Generative AI and Climate Change Are on a Collision Course

By 2025, artificial intelligence and climate change, two of the biggest social disruptors we face, will collide.

Summer 2024 broke the record It was the hottest day on Earth since data collection began, sparking widespread media coverage and public debate. This also happens to be a year for both parties Microsoft and GoogleTwo leading big tech companies have invested heavily in artificial intelligence research and development but failed to meet their climate goals. While it has made headlines and sparked outrage, the impact of artificial intelligence on the environment is still far from known.

In fact, the current “bigger is better” paradigm of artificial intelligence (reflected by technology companies pursuing larger and more powerful large-scale language models as a solution to every problem) brings very huge costs to the environment. These range from generating the vast amounts of energy to power the data centers running tools like ChatGPT and Midjourney, to pumping millions of gallons of fresh water through these data centers to ensure they don’t overheat, and the vast amounts of rare earth metals required to build the volumes they contain. of hardware.

The data center has been used 2% of global electricity. In countries such as Ireland, this figure is as high as one-fifth of electricity generation, prompting the Irish government to announce effective pause The new data center will continue to consume electricity until 2028. While the vast amounts of energy used to power data centers are officially termed “carbon neutral”, this relies on mechanisms such as renewable energy credits that technically do offset the emissions from electricity generation but do not alter its generation Way.

places like this Information Center Alley“In Virginia, it is primarily powered by non-renewable energy sources, such as natural gasenergy providers are delaying the retirement of coal plants to keep up increased requirements Artificial intelligence and other technologies. Data centers draw large amounts of fresh water from scarce aquifers, putting local communities in competition with data center providers in areas including: Arizona arrive Spain. exist TaiwanAmid the country’s worst drought in more than a century, the government has chosen to allocate precious water to wafer manufacturing facilities to stay ahead of growing demand, rather than allowing local farmers to use the water to water crops.

My latest research shows that switching from old standard AI models (trained to perform a single task like question answering) to new generative models can cost the most 30 times More effort just to answer the exact same set of questions. Tech companies are increasingly adding generative AI models to everything from search engines to word processors, but are also not disclosing the carbon cost of these changes — we still don’t know what ChatGPT is using when talking to or generating How much energy is used.

Much of the discussion around big tech’s environmental impact of artificial intelligence has followed two trajectories: Either it’s not a real problem (according to Bill Gates), or an energy breakthrough will appear and magically solve the problem (according to Sam Altman). What we really need is through something like AI Energy Star The project I lead will help users compare the energy efficiency of artificial intelligence models to make informed decisions. I predict that by 2025, such voluntary initiatives will begin to be implemented through legislation, from national governments to intergovernmental organizations such as the United Nations. By 2025, with more research, public awareness, and regulation, we will finally begin to have a handle on The environmental footprint of artificial intelligence and take necessary actions to reduce it.