Towards Crowdsourcing Science
Edith Law , Harvard

Area: Machine Learning

Abstract:

Human computation is the study of intelligent systems where humans are an integral part of the computational process. Well-known examples of human computation systems include crowdsourcing marketplaces (e.g., Amazon Mechanical Turk) which coordinate workers to perform tasks for monetary rewards, games with a purpose (e.g., the ESP Game) which generate useful data through gameplay, and identity verification systems (e.g., reCAPTCHA) that accomplish remarkable feats (e.g., digitize millions of books) through users performing computation for access to online content. One limitation of these human computation systems is that they are restricted to tasks that can easily be decomposed and solved by any person with basic perceptual capabilities and common-sense knowledge. There are many other tasks, however, that are complex and require expertise to solve, but for which human computation would be greatly beneficial. For example, many tasks that arise in the course of scientific inquiries involve data that is unfamiliar to people without formal training (e.g., images of micro-organisms, EEG signals). How can we enable a crowd of people to perform tasks that are difficult, and possibly hard to decompose? In this talk, I will illustrate the key challenges of human computation in the scientific domain, draw connections to my previous work on games with a purpose and crowdware systems, as well as highlight new directions.