AI's environmental impact

 

AI's  environmental impact

Source: Midjourney v7


Every time you ask ChatGPT a follow-up question, you're triggering an exponential multiplication of carbon emissions that most users never see coming.

The core problem lies in how AI processes conversations. Unlike a simple web search that handles one query independently, every time you send a follow-up prompt in a chat, the AI reprocesses the entire conversation history — every past message, every token, every time.

This creates what computer scientists call quadratic scaling, where computational costs increase exponentially with conversation length. The attention mechanism in transformers requires calculating scores between each token and every other token, meaning memory and computational complexity scale quadratically with input sequence length.

The math is stark:

  • A 1,000-token chat requires 1 million attention calculations

  • A 10,000-token chat requires 100 million

  • A 100,000-token chat requires 10 billion

Recent research testing 14 large language models found that complex questions produced up to six times more carbon dioxide emissions than simple ones, while "smarter" models with more reasoning abilities produced up to 50 times more emissions than simpler systems to answer the same question.

Your tenth prompt in a long conversation thread may carry 10 to 100 times the computational cost of your first one. Researchers estimate that a ChatGPT query consumes about five times more electricity than a simple web search, while one query uses approximately as much electricity as could light one light bulb for about 20 minutes.

A University of Cambridge report warns AI growth could raise tech sector energy use 25-fold by 2040, straining power grids and undermining climate targets. Meanwhile, Google and Microsoft report surging greenhouse gas emissions as they double down on AI, despite both companies having ambitious net-zero goals.

Data centers account for about 1.5% of global emissions currently, but that figure could grow to 8% by 2040, exceeding current aviation emissions. Goldman Sachs estimates data centers will be using 8% of total U.S. power by 2030, up from 3% in 2022.

Big Tech companies are deliberately vague about AI's energy requirements, leaving policymakers in the dark about the climate harms, said Bhargav Srinivasa Desikan, the lead author of the Cambridge report.

  • Google, despite claiming carbon neutrality since 2007 through offsets, wrote in its 2023 sustainability report that it was no longer "maintaining operational carbon neutrality" due to AI expansion

  • Microsoft, which pledged to be carbon negative by 2030, has similarly seen emissions surge

The Green Software Foundation, led by Executive Director Asim Hussain, has been working to educate developers about these hidden costs. In a recent LinkedIn post, Hussain credited researcher Nate Jones for helping illuminate how most people — including himself — had underestimated the quadratic scaling problem.

Approaches to reduce AI's environmental impact are emerging. Users can open focused chats for different tasks and front-load prompts with context to avoid long exchanges. Companies are developing linear attention mechanisms to replace quadratic scaling, though these often sacrifice accuracy. The UN Environment Programme recommends standardized procedures for measuring AI's environmental impact.

The accountability gap remains massive. Only 12% of executives using generative AI are measuring its environmental impact, while 42% are re-examining climate goals due to AI adoption.


Source: The Deep View 

Comments

Popular posts from this blog

IBM goes big to keep quantum computers cold

IBM - avances en mantener adecuadamente fríos computadores cuánticos.

Estructura multinivel de los sistemas de agentes de IA