Temnothorax rugatulus ants, individually marked with paint to track their activities. (Alex Wild.)
This excellent article by Ed Yong was published in National Geographic in July 2013. But its statement "individuals do a better job" is somewhat misleading for our purposes. The problem the ants face is that their algorithm for evaluating the Score Voting winner, is vulnerable to noise and to time limitations, which can cause bias if site A is discovered earlier than site B. This would not matter with the ants' algorithm if they had infinite time and an infinite number of ants and thus truly implemented score voting; but with finite time it biases the choice in favor of A, and with finite number of ants we get noise. Meanwhile , an individual ant if it sees all sites usually should make a good decision, but if only sees some, it may make a worse decision. Anyway, I want to remove from your minds the worry that with score voting, societal decisions are worse. This problem is not caused by score voting, it is caused by the ants' imperfect algorithm for approximating score voting.
Virtually every article or documentary about ants takes a moment to fawn over their incredible collective achievements. Together, ant colonies can raise gardens and livestock, build living rafts, run vaccination programmes, overpower huge prey, deter elephants, and invade continents. No individual could do any of this; it takes a colony to pull off such feats.
But ants can also screw up. Like all animal collectives, they face situations when the crowd's wisdom turns into foolishness.
Takao Sasaki and Stephen Pratt from Arizona State University found one such example among house-hunting Temnothorax ants. When they need to find a new nest, workers spread out from their colony to search for good real estate. In earlier work, Sasaki and Pratt have shown that, as a group, the ants are better at picking the best of two closely matched locations, even if most of the workers have only seen one of the options. It's a classic example of swarm intelligence, where a colony collectively computes the best solution to a task.
But Sasaki showed that this only happens if their choice is difficult. If one nest site is clearly better than the other, individual ants actually outperform colonies.
When a worker finds a new potential home, it judges the site's quality for itself. Temnothorax ants love dark nests, in particular; with fewer holes, it's easier to control their temperature or defend them. If the worker decides that it likes the spot, it returns to the colony and leads a single follower to the new location. If the follower agrees, it does the same. Through these "tandem-runs", sites build up support, and better ones do so more quickly than poorer ones. When enough ants have been convinced of the worth of a site, their migration gathers pace. Workers just start picking up their nestmates and carrying them to the new site.
In past experiments, the team have always found that ant colonies make better decisions than individual workers. Even though each worker might only visit one or two possible sites, the colony collectively explores all the options and weighs them against one another. And since many individuals need to "vote" for a particular site, "this prevents any one ant's poor choice from misleading the entire colony," says Sasaki.
This time, the team wanted to see if the colony keeps its superiority for easy tasks as well as difficult ones. They presented Temnothorax ants with two possible nests – one held in constant darkness and another whose brightness could be adjusted. Sometimes, the ants had an easy choice between a dark nest and a blindingly illuminated one. Sometimes, they had to choose between two similar sites, one just marginally dimmer than the other.
As the light difference between the nests got bigger and the task became easier, the ants, whether as individuals or colonies, made more accurate choices. The team expected as much. But to their surprise, the single workers showed the greatest improvements and eventually outperformed their collective peers. In the easiest tasks, they chose the darker nest 90 percent of the time, while the colonies peaked at 80 percent accuracy.
To understand why this happens, consider how the ants choose their nests. If an individual is working by herself, she might visit a few sites in a row and gauge the difference between them. If they're very similar, there's a good chance she'll make the wrong decision. But the colony doesn't work off the recommendations of any individual; it relies on a quorum, just like the up- and down-voting system of social websites like Reddit. Together, the colony can amplify small differences between closely-matched sites and smooth out bad choices from errant individuals.
Still, this system isn't perfect. If many ants happen to find a bad site very quickly, they might reach a quorum before other workers have time to rouse support for a better alternative. "A bad choice can happen even if one site is much better than the other, because the ants at the bad site will have no information at all about the existence of the much better alternative," says Sasaki.
A single ant isn't as vulnerable to this problem. "She will visit both sites, easily see that one is better than the other, and nearly always make the right choice," says Sasaki. Colonies, however, put less effort into comparing their options than lone individuals, which sometimes leads them astray.
Does that sound familiar? Perhaps the same vulnerability can explain why the collective intelligence of humans often flips into the so-called "madness of crowds". Sasaki certainly thinks so. "For example, I just went to an online site to buy a fan," he says. "Instead of comparing options carefully, I blindly bought the most famous one. This ant-like consuming behaviour may lead to a similar pattern – the crowd fails when quality of options is easy to distinguish."
Reference: Sasaki, Granovskiy, Mann, Sumpter & Pratt. 2013. Ant colonies outperform individuals when a sensory discrimination task is difficult but not when it is easy. PNAS http://dx.doi.org/10.1073/pnas.1304917110