In 2025, the two biggest social disruptors we face will collide: AI and climate change.
The summer of 2024 broke the record for the hottest day on Earth since data collection began, sparking widespread media coverage and public debate. This is also the year that Microsoft and Google, two of the largest technology companies that invest heavily in AI research and development, missed their climate goals. This too has made headlines and caused outrage, but the impact of AI on the environment is still far from common knowledge.
In fact, the current “bigger is better” paradigm in AI is epitomized by technology companies’ pursuit of bigger, more powerful, large-scale language models that are presented as solutions to all problems, and that results in large costs. That ranges from the generation of vast amounts of energy to power data centers that run tools like ChatGPT and Midjourney, to the millions of gallons of fresh water and large amounts pumped to keep data centers from overheating. and rare earth metals. Required to build the included hardware.
Data centers already use 2% of the world’s electricity. In countries like Ireland, this figure reaches one-fifth of the electricity generated, leading the Irish government to declare a virtual moratorium on new data center construction until 2028. Much of the energy used to power data centers is officially “carbon.” -Neutral”, this relies on mechanisms such as renewable energy credits, which technically offset the emissions generated by the generation of this electricity, but do not change the way it is generated.
Places like Virginia’s “data center array” are mostly powered by non-renewable energy sources such as natural gas, and energy providers are turning to coal-fired power to meet the growing demand for technologies such as AI. Delaying power plant decommissioning. Data centers suck vast amounts of fresh water from scarce aquifers, putting local communities and data center providers at odds from Arizona to Spain. As Taiwan faces its worst drought in more than 100 years, the government is allocating precious water resources to chip manufacturing facilities ahead of rising demand, rather than forcing local farmers to use precious water resources to water crops. selected.
My latest research shows that switching from an old standard AI model trained to perform a single task, such as question answering, to a new generative model can use up to 30 times more energy just to answer the exact same set of questions. may be consumed. Tech companies, which are increasingly adding generative AI models to everything from search engines to text processing software, also don’t disclose the carbon costs of these changes. We don’t yet know how much energy is used during conversations with ChatGPT and when generating messages. Gemini image from Google.
Much of the Big Tech discourse on the environmental impact of AI follows two trajectories. Either it doesn’t really matter (according to Bill Gates) or an energy breakthrough will arrive and magically fix things (according to Sam Altman). What we really need is more transparency about AI’s environmental impact through voluntary initiatives like the AI Energy Star project that I lead. This helps users compare the energy efficiency of AI models and make informed decisions. By 2025, I predict that such voluntary initiatives will begin to be implemented through legislation, from national governments to intergovernmental organizations such as the United Nations. By 2025, more research, public awareness, and increased regulation will finally begin to understand the environmental impact of AI and take the necessary steps to reduce it.
