Design tweaks promote responsible AI use for environmental protection, research shows

Artificial intelligence (AI) systems that require users to pause and reflect on their energy use and environmental impacts can reduce unnecessary AI usage.
AI has a significant environmental footprint, with the electricity needed to train a large language model comparable to powering 120 homes for a year.
AI’s impacts are often not disclosed or effectively communicated to the average user, limiting their ability to make environmentally conscious decisions.
“Design friction” mechanisms, like prompting users to search for existing resources or specifying image details before generating a new one, can increase environmentally responsible AI use.
Persuasive messaging (“cue-based friction”) about the environmental effects of AI increased user trust but did not significantly affect their intentions to use AI responsibly.
Mechanisms for encouraging responsible AI use are critical as AI’s energy consumption is expected to grow significantly in the coming years.
Responsible AI use can include relying on AI only when no comparably effective tools exist, avoiding redundancies, and exiting the tool once the task is done.