Human-in-the-Loop Oversight: Why Reviewing AI Output Fails at Scale
Human-in-the-loop oversight is a feedback structure with capacity, vigilance and skill limits. Why review degrades and how to design it to work.
Human-in-the-loop oversight is a feedback structure with capacity, vigilance and skill limits. Why review degrades and how to design it to work.
Burnout as a systems outcome: the workload and attrition loops, a simulation showing why overtime backfires after eight weeks, and structural levers.
Why COP outcomes move in increments: consensus as a ceiling, package deals, the ratchet of agreed language, and what to watch at COP31.
A tested tutorial: load a Vensim model with PySD, run it, validate against the analytic Bass solution, and sweep parameters.
A compartment model of misinformation, a toy simulation of supersharers, the empirical studies, and where the epidemic analogy breaks.
How local faults become system-wide collapse: load redistribution, thresholds and coupling, the 2003 and 2025 blackouts, and design rules that contain cascades.
Weak versus strong emergence, a traffic-jam simulation you can reproduce, and what emergence means for modelling and managing complex systems.
Metastable failures keep systems overloaded after the trigger ends. How retry storms work, what the research shows, and how to prevent them.
IEA and Berkeley Lab data show why AI data center power demand outruns the grid, analysed as stocks, flows and delays.
Why errors compound across AI agent steps, what the MAST research and OWASP ASI08 say, and design patterns that stop a cascade.