For this post, I would like to focus on Mary Gray’s video on the hidden cost of ghost work in algorithms. As she points out, even with machine learning techniques, there are still humans who perform the new work upon which artificial intelligence algorithms rely. This takes on the form of data entry, data labeling, and content review. Gray’s main topic is what she labels the “human-in-the-loop” services that require humans to work on the back end of the algorithms to ensure that they run smoothly.
She goes on to describe how the process works; with “requesters” on the left, who then interact with the “platform” through the internet, when finally any number of human workers with accounts to the platform supply the labor required to complete the initial request. It is through this process, Gary argues, that workers are devalued and isolated due to an over-reliance on code.
Around the 6-minute mark, Gray introduces an “online-to-online” process in which companies access data online and contextualize it with other data sets to maximize profit. Thinking about the process(es) companies go through to obtain such data and contextualize it led me to wonder how future historians will grapple this same data.
For future historians working on economic or labor issues of the early 21st century, what digital sources and data might they discover in their research? How will this information be archived, organized, and preserved over time? How will they be able to link the human experience (whether as worker or consumer) with the different processes the Gray addresses in her lecture?
In Laura Putnam’s piece that we read earlier this semester, she explored the possible shadows that digitized sources cast over certain historical subjects. What can we say about the current digital processes that inherently cast shadows over the human laborers? Like historians now, perhaps future historians will get a glimpse into the conditions through ethnographies, personal testimonies, and second-hand accounts. Or perhaps the data collected, analyzed, and contextualized will be stored in a way that it will be accessible decades and centuries from now.
mario
April 1, 2020 — 9:43 am
I have thought about similar questions before, though fair warning, I’m not trained as a historian. It seems to me that understanding current practices of hiding labor, such as the requester mechanisms in online to offline and online to online services that Gray discusses, provides an implicit sense of comparable instances in history. In other words, we can imagine (and of course find evidence) of how labor was concealed in previous regimes of production, such as Taylorism, Fordism, or Toyota’s just-in-time production. Then and now, technologists and industrialists will have you believe that their histories are of innovation, increased efficiency, automation, etc. while communicating the shifting labor processes from the workers’ perspective often rely, as you write, on ethnographic and testimonial knowledge.
At the same time, content moderation, crowdsourced microtasking, and digital work overall have their own contexts and dynamics. To me, this suggests that future historians will need to theorize what alternative and revisionist approaches might look like based on available (and unavailable data). I would imagine that their work could be improved by paying more attention to these labor systems now.