New protocol quantifies hidden risks of automation to preserve human roles
Standard ROI misses four systemic risks—this five-gate protocol decides when to keep humans in the loop.
A new paper on arXiv from Jose Manuel de la Chica Rodriguez, Jairo Rodriguez Arias, and Spyridon Chouliaras presents a formal protocol for deciding when NOT to automate roles in AI-optimized organizations. The Protocol for Human Preservation in AI-Optimized Organizations (PHP-AIO) addresses a critical gap: standard automation ROI ignores four categories of systemic risk that degrade long-term organizational performance. These risks include tacit knowledge erosion (the loss of unspoken expertise), resilience reduction (inability to handle edge cases), regulatory exposure (compliance failures), and socio-institutional capital degradation (damage to trust and culture).
PHP-AIO is a sequential five-gate decision protocol with a final composite check. Each gate evaluates one risk category at the role level, producing auditable outcomes: automate, augment, hybrid, or preserve. The paper also introduces a closed-form automation-debt measure (ρ(P)) that quantifies how role-level automation decisions accumulate across multi-step processes. This debt warning can only be neutralized by a regulator-mandated human-in-the-loop anchor. When applied to stylized internal role profiles, PHP-AIO produced distinct decisions for candidates that standard cost-benefit analysis would have uniformly automated. Threshold sensitivity analysis confirmed the gate decisions are robust to upward perturbations of at least 14% in three out of four representative cases. The framework is particularly relevant for financial services and other regulated industries where automation decisions have high systemic stakes.
- PHP-AIO uses five gates to evaluate four systemic risks: tacit knowledge erosion, resilience reduction, regulatory exposure, and socio-institutional capital degradation.
- The protocol introduces an automation-debt measure (ρ(P)) that accumulates across multi-step processes, neutralizable only by a mandated human-in-the-loop anchor.
- Applied to internal role profiles, PHP-AIO produced non-automate outcomes (augment, hybrid, preserve) where standard ROI would have uniformly automated, with ≥14% perturbation robustness.
Why It Matters
A rigorous, auditable framework to prevent blind automation from eroding organizational resilience, knowledge, and regulatory compliance.