ChatGPT currently handles approximately 2.5 billion queries daily
Alina Vytiuk / Alamy Stock Photo
UN researchers are urging people to be less polite towards artificial intelligence after a report says cutting words from prompts could reduce ChatGPT’s energy consumption by up to 25 percent.
A report by the United Nations University Institute for Water, Environment and Health (UNU-INWEH) found that removing unnecessary words such as “please” and “thank you” from AI prompts could save between 87 and 98 gigawatt hours of electricity annually. This is equivalent to the annual electricity use of up to 760,000 households in sub-Saharan Africa.
To reduce energy consumption and carbon emissions, people should write concise prompts, avoid getting caught in conversational loops and avoid starting relationships with AI, the researchers said.
“I’m not saying we should be disrespectful towards AI, but we shouldn’t fall into the trap of interaction and we shouldn’t get carried away with it,” says UNU-INWEH’s Kaveh Madani.
The large language model behind an AI chatbot processes text in small units called tokens. Madani says concise prompts can save energy because they can reduce both the number of tokens a model has to process and the number of tokens it generates in response. In some cases, shorter prompts can simplify the task and further reduce power requirements.
One of the most comprehensive assessments of the environmental costs of AI to date, a United Nations study warns that the growing adoption of AI technologies is rapidly increasing energy, land and water use.
Today, ChatGPT alone processes approximately 2.5 billion queries every day, and Google processes 16 billion queries, the majority of which have AI summaries integrated.
Since high-tech companies rarely disclose information about their energy use, the researchers used data available in their own data centers.
AI currently accounts for about 20% of the energy used in data centers, and that proportion is expected to double to about 40% in the next few years. By 2030, AI alone could consume approximately 378 TWh per year, and data centers could consume a total of 945 TWh, representing nearly 3% of projected global electricity usage.
The 9.3 trillion liters of water projected to be needed for data centers by 2030 is enough to meet the annual minimum domestic water needs of all 1.3 billion people in sub-Saharan Africa.
“We’re seeing some technologies being introduced on a global scale faster than any other technology in the history of technology. That’s why energy use is increasing so rapidly,” says UNU-INWEH’s Miriam Axel.
The researchers said that while AI companies should be required to publish their energy consumption and governments should introduce energy limits for businesses and individuals, it is also important to educate the public on how to use AI efficiently.
Madani says people should be encouraged to avoid using AI unnecessarily, and when using AI, use fewer words and less powerful models.
It’s also important to note that generating an image uses 60 times more energy than a text query, which is enough to power a 10-watt LED light bulb for about 17 minutes.
A complex video uses up to 8000 times more power than text and can power the same light bulb for about 1.7 days.
“We’re not saying AI is bad,” Madani said. “We’re just saying, let’s use it the right way. It’s like a knife. You can save a patient’s life in the operating room, but you can also use it to kill someone.”
topic:
