All about Artificial intelligence
THE artificial intelligence It’s already part of the daily lives of many people. Whether to do a search, help homework or write an email. But you have already thought about how to use this tool affects the environment?
The data centers in which these AI models are hosted consume clean water to keep computers cold. And can consume more electricity than an entire city.
The US Department of Energy estimates that AI can triple data centers’ electricity consumption, increasing its participation in the country’s energy use from 4.4% to 12% by 2028. To meet this demand, some plants may have to burn more coal and natural gas, increasing greenhouse gas emissions.
The information is from The New York Times.
To decrease the environmental impact, users can follow some tips:
- Keep your questions short and direct.
- Ask the IA to be concise when you don’t need long answers.
- No need to say ‘please’ and ‘thank you’, the models consume more energy with each extra word they process.
- Look for simple models for simple questions, the goal should be to choose the right model for the right task.
- For simple research use traditional search engines. A Google search consumes about 10 times less energy than a chatgpt query, for example.
- Use AI only for tasks that cannot be done in other ways. If you need to make an account, for example, use a calculator.
What do the research say?
Some chatbots are more polluting than others. Research conducted by Professor Gudrun Socher, from the University of Applied Sciences of Munich, compared the performance of 14 AI language models and identified a clear relationship between the size of the model, the accuracy of the answers and energy consumption. The results showed that larger models tend to provide more accurate answers, but consume several times more energy than lower models – which represents a significant environmental cost.
Use it when it makes sense to use it. Don’t use it for everything.
Gudrun Socher
She points out that for simple tasks, it is often not necessary to resort to more advanced models and that smaller and more efficient versions can be enough.
In addition, CO₂ emission associated with AI use also varies according to the region where the models are executed, as some countries use renewable energy sources, while others still depend on fossil fuels.
In 2022, scientist Jesse Dodge, senior researcher at the Allen Institute of Artificial Intelligence (Allen Institute for AI), led a study that compared greenhouse gas emissions generated by AI model training in 16 different regions of the world. The research showed that, depending on the time of year, regions such as the center of the United States may have a carbon intensity up to three times larger than regions with cleaner energy matrix, such as Norway.
According to him, even with limitations, current studies help fill an important gap in understanding the balance between the energy cost and the accuracy of the models.
Everyone knows that as the size of the model increases, it tends to become more capable, but also consumes more electricity and generates more emissions.
Jesse Dodge
