DESCANT - fAIre MRG

DESCANT: Detecting Stereotypes in Human Computational Tasks

About the Project

Crowdsourcing has enabled the development of novel “hybrid human-machine information systems,” which benefit from including humans in the loop when facing computational tasks that are still better solved by humans than machines (e.g., describing the emotional content of a given text or the visual content of an image). However, hybrid systems are only as good as the data with which they are built and ensuring the quality of workers’ contributions is often non-trivial. DESCANT addresses a particular concern for crowdsourced data quality: the expression of social stereotypes in the data collected. There is no shortage of media coverage on popular hybrid systems (e.g., search engines, machine translators, chat bots) that have been observed exhibiting sexist or racist behaviours. At the same time, the research community is making serious attempts to better understand how social biases end up in these systems and what can and should be done to redress this. Furthermore, recent research has made it clear that systems and algorithms trained on human-produced data exhibit the same implicit biases – such as the expression of racial and gender stereotypes – that humans do. Although the source of the training data for these systems varies, DESCANT focuses specifically on systems that leverage paid micro-tasks crowdsourcing. The project extends upon our previous work, which focused on developing conceptual and computational models to detect stereotypes in search engine results. In DESCANT, we will generalize the approach, carrying ou experiments on micro-tasks involving a number of different media (i.e., judgments of a variety of characteristics of images, sound files, and video). Based on our results, we shall develop a set of guidelines for researchers and entrepreneurs who use micro-task crowdsourcing, which will help them to understand how the design of their human intelligence tasks might result in stereotyped data and will offer alternative solutions depending on their goals.

Acknowledgments

This project has received funding from the Cyprus Research and Innovation Foundation under grant EXCELLENCE/0918/0086 (DESCANT), the European Union’s Horizon 2020 Research and Innovation Programme under Grant Agreement No. 739578 (RISE), and the Government of the Republic of Cyprus through the Deputy Ministry of Research, Innovation and Digital Policy.

Project Lead

Dr. Jahna Otterbacher MRG Leader

Demonstration Tools

View the generated demonstration tool developed in the framework of the DESCANT: “Detecting Stereotypes in Human Computational Tasks” research project.

Generated Publications

View the generated publications by visiting the project’s active Zenodo repository.

Deliverables

Find the project’s generated deliverables in the list below.