Google’s showcase at its recent I/O conference saw a fresh emphasis on its Search Generative Experience (SGE), an initiative to embed generative artificial intelligence (AI) into Google Search. The goal of SGE is to provide AI-generated answers to over a billion users by 2024, utilizing Google’s large language models Gemini to produce human-like responses to search queries. This marks a departure from the conventional Google Search procedure, as users will now be presented with AI-generated summaries alongside other results.
However, the environmental implication of Google’s new SGE is a topic of concern. Traditional search engines retrieve data from the internet but generative AI systems like SGE create entirely new content for each query. This process is computationally intensive and requires much more energy compared to the standard search methods.
It is estimated that Google conducts between 3 to 10 billion searches every day, which means that even applying AI to a small percentage of these searches could result in drastic environmental impacts. According to Sasha Luccioni, an AI researcher at Hugging Face, employing AI for search generation could require 30 times more energy than conventional searches. In a study conducted by her team, they found training a single large language model could emit greenhouse gases equivalent to driving 49 miles in a gas-powered car in an average day.
Moreover, data centers which house these AI servers for Google are projected to double their power consumption by 2026. Other pertinent issues include the slow and complex transition from fossil fuel to renewable energy sources for data centers and the sporadic nature of renewable energy production, which often fails to meet the constant power demand of data centers.
SGE’s water footprint also warrants scrutiny, with data center water consumption expected to spike due to the introduction of these AI-generated search results. It’s estimated that each interaction with AI systems like SGE could use up as much as 500ml of water. If 10% of Google’s average daily search volume utilizes SGE, we could see a daily water consumption equivalent to a city of half a million residents.
While strides are being made towards more energy-efficient AI hardware and a greater reliance on renewable energy, there is notable uncertainty around the environmental footprint of AI. As AI becomes increasingly integrated into our everyday internet usage, it is crucial for tech companies to address and mitigate these environmental concerns. It remains to be seen whether the global implementation of AI experiences like Google’s SGE can be sustained on a mass scale without exacerbating environmental issues.
