Climate crisis: Generative artificial intelligence, the technology that is revolutionising numerous sectors, could have a dark side: a worrying environmental impact. We don’t realise it, but we are taking the planet to the brink of the precipice. Some people, in an interview with AFP on the sidelines of the “ALL IN” conference on artificial intelligence in Montreal, pointed out that generative AI uses 30 times more energy than a traditional search engine. “I find it particularly disappointing that generative AI is being used to search the Internet,” the researchers said.

Energy-sucking models

The problem lies in the very nature of the language models on which programs such as ChatGPT or Midjourney are based. These require enormous computing capacities to train on billions of data, necessitating powerful servers. In addition, there is the energy used to respond to the requests of each individual user to consider. Unlike a search engine that simply extracts information, AI programs generate new information, making the process much more energy-intensive. The language models on which the programs are based require enormous computing capacities to train on billions of data points, necessitating powerful servers.

The figures are alarming: according to the International Energy Agency, the combined AI and cryptocurrency sectors consumed nearly 460 terawatt-hours of electricity in 2022, or two per cent of total global production. In 2020, ‘CodeCarbon’ was developed, a tool for developers that quantifies the carbon footprint of running a piece of code. Now, as head of climate strategy at the start-up Hugging Face, he is working on creating a certification system for algorithms.

Limiting energy consumption

This system, like the US Environmental Protection Agency’s program that assigns scores based on the energy consumption of electronic devices and household appliances, would provide insight into the energy consumption of an AI product. CodeCarbon is an open-source project that aims to address the environmental impact of the IT sector, particularly that of running code. It measures hardware’s energy consumption during code execution and converts it into equivalent carbon emissions, providing developers with a concrete estimate of their work’s carbon footprint.

This tool promotes awareness and encourages more sustainable software development practices by helping developers identify the areas of code that consume the most energy and suggesting possible optimizations. In this way, CodeCarbon contributes to mitigating the environmental impact of the IT sector, encouraging a more responsible and efficient approach to software development. The idea here is not to oppose AI but to choose the right tools and use them better.

We need transparency

However, we need more, such as more transparency from big technology companies. Microsoft and Google, despite pledging to achieve carbon neutrality by the end of the decade, have seen their greenhouse gas emissions increase dramatically in 2023 due to AI: a 48 per cent increase for Google compared to 2019 and a 29 per cent increase for Microsoft compared to 2020. We need action from governments, which are currently flying blind without knowing the details of the datasets or how the algorithms are trained. Once we have transparency, we can start legislating. Making laws in doubt is of no use to no one.