Back to the MIT repository
6. Socioeconomic and Environmental2 - Post-deployment

Impact on the environment

AI, and large generative models in particular, might produce increased carbon emissions and increase water usage for their training and operation.

Source: MIT AI Risk Repositorymit1333

ENTITY

2 - AI

INTENT

3 - Other

TIMING

2 - Post-deployment

Risk ID

mit1333

Domain lineage

6. Socioeconomic and Environmental

262 mapped risks

6.6 > Environmental harm

Mitigation strategy

1. **Decarbonize Power and Strategically Site Infrastructure** Accelerate the transition to clean energy sources for the electricity grids powering data centers. Concurrently, enforce policies that mandate the strategic siting of new AI computing infrastructure in regions with low water stress and high renewable energy capacity to minimize both carbon emissions and strain on local water supplies. 2. **Optimize AI Models for Algorithmic and Hardware Efficiency** Prioritize the implementation of computational optimization techniques—including model pruning, quantization, knowledge distillation, and transfer learning—to reduce the energy and computational demands for both training and inference. Utilize energy-efficient hardware specifically designed for machine learning workloads, such as TPUs and neuromorphic chips. 3. **Adopt Advanced Water-Efficient Cooling and Ensure Transparency** Require data centers to adopt and invest in advanced cooling technologies, such as closed-loop liquid cooling systems, immersion cooling, and the use of non-potable or recycled water. Establish and enforce standardized reporting protocols for corporate energy consumption and water withdrawal/consumption to ensure accountability and enable carbon-aware operational scheduling.