Ghost workers - the hidden force that trains AI systems

© istock.com

© istock.com

Siddharth Suri (Microsoft Research) came to present his latest research on Ghost Work - The labor that powers AI.
His seminar is part of the Computational Social Science and Economics launched jointly between the College of Management of Technology and the School of Computer & Communication Sciences.

This seminar is the 3rd edtion of an initiative aimed at bringing in leading guest speakers to examine questions that are of great importance for EPFL and society more broadly. These seminars are hosted by Prof. Kenneth Younge, Technology & Innovation Strategy Lab, EPFL and Prof. Robert West, Data Science Lab, EPFL

Ghost Work: The Labor that Powers AI

February 28th, 2020

Think about the last time you searched for something on the web. Ever wonder who trained the search engine ranking algorithm? Ghost work is the hidden human labor that powers modern AI systems not only for web search, but also for content moderation, image recognition, spam detection, and a host of other problem domains. This talk will shed light on this hidden workforce by introducing the humans that train modern AI systems and step in when these systems have low confidence. Through the combination of qualitative and quantitative research I’ll show how workers self-organize by building their own collaboration network. We will also show how workers constantly have to hustle to find good work and that those creating the work have the majority of the power in these markets. We'll close by discussing the implications of ghost work on the overall future of work. Most of the research in this talk is from the recently published book, Ghost Work, with Mary L. Gray.


Siddharth Suri is a Principal Researcher at Microsoft Research - AI. Sid is a computational social scientist whose work lies at the intersection of computer science, behavioral economics, crowdsourcing and the gig economy. His early work analyzed the relationship between network topology and human behavior. Since then he became one of the leaders in designing, building, and conducting "virtual lab" experiments using Amazon's Mechanical Turk. He used this methodology to study cooperation, honesty, group problem solving, and display advertising. Most recently, he has been studying the gig workers who power many modern apps, websites, and AI systems which culminated in a book he coauthored titled Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass, published in May 2019.
Sid earned his Ph.D. in Computer and Information Science from the University of Pennsylvania in 2007 under the supervision of Michael Kearns. After that he was a postdoctoral associate working with Jon Kleinberg in the Computer Science department at Cornell University and then he moved to the Human & Social Dynamics group at Yahoo! Research led by Duncan Watts. Sid was a founding member of Microsoft Research - New York City and he recently joined the Adaptive Systems and Interaction group of Microsoft Research - AI.

Censored: Distraction and Diversion Inside China's Great Firewall?

June 13th, 2019

Censored: Distraction and Diversion Inside China's Great Firewall (2018, Princeton University Press) describes how incomplete and porous censorship in China have an impact on information consumption in China, even when censorship is easy to circumvent. Using new methods to measure the influence of censorship and propaganda, I present a theory that explains how censorship impacts citizens' access to information and in turn why authoritarian regimes decide to use different types of censorship in different circumstances to control the spread of information.

About the speaker
Margaret Roberts studies political methodology and the politics of information, with a specific focus on the methods of automated content analysis and politics of censorship in China. She uses large, online datasets to reverse-engineer the content of censorship and online propaganda in China and estimate its impact on citizens' access to information. Professor Roberts received her Ph.D. from Harvard University in 2014 and M.S. in statistics from Stanford University in 2009.
Currently, she is working on a variety of projects that span censorship, propaganda, topic models and other methods of text analysis.

    The Science of Science: What makes science effective?

    May 3rd, 2019


    Our first high-level guest speaker will be Prof. James A. Evans, from the University of Chicago. His research on the Science of Science provides insights into the conditions that drive creativity and the genesis of scientific breakthroughs. Ultimately, these insights help to develop tools and policies with the potential to accelerate discoveries and advance science.

    His research, published in Science and Nature, has analyzed more than 110 million articles, patents, and software packages from 1954 to 2016, and examined questions related to collaboration, team composition, and scientific impact factor. His results shed light on how to balance the demands of multi-expert teams with goals of fostering innovative, breakthrough results.

    Professor Evans’ research is of major interest for all researchers across EPFL seeking to have greater scientific impact and meaningful results.


    Author: Alexandra von Schack