Ants, and other social insects, use simple rules to create complex emergent actions. This observation, that large numbers of participants can make small choices which add up to collectively complex systems, is the basis for all sorts of amazing and exciting work in computation, robotics and complexity theory.
The idea goes something like this: if we can figure out how ants use a few dozen pheremones to communicate with each other in ways which allow them to, say, go invade that picnic twenty meters away, then we ought to be able to figure out how to program large numbers of relatively dumb parts to work together using similar rules to say, guide traffic on freeways, or keep windmill farms working at optimum efficiency. This is known as an "agent-based, bottom-up" approach. It's a valuable insight, which has already lead to some startling breakthroughs.
However, like all good working metaphors, the metaphor of ant-like emergence has a dark side. In this case, it's the circular mill. As described by Eric Bonabeau, a circular mill occurs when ants travel around an obstacle, say a tree trunk, leaving behind them two parallel pheremone trails which say "follow me" - one going there and one coming back. More ants follow that trail, leaving their own "follow me" trails as usual, but then something goes wrong: an ant going there takes a wrong turn and crosses over to the coming back trail, then wanders again to the going there trail, creating a circle of "follow me" pheremones around the tree.
Other ants pick up this circular trail, and, by following it, reinforce it. Sometimes a colony's entire worker ant population will get stuck in one of these patterns, going around the tree faster and faster, building up a stronger and stronger circular trail, until they create a circular mill, from which no ant will vary, and run around in a circle until they all die.
Draw your own conclusions about meaning and import, but the next time someone mentions programming a system to work like ants or bees, remember the circular mill.