Study finds <1% of AI education papers report LLM environmental costs
Most AIED researchers ignore the carbon footprint of their LLMs—and a new tool aims to fix that.
A new study published on arXiv (arXiv:2606.11215) reveals a stark gap in how the Artificial Intelligence in Education (AIED) community handles the environmental cost of large language models. Researchers led by Eimler et al. conducted a systematic literature review of every paper from the AIED 2025 conference proceedings. Their finding: although the vast majority of projects now leverage LLMs, barely any authors disclose the computational resources used, and nearly zero discuss the carbon footprint as an ethical concern. This lack of standardized reporting means the environmental impact of AI in education remains largely invisible—a problem that grows more pressing as adoption accelerates.
To close this gap, the team introduces an open-source methodology that systematically measures both computational expense and environmental impact. Their software works for local hardware and cloud-based setups, accurately tracking carbon emissions. They also provide a simple formula to estimate the computational cost of frontier LLMs even when the exact number of parameters is unknown—a common hurdle for proprietary models. By offering these tools freely, the authors hope to motivate colleagues to adopt transparent reporting, turning hidden costs into a standard part of AIED research practice.
- Literature review of AIED 2025 proceedings found almost zero reporting of environmental costs despite widespread LLM usage.
- Proposed open-source method measures carbon footprint for both local and cloud-based hardware.
- Includes an easy-to-use formula to estimate computational expense of frontier LLMs when exact parameters are unknown.
Why It Matters
As LLM use in education explodes, this framework could force transparency on the hidden environmental costs of AI.