Collective Animal Behavior from Bayesian Estimation and Probability Matching
https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1002282
In our class we talk about Bayes Rule in Chapter 16, Information Cascades. It correlates to computing the probability of various events, and we use it where a person needs to make decisions on whether a certain choice is the best one, given that the person has received other private information or observed this information. For example, in the book it is talked about in e-mail spam detection. The reference to the subject line being “Check this out” can lead to a person knowing it is more likely that is it spam, even though spam is less than half of their incoming e-mail, with no other information, according to Bayes Rule.
This article uses Bayes Rule in a different way where animals can make collective movement decisions based on probability estimations and the presence of uncertainty. In this experiment, the experimenters made a Bayesian estimation in which behavior would be best performed when only taking into account personal information particularly environment and social surroundings. This information was obtained from watching the behaviors of other animals. The model that was made was to estimate the behavior that was best performed. In this model, the experimenters gave the animals two spatial locations, x and y. They had the animals choose the best location, where “best” meant safest, a place with high destiny of food, or any other assortment of reasons. It can be assumed that each animals uses both social and non-social information in order to make the best decision. Social information was the information that the other animals made to convince this specific subject, non-social being sensory information about the environment which included potential predators or food items. Each individual then estimates the probability that each location, for example, (Y), is the best using non-social (C) and social information (B).
Above is how Bayes Rule was used in this experiment in order to try to find a estimation to what was the best location. With this model and another probabilistic estimation the authors made the conclusion that all the data had a good fit with the social reliability parameter s, where s=2.5. This means that for certain relevant behaviors animals make the right choice almost 2.5 times more than they make the wrong choice. The Bayes Rule can be applied in many ways, like in class where we talked about the probability of picking from a urn with blue or red marbles, or here in this article, where it is used to find the probability that an animals would pick the best location in terms of movement for their species.