Ditch the niceties in AI prompts to save energy use, say researchers

UN researchers are urging people to be less polite to artificial intelligences after a report found that cutting words from prompts could reduce ChatGPT’s energy consumption by up to 25 per cent.

Removing “please”, “thank you” and other unnecessary words from AI prompts could save 87 to 98 gigawatt-hours of electricity per year, the report from the UN University Institute for Water, Environment and Health (UNU-INWEH) found. That is the equivalent of the annual residential electricity use of up to 760,000 people in sub-Saharan Africa.

To reduce their energy consumption and carbon footprint, people should write concise prompts, avoid getting sucked into conversation loops and refrain from starting relationships with AI, the researchers said.

“We are not saying be rude to your AI. But don’t fall into the interaction trap and don’t go falling in love with it either,” says Kaveh Madani at UNU-INWEH.

The large language models behind AI chatbots process text in small units known as tokens. Madani says concise prompts can save energy because they can reduce both the number of tokens the model has to process and the number it generates in response. In some cases, shorter prompts may also simplify the task, further reducing the power required.

The UN study – one of the most comprehensive assessments of the environmental costs of AI to date – warns of rapidly increasing energy, land and water use due to the growing adoption of the technology.

ChatGPT alone now processes around 2.5 billion queries every day and Google 16 billion, the majority of which have integrated AI summaries.

Tech companies disclose little information on their energy use, so the researchers used the available data for their data centres.

AI currently accounts for about 20 per cent of the energy used by data centres, but that share is projected to double to around 40 per cent in the next few years. By 2030, AI alone could consume around 378 terawatt-hours a year and data centres could use 945 TWh in total — almost 3 per cent of projected global electricity use.

The 9.3 trillion litres of water projected to be needed by data centres by 2030 is enough to meet the minimum annual domestic water needs of all 1.3 billion people in sub-Saharan Africa.

“You’re looking at something on a global scale that is being adopted faster than any other technology in the history of technology, so the energy use is increasing very rapidly,” says Miriam Aczel at UNU-INWEH.

The researchers said AI companies should be required to publish their energy consumption, while governments should introduce energy caps on companies and individuals, but it is also crucial to educate the public on how to use AI efficiently.

People should be encouraged to avoid using AI unnecessarily and, when they do use it, to cut words and use less powerful models, says Madani.

They should also be aware that generating an image uses 60 times more energy than a text query, enough to power a 10-watt LED bulb for about 17 minutes.

A complex video uses up to 8000 times more than text and could power the same bulb for about 1.7 days.

“We are not saying AI is bad,” says Madani. “We are just saying let’s use it in a proper way. It’s like a knife: you can save a patient’s life in the operating theatre, but you can also kill someone with it.”