Field workers, like farmers and radiologists, play a crucial role in dataset collection for AI models in low-resource settings. However, we know little about how field workers' expertise is leveraged in dataset and model development. Based on 68 interviews with AI developers building for low-resource contexts, we find that developers reduced field workers to data collectors. Attributing poor data quality to worker practices, developers conceived of workers as corrupt, lazy, non-compliant, and as datasets themselves, pursuing surveillance and gamification to discipline workers to collect better quality data. Even though models sought to emulate the expertise of field workers, AI developers treated workers as non-essential and deskilled their expertise in service of building machine intelligence. We make the case for why field workers should be recognised as domain experts and re-imagine domain expertise as an essential partnership for AI development.
https://dl.acm.org/doi/abs/10.1145/3491102.3517578
The ACM CHI Conference on Human Factors in Computing Systems (https://chi2022.acm.org/)