收听本期播客
阅读正文
A recent report has ignited debate by questioning the widely held belief that artificial intelligence (AI) can effectively address the challenges of climate change. The study suggests that many assertions regarding AI’s environmental benefits are, in fact, instances of “greenwashing” – a practice where companies portray themselves as more eco-friendly than their actual operations indicate.
Ketan Joshi, an energy analyst and author of the report, contends that the technology industry frequently employs “diversionary tactics.” He highlights a crucial distinction, noting that companies often conflate traditional AI, such as machine learning, with the newer, significantly more energy-intensive generative AI tools like chatbots and image creators. While traditional AI may indeed offer some potential climate solutions, research indicates that the popular generative AI applications, which are currently driving the sector’s rapid expansion, have not led to any substantial reductions in planet-warming emissions. A particular study found no concrete evidence to suggest that widely used generative AI tools are making a “material, verifiable, and substantial” difference in mitigating carbon pollution.
Joshi draws a parallel between this strategy and the approach sometimes taken by fossil fuel companies, who may publicise minor investments in renewable energy while their core business continues to generate vast emissions. He suggests that big tech companies have not only adopted but also expanded upon this tactic. The report meticulously examined claims found in an International Energy Agency document and various corporate records. It revealed that many of these claims lacked robust evidence, often relying on internal blog posts or anecdotal company experience rather than rigorous, independent academic research. For example, a frequently quoted statistic claiming AI could reduce 5-10% of global greenhouse gas emissions by 2030 was traced back to a consulting firm’s report, commissioned by Google, which offered only weak substantiation.
The escalating deployment of generative AI fundamentally depends on vast data centres, which consume immense quantities of electricity. Globally, these centres currently account for approximately 1% of the world’s total electricity usage. Projections indicate a significant increase, with electricity consumption potentially doubling in the United States alone by 2035. Energy researchers express concern that while a simple text query to a large language model demands minimal energy, more intricate functions, such as video generation and in-depth research, require considerably greater power, exacerbating the overall energy footprint.
Critics like Joshi argue that presenting AI as a viable climate solution, concurrently with the unchecked expansion of energy-hungry data centres, serves primarily as a distraction. This approach, they believe, diverts essential attention away from the severe environmental impact resulting from the unrestrained growth of these facilities.