Trying to make industrial AI just a tiny bit more accurate is starting to cause a massive, scary spike in carbon emissions.
SSRN · March 18, 2026 · 6429425
The Takeaway
The prevailing wisdom in Industry 5.0 is that more accuracy is always better for the environment. This research identifies a 'dual-regime' where initial AI improvements help sustainability, but past a certain point, the energy and carbon costs of marginally better predictions become a net negative for the planet.
From the abstract
The transition to Industry 5.0 demands industrial systems that are not only reliable but also sustainable and resilient. Predictive maintenance (PdM), a cornerstone of such systems, remains dominated by a single-objective paradigm that maximizes technical accuracy while treating critical sustainability impacts - energy, cost, and carbon footprint - as externalities. This paper addresses this gap by proposing a Multi-Objective Optimization (MOO) framework that transforms sustainability into a fir