Answers are grounded in this paper and cite their sources.
Artificial intelligence (AI) design traditionally prioritizes full automation and broadest data coverage yet often overlooks the contextual realities of pre-existing biases in datasets and human agency in real-world applications. This article explores how cross-disciplinary collaboration can transform this paradigm by opening AI's ‘black box' of full automation to public accountability, human operation and real-world use. Based on anthropologists’ direct collaboration with computer scientists in a machine learning (ML) AI design project, this project proposes cross-disciplinary methodological innovations for algorithm design and takes up the fairness issue as an exemplary domain to experiment with our new ML model framework. We developed a new ML framework that moves beyond conventional accuracy-coverage trade-offs to incorporate a human-operable three-way balance between accuracy, fairness, and coverage. In this framework, we foreground the real-world contexts and human agency at multiple stages of AI design, from data training, decision-making, interface design to model testing. The results demonstrate both the promise and challenges of bridging academic disciplines and connecting lab-based AI development with real-world needs. While our collaboration has made AI design more accountable, it also highlights enduring tensions in balancing competing priorities like fairness, accuracy, and coverage.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: