The energy consumed in training generative and other
computationally-intensive forms of artificial intelligence
(AI) is attracting increasing attention from computer
scientists, energy modelers, policy makers, and the
public. However, the development and use of AI has
other potential energy and environmental impacts.
Building on bibliometric analysis, this article describes
impacts that are the focus of current research—energy
use in training AI—those starting to be characterized in
the research literature—energy use in inference—and
those impacts that exist or are hypothesized to exist but
have drawn little attention from researchers—including
indirect impacts from the use of AI, rebound effects, and
misleading inferences in environmental management
and policy relying on the use of AI.