‘Sometimes it’s more expensive than having humans’: Honeywell and Ecolab get real about AI’s limits in the physical world

The latest AI models pose challenges when integrated into buildings, factories, and kitchens compared to chatbots, according to executives from Honeywell and Ecolab. Honeywell's CTO, Suresh Venkatarayalu, highlighted the difficulty in achieving high accuracy, with customers demanding near-perfect precision at 99.9999%, whereas frontier models only reach 85%.

Ecolab's AI Officer, AJ Wijesinghe, emphasized another constraint: cost. Implementing top-tier models for high-volume tasks can become prohibitively expensive, even surpassing the expense of human labor. However, Ecolab managed to mitigate costs by 70% to 80% through model optimization. Both executives cautioned against expecting a fully autonomous future, advocating for a semi-autonomous approach where operators build trust in the systems without completely removing human oversight.

Physical equipment's difficulty in receiving updates compared to software was also noted. Venkatarayalu explained that over-the-air upgrades are under consideration for commercial buildings, but their implementation must be cautiously managed to avoid prolonged shutdowns. The companies are leveraging partnerships and fine-tuning open-source models to address these concerns, ensuring data sovereignty and safety.

Implementing AI in various systems, such as HVAC and security, is seen as a pathway to significant cost savings and operational improvements.