Google's AI-powered searches raise environmental concerns

The introduction of AI Overviews by Google, a feature that delivers instant answers generated by AI for common queries, is raising environmental concerns. Google, managing over 90% of global internet searches, plans to roll out this feature to a billion users by the end of 2024, with other tech giants like Microsoft also adopting similar AI functionalities.

AI Overviews rely on Gemini, Google's family of large language models, to generate responses. These models, while advanced, are energy-intensive. Traditional search engines retrieve pre-existing data, but AI systems must generate new information, consuming approximately 30 times more energy per query. Researchers estimate that the environmental impact includes significant greenhouse gas emissions and high water usage for cooling AI servers. For instance, the BLOOM language model emits greenhouse gases equivalent to driving 49 miles in a gas-powered car per day of use, and generating two images with AI can use as much energy as charging a smartphone, Scientific American has reported. 

The financial and environmental costs of integrating generative AI into search engines are substantial. John Hennessy, chair of Alphabet, Google's parent company, told Reuters last year that interactions with large language models could be ten times costlier than traditional searches. Analysts predict that AI-generated answers for half of Google's queries could cost up to $6 billion annually.

Data centres, the backbone of AI operations, currently consume about 1.5% of global energy and are expected to double their consumption by 2026, potentially matching Japan's current power usage. Generative AI alone is anticipated to consume ten times more energy in 2026 than in 2023. In response, Google is investing heavily in new data centres and aiming to power its operations with 100% carbon-free energy by 2030, though challenges in aligning renewable energy supply with constant power demands remain.

The tech industry is exploring solutions such as better scheduling of computing needs and investing in energy storage to enhance the use of renewable energy. Efforts are also underway to make AI systems more energy-efficient, potentially reducing costs over time. However, the public's ability to connect their digital activities to environmental impacts is limited, often due to the perceived low cost of cloud computing services.

Initiatives like providing Energy Star ratings for AI models and tasks could help users make more informed choices, recognizing the real-world environmental costs associated with their digital actions.