Spending all day glued to your smartphone is no longer a novelty, but do you have any idea how much energy is used to create images using artificial intelligence? Recent research in the US has shown that generating a single image using AI can consume energy equivalent to a full charge of your cell phone. This study analyzed different language models to understand how each one consumes electricity.
Read now: Do Cell Phones Really Transmit Radiation to the Brain?
AI and Variations in Energy Consumption
The study revealed a large difference in energy consumption between the technologies studied. While AI chatbots showed relatively low consumption, transforming text into an image consumed considerably more energy.
The research, conducted by Carnegie Mellon University in conjunction with the organization Hugging Face, carried out 1,000 requests using different technologies. On average, the consumption for text responses was 0.42 kWh (kilowatt-hours). Generating an image required 1.35 kWh, which can vary depending on the size of the file and the complexity of the details.
The Challenge of Energy Efficiency
One of the least efficient models surprised by presenting a consumption of 11.49 kWh, equivalent to approximately 950 full charges on a smartphone. This shows the importance of balancing the accuracy of the responses with attention to energy consumption on the part of the developers of these technologies.
Understand more: Cell Phones in Classroom Distract Students and Lower Grades, PISA 2022 Report Warns
Environmental Impact of AIs
In addition to energy consumption, the study also assessed the CO2 emissions of artificial intelligence models. Again, technologies focused on text-based responses showed a lower environmental impact compared to those focused on image generation.
The results indicated that AI models focused on specific tasks emitted fewer polluting gases compared to generalist models. These, which analyze the entire database before delivering answers, demonstrated a greater environmental impact.
Carbon Emissions and AI Efficiency
While AIs focused on specific tasks recorded emissions of 0.3 g of CO2 to 10 g, those focused on image generation presented values between 4 g and 30 g. The study highlights that AI training also requires more energy and emits more CO2, due to the number of parameters to be analyzed.
Indeed, the study considers this a first step towards developing more efficient technologies and hopes that further research will guide developers in the search for more sustainable approaches to AI innovation.
See also: Those registered in the Single Registry can request a free chip and cell phone from the Government
