[Asis-l] Mechanical Turk is not anonymous

Matt Lease ml at ischool.utexas.edu
Fri Mar 8 07:37:22 EST 2013


This may be of interest to those in community using Amazon's Mechanical 
Turk platform for research, as well as those more generally interested 
in how online data can be linked in ways that can be surprising to 
people in practice and compromise their privacy in a manner they didn't 
expect.

Several collaborators and I have just announced discovery of a
vulnerability on Amazon's Mechanical Turk platform, with potential
implications for IRB governance of human subjects research using AMT at
US universities. In particular, this vulnerability can be exploited to
obtain personally identifying information (PII) and other private
information of some workers, who may have shared this information online
in a way they did not recognize could be linked to their WorkerIDs.

This may impact IRB oversight of research conducted at UT with AMT, as
well as what research is classified as human research and subject to IRB
governance.  I am just starting to follow up on this now with our IRB
coordinator here at UT Austin.

The announcement of our finding is below:

Blog post: http://crowdresearch.org/blog/?p=5177
Paper: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2228728

We are now trying to get the word out to be AMT workers, as well as 
researchers whose might be impacted or who may have posted WorkerIDs 
online which could be compromised via this vulnerability. We would 
appreciate your help with this.

We are also specifically advocating *against* online posting of 
WorkerIDs due to the risk of workers not having realized that 
information they have shared could be linked with their worker accounts. 
Regardless of the vulnerability, we have also found explicit requests 
from workers to not post such uniquely identifying information.

Thanks,
Matt

-- 
Matt Lease
Assistant Professor
School of Information
University of Texas at Austin
Voice: (512) 471-9350 · Fax: (512) 471-3971 · Office: UTA 5.442
http://www.ischool.utexas.edu/~ml


More information about the Asis-l mailing list