Buying a new laptop can be an exciting but complex decision. One important factor to consider is energy usage. Imagine finding a laptop that offers impressive features but consumes significantly more electricity than your current one. Unfortunately, salespeople cannot provide exact figures because energy usage is often a closely guarded company secret.
But that’s not all. This laptop also comes with a funnel on top, requiring a water refill every time you request a joke or a fun image. The laptop doesn’t disclose how much water it needs, adding to the uncertainty. For those concerned about the environment and climate change, this upgrade may not be worth it.
This scenario reflects the current AI gold rush and its impact on the environment. The true carbon dioxide emissions and water usage associated with every AI prompt remain hidden. While researchers can provide rough estimates, major players like Google, Microsoft, and OpenAI could offer more precise information if they chose to do so.
However, since the launch of ChatGPT in 2022, there has been a general crackdown on information. Companies that offer AI tools refuse to disclose energy usage and carbon footprint information. This lack of transparency is frustrating to climate-conscious individuals like Sasha Luccioni, an expert in AI energy usage research and the climate lead at Hugging Face.
In contrast, companies like Google and Microsoft readily provide information on the carbon emissions of activities such as plane flights. Yet, when it comes to AI-generated content like term papers or AI-created images, they remain silent. The reason behind this compartmentalization may be the fear of public shaming if the environmental cost of AI products were widely known.
It is worth noting that while AI models are energy-intensive, their energy usage may be overshadowed by other power-hungry technologies in data centers, such as cryptocurrency, streaming apps, and online games. However, Luccioni argues that comparing AI to these verticals is unfair because AI is a horizontal tool used across various industries and applications.
Even if AI’s energy usage is not the highest, its exponential growth poses significant challenges. Google’s sustainability report revealed a 48 percent increase in greenhouse gas emissions between 2019 and 2023, primarily driven by AI-related activities. Microsoft’s report also showed a 29.1 percent rise in emissions since 2020. Third-party data centers, designed to support AI workloads, contribute to this surge in emissions.
Reducing emissions from AI is challenging due to the increasing energy demands of AI compute. Data centers’ carbon dioxide emissions are projected to more than double between 2022 and 2030. The finger can be pointed at the greater intensity of AI compute, which exacerbates the problem.
AI’s environmental impact is not limited to greenhouse gas emissions. It also has a significant thirst for water. OpenAI’s training of its GPT-4 model required 11.5 million gallons of water, equivalent to 6 percent of the entire district’s water. Similar issues have arisen in other locations with data centers, highlighting the strain on water resources.
While data centers are increasingly using non-potable water sources and implementing water-saving measures, the exponential growth of AI outpaces the growth of renewable energy. AI queries can be routed to data centers powered by different energy sources, making it challenging to determine the environmental impact accurately.
To address the gap, tech companies rely on carbon credits. However, these credits do not remove emissions from the atmosphere, leading to concerns about outdated carbon accounting rules. Microsoft and Amazon, for example, rely on credits for over 50 percent of their renewable energy. Meta performs better in this regard, with only 18 percent of its green energy coming from carbon credits.
Despite the potential for AI to aid climate research and carbon capture solutions, the widespread usage of AI by everyday individuals contributes to its environmental impact. From Gen Z students using AI to write papers to older generations sharing AI-generated cat pictures on social media, few are actively using AI to combat climate change.
In conclusion, the environmental consequences of AI remain hidden. The lack of transparency from major AI players hinders an accurate assessment of AI’s impact on carbon emissions and water usage. While efforts are being made to improve energy efficiency and water usage in data centers, the rapid growth of AI poses challenges for renewable energy adoption. To truly address the environmental impact of AI, greater transparency and accountability are needed from all stakeholders involved.
