Environmental alignment for artificial intelligence

There is a missing dimension between AI Alignment safety research and environmentalists’ efforts on AI’s energy and use: How to design AI models that are aligned with environmental goals. As AI models quickly weave into the core digital infrastructure underlying and driving human activity and become more autonomous, the goals and values that we imbue into AI models will scale environmental outcomes exponentially. This paper’s goal is to introduce the concept of environmental alignment in AI system development and to anchor questions on what goals to align to, how to develop such alignment, and how to course correct. By naming the problem explicitly, this work invites further exploration and solution-building by the broader community.