The term 'Human-in-the-Loop' (HIL), once a clarifying concept for AI systems reliant on human input, now obscures more than it reveals. Its monolithic application to a vast and expanding range of human-AI interactions has led to a conceptual fragmentation that impedes scientific progress and responsible systems' development. This research confronts this ambiguity by proposing a new, tripartite framework that offers a more granular and robust typology for understanding human-AI collaboration. This framework is the result of a systematic literature review designed to synthesize a fragmented, interdisciplinary body of research. We first establish a foundational dichotomy based on the locus of control and introduce a tripartite framework consisting of: HIL (AI-led), where AI is autonomous, and humans provide input for improvement (e.g., data labeling for self-driving cars); AI2L (Human-led), where humans are in control, and AI augments their capabilities (e.g., AI-powered diagnostic tools for doctors); And Hybrid Intelligence (HI), a more advanced paradigm focused on symbiotic, co-creative partnerships between humans and AI, often enabled by generative AI (e.g., generative design, collaborative software engineering). It demonstrates that the choice of the human-AI collaboration paradigm is strongly influenced by the domain's primary objectives: efficiency for HIL, accountability for AI2L, and creativity/innovation for HI. It also establishes that ethical considerations must move beyond simple human oversight ("human in the loop") to a more robust and systemic approach called "participatory governance," which involves diverse stakeholders throughout the AI's lifecycle to ensure fairness and accountability. This resolves a central paradox in human-AI studies, re-conceptualizes the landscape of collaboration, and establishes a new foundation for the design and governance of socio-technical systems.
Patil et al. (Fri,) studied this question.