Abstract: In this talk, we discuss the general question of fundamental capacity limits of human-based
crowdsourcing and freelance systems to perform intelligent work, as well as practical schemes that
approach these limits. We specifically focus on two types of crowd-based frameworks: skill-agnostic
microtasking, where simple jobs are performed by crowd workers for a small payment, and skillbased
crowdsourcing, where jobs can only be performed by workers with requisite skills. A main
challenge for skill-agnostic microtasking is the unreliability of random crowd workers, whereas for
skill-based platforms it is the dynamic, time-varying, and random (un)availability of skilled workers.
Information and coding theory are well-suited to understand the limits of reliable computation in a
system constructed from unreliable elements. On the other hand, queuing theory is naturally suited
for studying limits of systems with dynamic and random resource availability. We also briefly discuss
how certain important crowdsourcing scenarios necessitate a joint information- and
queuing-theoretic treatment, leading to a union of these two mathematical fields in a novel way.
