Environmental risk
AI models are often trained using large amounts of computation. This process is very energy intensive, potentially leading to significant greenhouse emissions depending on the energy sources [132]. Experts believe drastically increasing carbon emissions could accelerate climate change, which may constitute a catastrophic risk [133].
ENTITY
3 - Other
INTENT
2 - Unintentional
TIMING
1 - Pre-deployment
Risk ID
mit1393
Domain lineage
6. Socioeconomic and Environmental
6.6 > Environmental harm
Mitigation strategy
1. Prioritize Location-Intelligent Compute Scheduling: Strategically relocate or time-shift energy-intensive AI training and non-critical inference workloads to data centers operating with verifiable access to low-carbon or carbon-free energy (CFE) sources (e.g., hydroelectric, geothermal, solar, wind) to directly minimize Scope 2 operational greenhouse gas emissions. 2. Employ Model Distillation and Optimization Techniques: Mandate the deployment of smaller, specialized, or distilled models for specific tasks over large, general-purpose models to significantly reduce computational and energy demand. Implement model compression methods, such as weight pruning and quantization, to decrease the requisite memory and power consumption per inference task. 3. Advance Energy-Efficient Hardware and Edge Processing: Accelerate the development and institutional adoption of specialized AI hardware accelerators (e.g., TPUs, ASICs, FPGAs) that are architecturally optimized for AI computations, demonstrating superior energy efficiency. Simultaneously, shift viable workloads to on-device (edge) processing to reduce reliance on energy-intensive cloud data center infrastructure and high-power data transmission.