Back to the MIT repository
6. Socioeconomic and Environmental3 - Other

Risks to the environment

Growing compute use in general- purpose AI development and deployment has rapidly increased energy usage associated with general- purpose AI. This trend might continue, potentially leading to strongly increasing CO2 emissions.

Source: MIT AI Risk Repositorymit780

ENTITY

1 - Human

INTENT

2 - Unintentional

TIMING

3 - Other

Risk ID

mit780

Domain lineage

6. Socioeconomic and Environmental

262 mapped risks

6.6 > Environmental harm

Mitigation strategy

1. Implement Algorithmic and Model Optimization: Prioritize the development and deployment of lightweight, specialized, or compressed AI models (e.g., using techniques like quantization, pruning, and knowledge distillation) instead of large, general-purpose models. This strategy directly addresses the computational intensity of AI workloads, offering substantial reductions in energy consumption per inference and training effort. 2. Decarbonize the Energy Supply and Optimize Compute Location: Mandate the transition of AI data centers to carbon-free energy sources through Power Purchase Agreements and site selection in regions with low-carbon power grids. Additionally, implement carbon-aware computing principles to dynamically schedule and shift workloads to coincide with periods of peak renewable energy availability. 3. Advance Hardware and Data Center Efficiency: Invest in energy-efficient hardware, including specialized AI accelerators (e.g., TPUs, NPUs), and aggressively optimize data center infrastructure to improve Power Usage Effectiveness (PUE) through advanced cooling systems (e.g., liquid cooling) and power management techniques. Concurrently, promote the shift toward on-device/edge AI processing to mitigate the energy burden of cloud data transmission and centralized computation.