Human Agency, Causality, and the Human Computer Interface in High-Stakes Artificial Intelligence
New research argues AI's biggest threat isn't trust—it's the loss of human causal control at the interface.
A new research paper by Georges Hattab, submitted to the 2026 CHI Workshop, presents a fundamental critique of current AI ethics discourse. The paper argues that the dominant focus on 'trustworthy' and 'responsible' AI overlooks a more critical Human-Computer Interaction (HCI) crisis: the systematic erosion of human agency. Hattab posits that the primary challenge with high-stakes AI systems—like those in healthcare, finance, or autonomous vehicles—is not establishing trust, but preserving human causal control. The research introduces the powerful metaphor that 'bad AI' will function as 'bad UI,' akin to catastrophic interface failures that misrepresent system state and lead to human error.
Applying media theorist Marshall McLuhan's concepts, the paper frames AI as a technology of 'augmentation' that simultaneously 'amputates' the user's direct perception of causality. This creates a 'double uncertainty' at the interface, combining human uncertainty with the probabilistic model's uncertainty. Hattab critiques current Explainable AI (XAI) methods for their correlational focus and failure to adequately represent this uncertainty. In response, the paper concludes by proposing a rigorous, nested Causal-Agency Framework (CAF). This framework is designed to integrate causal models, formal uncertainty quantification, and human-centered evaluation metrics with the explicit goal of restoring genuine human agency at the point of interaction.
- Critiques 'trustworthy AI' focus, argues core crisis is erosion of human causal agency in high-stakes systems.
- Introduces 'bad AI as bad UI' metaphor, comparing AI failures to catastrophic interface design that misleads users.
- Proposes a Causal-Agency Framework (CAF) integrating causal models and uncertainty quantification to restore control at the interface.
Why It Matters
Shifts AI safety focus from abstract ethics to concrete interface design, crucial for medical, financial, and autonomous systems.