Being polite to chatbots is a waste of energy

Experts have suggested that cutting unnecessary pleasantries from AI prompts could save a considerable amount of energy.

Saying “please” and “thank you” to artificial intelligence chatbots could be wasting vast amounts of energy, according to researchers who say shorter prompts would significantly reduce the technology’s environmental impact.

A study by the United Nations University estimated that removing unnecessary pleasantries from conversations with AI systems could save between 87 and 98 gigawatt hours of electricity every year.

Researchers said that amount of energy would be enough to power a city the size of Leicester, with a population of around 330,000 people, for an entire year.

The findings highlight the hidden environmental cost of everyday interactions with AI tools.

Chatbots process requests by breaking users’ inputs into units known as “tokens”, with each additional word requiring computing power to analyse and generate responses.

As a result, greetings, expressions of gratitude and lengthy explanations all increase the amount of energy consumed.

ChatGPT, currently the world’s most widely used chatbot, handles approximately 2.5 billion prompts every day from its estimated 900 million weekly users.

Researchers said even seemingly harmless habits, such as beginning prompts with “hello”, contribute to the enormous number of tokens processed globally.

Kaveh Madani, who led the study, stressed that it was not encouraging users to abandon basic manners altogether.

He said: “We are not saying be rude to your AI.

“We are just saying let’s use it in a proper way.”

Madani encouraged people to write shorter, clearer prompts instead of lengthy instructions packed with unnecessary detail.

However, the issue remains complicated.

Previous studies have suggested that polite language can improve the quality of responses generated by AI systems.

Last year, OpenAI boss Sam Altman joked about the financial cost of users thanking chatbots.

He said: “Tens of millions of dollars well spent… You never know.”

The UN report also warned that the environmental demands of AI extend beyond electricity.

It estimated that the world’s data centres, which process AI requests, could require 9.3 trillion litres of water annually by 2030, equivalent to the yearly needs of around 1.3 billion people.