RunTheSim
The two-bridge experiment
The setup looks too plain to be science. One nest. One food source. Two bridges between them. Then you watch which bridge the colony ends up using.
Start with both bridges the same length. There is no better option to find, so you might expect the traffic to split down the middle and stay there. It does not. The colony picks one bridge and commits to it, and across repeated runs each bridge wins in about half the cases, as Scholarpedia's account of the experiment describes it.
Now make one bridge clearly shorter. The ants that happen to pick the short one reach the food first, and they are also the first ones back. Their trail gets laid twice while the long-bridge ants are still walking. From that moment the fork smells short, and the bias grows on its own.
So what is the colony actually measuring?
Time. Not distance.
That difference is bigger than it sounds. A rule that measures distance needs a surveyor, a memory and a way to compare. A rule that counts round trips needs none of that. Every ant answers one question at the fork: which side smells stronger right now. The colony's idea of "short" is a side effect of traffic, and traffic is something a chemical can count by itself. Goss and colleagues fitted that choice to a short probability formula in which a branch pulls harder than its share of traffic would suggest. The formula is the algorithm. There is no manager in it.
| Setup | What the colony does | What actually drives it |
|---|---|---|
| Both bridges the same length | Locks onto one bridge instead of splitting | An early random fluctuation, amplified by every ant that follows |
| One bridge clearly shorter | Locks onto the short bridge | Faster round trips reinforce that trail sooner and more often |
| Artificial ants in software | Find good routes, and are made to forget on purpose | Evaporation lowers every trail each round, which keeps other options alive |
Forgetting is the other half of the trick
A trail that only ever grows is a trap. The first lucky path would harden forever, and the colony could not react when the food runs out or a wall goes up. Evaporation is what keeps the map editable, and the same mechanism sits in the software versions: pheromone evaporation, in the words of that Scholarpedia article, "implements a useful form of forgetting, favoring the exploration of new areas in the search space".
That is why the rule survives outside biology. Simple agents, a shared signal, and a signal that decays. Watch it once and you start seeing the pattern in traffic, in supply chains, in markets. The same shape shows up in a business simulation you run in the browser, where nobody plays the coordinator and the coordination happens anyway.
The trick also routes Swiss delivery trucks
In the 1990s this stopped being a biology story. Marco Dorigo and Luca Maria Gambardella turned trail-laying into an optimization method, much of it at IDSIA, the Swiss AI lab in Lugano. Their artificial ants walk a graph instead of a bridge and communicate through pheromones that evaporate over time.
The lab reports the results in unglamorous places: vehicle routing, truck fleet management, logistics for a large Mediterranean container terminal, and goods distribution for the largest Swiss retailer. Every parcel route is a shortest-path problem with no ant in sight, solved by a rule that a colony has been running for a very long time.
Reading about a feedback loop is one thing, watching it settle is another: the ant colony runs in your browser, free, no account needed.
Run the sim