mushroom body, which endows weaver workers with above average memory.
“That memory sharpens the iterative component of weaver swarming intelligence. Because swarming intelligence is all about data exchange. What we call”-she wrote a word on the board-“ stigmergy. Stigmergy is where individual parts of a system communicate indirectly by modifying the local environment. In the case of weaver ants, they exchange data mostly through pheromones.” She started drawing lines that represented ant paths. “If they encounter a source of food or an enemy, they return to the nearest nest, all the while laying down a specific mix of chemical pheromones in a trail that communicates both what they’ve encountered-food or threat-and the degree to which they encountered it-lots of food or a big threat. Half a million individual agents moving about simultaneously doing this creates a network of these trails, known as the colony’s pheromone matrix, holding dozens of different encoded messages. This matrix fades over time, which means it represents in effect the colony’s current knowledge. As weavers encounter these trails, they’re recruited to address whatever message the trail communicates-for example, to harvest food or fight intruders. As they move along the trail, they reinforce the chemical message-sort of like upvoting something on Reddit or ‘Liking’ someone’s Facebook status. As that pheromone message gets stronger, it recruits still more workers to the cause, and soon, clusters of ants begin to form at the site of the threat or opportunity.”
Expert Two, the blond man, nodded. “Meaning it goes viral.”
McKinney nodded. “Basically, yes. In this way, weavers manage everything from nest building, food collection, colony defense, and so on. At each iteration of their activity, each ant builds a solution by applying a constructive procedure that uses the common memory of the colony-that is, the pheromone matrix. So, although individual weaver ants have very little processing power, collectively they perform complex management feats.”
McKinney dropped the marker in the tray at the base of the whiteboard. “In fact, if I were going to create an autonomous drone-and I had no ethical constraints-swarming intelligence would be a logical choice. Lots of simple computational agents reacting to each other via stigmergic processes. That’s why weaver ants don’t need a large brain to solve complex puzzles. They can solve problems because they can afford to try every solution at random until they discover one that works. A creature with a single body can’t do that. A mistake could mean biological death. But the death of hundreds of workers to a colony numbering in the hundreds of thousands is irrelevant. In fact, the colony is the real organism, not the individual.”