Envision a future where algorithms tackle food shortages through enhanced crop production, where planning an entire weekend getaway can be entrusted to a bot, and where poets can create captivating images without the need for paintbrushes. This is the realm of possibilities presented by advancements in generative AI.
The Latest Frontier
Artificial Intelligence has made significant strides since its inception in the 1950s. Recently, the rise of generative AI—an algorithm, exemplified by ChatGPT, trained to autonomously execute tasks and generate content such as audio, code, images, tests, simulations, and videos—has established itself as a valuable asset for both professional and personal applications.
Currently, generative AI is employed for managing customer service chatbots, streamlining tasks like data entry, adjusting the tone of documents, drafting sick notes for work, coding software, and aiding in medical diagnostics related to radiology, tuberculosis, and oncology.
However, this productivity does come with a price tag—one that extends beyond financial implications.
Emerging data indicates that AI significantly influences our energy systems and the fight against climate change through aspects like energy consumption, water usage, and carbon emissions.
“The impact begins right at the start of the supply chain lifecycle,” explains Shaolei Ren, a researcher at the University of California, Riverside, specializing in responsible AI aimed at creating a resilient, sustainable, and fair future. “The environmental ramifications during the manufacturing phase far surpass those during the usage stage.”
The High Costs of Creation
Creating advanced interfaces like ChatGPT-3, Meta’s Llama models, and BERT—collections of large, multimodal language models, which are part of generative AI—requires substantial computational resources to analyze extensive datasets of human-generated text.
For instance, a recent case involving a major company noted by the World Health Organization reported that this analysis phase consumed approximately 3.4 GWh over two months, equating to the annual energy usage of around 300 American households.
Moreover, during the development of GPT-3, an estimated 552 tons of carbon dioxide emissions were produced, which is akin to what 123 gasoline-fueled cars would emit over a year.
This process also requires vast amounts of water to manage the heat generated during these calculations.
User-Generated Expenditures
Every interaction with an AI tool contributes to additional energy consumption. According to a recent study by Ren, GPT-3 has been found to consume enough water equivalent to a 500 mL bottle for every 10 to 50 responses, and the forthcoming ChatGPT-4 is predicted to demand even more.
Water consumption in the context of generative AI differs from traditional water withdrawal, such as during a shower; the latter is typically returned to wastewater systems, whereas consumption here refers to evaporation. This raises potential disparities in water availability across different regions.
“The environmental repercussions of AI tend to be very localized,” Ren states. For example, one company’s data center in an Oregon city consumed over a quarter of the city’s total water use.
Carbon emissions present a similar localized concern, exacerbated by the reliance on fossil fuels for the majority of the energy supply, despite some AI technologies using renewable sources.
A Future We Can Live With?
As the number of data centers grows along with new iterations of AI technologies, future energy demand is set to rise.
This situation requires companies to seriously consider how to develop advancements responsibly while also addressing the myriad complex social issues involved. Additionally, AI’s environmental effects should be factored into discussions about its future.
Consider the following questions:
- Given the energy requirements for AI advancement, could resource scarcity limit our progress, or might we encounter critical challenges in other areas?
- If we prioritize renewable energy resources for AI and its associated infrastructure, are we accepting increased fossil fuel usage elsewhere?
- Should AI applications be restricted to vital functions, thereby limiting personal inquiries to maintain its significance for fields like research, data analysis, and healthcare?
- What level of transparency is necessary from companies regarding their AI-related carbon footprint, water usage, and other environmental impacts—should regulations be established to mandate such disclosures?
This article was initially published in the November 2024 edition of Thewindowsclubs magazine.
