The discourse on AI has mainly been focused on the “artificial” and “autonomous” aspects of the new technologies, what they are capable of and how to make use of them. What is often left out is the human labor that makes AI possible in the first place. Humans are needed to prepare training data and to verify and correct results of algorithmic processes. The structure of the industry makes it difficult to notice the ‘humans-in-the-loop’ – most of the work is outsourced and distributed to networks of crowd workers, many of them working from home. Yet, the manual labor is far from marginal. Lionbridge, a company offering data training and related services states that they have access to ‘500,000+ vetted contributors across 300 languages’. Appen, a competing company, advertise a ‘global crowd of over one million skilled contractors. They speak 180+ languages in 130+ countries’. Platforms such as Amazon Mechanical Turk, Clickworkers and Microworkers connect clients and workers in similar ways. Taking this workforce into account, AI seems to be highly labor intensive. The industry is in fact full of contradictions: It is a high-tech enterprise – and a cottage industry; it is fueled by a rhetoric of the spectacular, and based on the heavily mundane; AI is about computers taking the jobs of humans, and human impostures doing what computers were supposed to do, ‘when humans, so to speak, steal computers’ jobs’.
This breakfast we will discuss various aspects of AI and manual labor: how it is organized, the tasks that the workforce performs, how to make sense of the humans-in-the-loop.
Link to Zoom Meeting: