We explore why individual efforts to "opt out" of AI, while potentially offering personal clarity, are ultimately insufficient to address the rapidly escalating environmental toll caused by the technology’s physical infrastructure. The core environmental harm is driven not just by initial model training, but overwhelmingly by inference , which is the continuous, real-time use of AI in billions of applications, causing immense and constant demands for electricity, water, and hardware materials extracted through mining. Instead of blanket avoidance, we explore the need for a strategic, ethical use of AI combined with collective limits and systemic pressure on corporations, which are currently incentivized only toward unchecked expansion and growth. We reflect on how making AI “vastly less harmful” will require both structural changes —like binding regulations on scale and resource use—and principled participation, where the AI technology is deployed only to support justice, care, or resistance, rather than trivial or extractive domains.
