Swarm intelligence : Definition SI


A definition of SI

We propose an overarching definition of SI that covers phenomena observed in both animals and humans: two or more individuals independently, or at least partially independently, acquire information and these different packages of information are combined and processed through social interaction, which provides a solution to a cognitive problem in a way that cannot be implemented by isolated individuals. Essentially, SI is a mechanism that individuals can use to overcome some of their own cognitive limitations. Therefore, owing to individual variation in cognitive abilities in animal (and human) populations, some individuals might solve a cognitive problem only through joining a group and using SI; whereas others might be able to do this alone through insight. Our focus is on the questions of how and when SI is used rather than on the issue of whether SI always provides an exclusive solution to cognitive problems that cannot be solved by single individuals. However, whenever SI enables grouping individuals to solve a cognitive problem, then the way in which this is done (information processing through interaction) is unique to grouping and cannot be implemented by singletons (even if they are capable of solving the problem in other ways). The terminology in the literature can be confusing and different names are applied: such as SI, collective intelligence and collective cognition. We consider these all to be essentially the same phenomena, and refer to them as SI.

What SI is not

The above criteria make clear that (by our definition) not all kinds of grouping and/or collective behaviour should be regarded as evidence of SI. A flock of birds or a crowd of humans in which individuals simply stay together through social attraction are not examples of SI. The fact that animals group and show collective behaviours, including consensus decision making, only tells us that decisions are made in a social context and that animals have evolved forms of decision making that can result in agreement but are not a conclusive indication of SI. After all, grouping is known to be advantageous for many reasons other than increased cognitive abilities (e.g. reduced predation). It is probable, however, that whenever individuals live in groups, there is a potential for SI. It needs to be evaluated on a case-by-case basis whether and how this potential is utilised (and it will be interesting to reexamine some of the recent work on cockroach collective behaviour and human pedestrian behaviour to look for SI evidence). SI is not different in this regard from individual cognition, where criteria such as brain size (relative to body size) and structure might give us some approximate idea of the cognitive potential that animals might have. However, the ultimate test of the abilities of single animals usually comes in the form of the particular types of problem that require a solution.

Possibilities and limitations of SI

Media articles often suggest that SI (or “the wisdom of the crowds”) could be the answer to every decision-making or forecasting problem in modern society. The difference between areas in which SI can and cannot contribute is, however, easily illustrated by a simple example. At a biomimetics exhibition in Berlin, Germany, we presented the general public with two tasks. In the first problem, following Galton, they needed to estimate the number of marbles in a large glass jar. For the second problem, they had to estimate how many times a coin needs to be tossed for the probability that the coin shows heads (and not tails) on all occasions to be roughly as small as that of winning the German lotto. For the first problem, the collective estimate came within 1.5% of the real value: an impressive performance ofSI (despite the high variance of the individual guesses). In the second case, however, the collective guess was poor. For a person with a background in combinatorics, this second task involves only a quick calculation that always arrives at the correct answer of 24 coin tosses. Clearly, expert knowledge would be superior here. An interesting difference is that between imprecision and bias in this context. For the marbles, the individual estimate is imprecise and uncorrelated guesses can result in a close approximation of the real value. In the second case, there is a huge systematic bias preventing useful information extraction. Most real-life problems will have components of both imprecision and bias, and the general rule would be that the greater the imprecision component (relative to the bias component), the greater the potential for SI solutions.