A network of integration, sharing, affectivity and union between the proponents and an (already selected) group of Turkers, in an attempt to conceive other modes of training machines, different from those made available by Amazon.
About 100,000 people perform Human Intelligence Tasks (HITs) on the Amazon Mechanical Turk (MTurk) platform. Low pay and alienation mark the essential labor of AI training.
These individuals are labelled by Amazon as Turkers, a reference to The Turk, a machine supposedly capable of playing chess (in fact, there is someone inside, moving the pieces). In the 18th century, Von Kempelen traveled Europe with this spectacle, an ode to intelligent machines – fascinating, but if examined carefully, replete with human exploration.
Since 2017, we have experimented with MTurk by getting closer to the workers, and stimulating their visibilities/subjectivities. Moreschi has also done fieldwork as a Turker. Through these processes, we became familiar with the system and some of its participants.
As part of this residency, we will create and continuously update a website with multimidia dialogues (chat windows), each featuring a Turker as a “special guest”; as well as a general forum. Possible dialogue themes: obtuse meanings of HIT labels; traumatic experiences in data cleaning; problematic behaviors of requesters; and the diverse (and unheard) opinions of these crowdworkers. Visitors of the website will be able to interact.
These conversations have a wider goal of subverting/rethinking the MTurk enterprise. Together, we will contribute to possible less-commercial ways of machine learning, based on ideas of solidarity and social justice. The grant will be split between us and the Turkers (about 5 of them).
SUBTITLE OF THE HEADER IMAGE:
Image created by the proponents from the studies of Stafford Beer & Herman Schwember (Concepts and Tools of Computer-Assisted Policy Analysis, Basel: Birkhäuser, 1977) and Eden Medina (Cybernetic Revolutionaries, Technology and Politics in Allende’s Chile, Cambridge: MIT Press, 2011).