Study Reveals Link Between AI Model Precision and Environmental Impact

German researchers have established a link between the performance of artificial intelligence and its environmental impact. Their study, published in the scientific journal *Frontiers in Communication*, involved an analysis of 14 open language models. Commercial AI systems, such as ChatGPT from OpenAI and Claude from Anthropic, were excluded from the investigation due to the lack of access to their internal workings. Each model was presented with 500 multiple-choice questions, along with the same number of open-ended queries.

The findings revealed that larger and more precise models tend to consume more energy and generate higher levels of carbon dioxide emissions. This was particularly evident in models that break tasks into smaller steps and solve them sequentially, referred to as «reasoning» models.

The model that exhibited the highest emissions was DeepSeek, which also delivered very accurate responses. Some models, like Cogito 70B, deviated slightly from the general trend, achieving slightly better accuracy than DeepSeek while using less energy. However, the overall correlation was clear: the reliability of AI responses is directly linked to its environmental damage.

The study supported a well-known observation in the community that as model sizes increase, their capabilities expand, albeit with a concurrent rise in energy consumption and emissions. This was noted by Jesse Dodge, a researcher at the Allen Institute, who was not involved in the German study but conducted a similar analysis.

One of the study’s authors, graduate student Maximilian Downer, pointed out that it is not necessary to employ the largest and most powerful models for responding to simple questions. More compact AIs can efficiently tackle narrow tasks as well. He emphasized the importance of selecting a model based on the specific task at hand.

This research raises questions about the necessity of deploying AI in every possible area. Even brief AI-generated responses in search engines, such as Google, can contribute to pollution, even if the user did not specifically request them. Each individual instance might seem trivial, but collectively, they could significantly impact the climate.

Previously, OpenAI’s CEO Sam Altman remarked that over time, a substantial portion of all the energy produced on the planet will need to be allocated to the needs of artificial intelligence.