Counter-Intuitive Characteristics of Rational Decision-Making Using Biased Inputs in Information Networks

We consider an information network comprised of nodes that are: rational-information-consumers (RICs) and/or biased-information-providers (BIPs). Making the reasonable abstraction that any external event is reported as an answer to a logical statement, we model each node's information-sharing behavior as a binary channel. For various reasons, malicious or otherwise, BIPs might share incorrect reports of the event regardless of their private beliefs. In doing so, a BIP might favor one of the two outcomes, exhibiting intentional or unintentional bias (e.g. human cognitive biases). Inspired by the limitations of humans and low-memory devices in information networks, we previously investigated a graph-blind rational-information-consumer interested in identifying the ground truth. We concluded that to minimize its error probability, graph-blind RIC follows a counter-intuitive but tractable rule. In this work, we build on this foundational knowledge: "graph-blind RICs prefer the combination of information-providers that are all fully-biased against the a-priori likely input, over all other combinations." Upon studying RICs with partial knowledge of the network graph, we find that they act similar to graph-blind RICs when their BIPs "listen to" sufficiently many information-providers of their own. Furthermore, if a common node is informing/influencing all n BIPs of a partially-aware RIC, that RIC anticipates its discovery of the "influential node" to diminish the average error probability by a factor that increases exponentially with n . However, from the partially-aware RIC's perspective, choosing n fully-, similarly-biased BIPs outweighs the discovery of influential nodes among its BIPs' sources. These insights might inform the design of consumer-centric information networks.