{"title":"The attribution problem of a seemingly intangible industry","authors":"","doi":"10.1016/j.envc.2024.101003","DOIUrl":null,"url":null,"abstract":"<div><p>We promote a more comprehensive understanding of the environmental costs associated with AI and “clouds” by raising awareness about the resource and emission-intensive nature of Artificial Intelligence through the use of data centres. While acknowledging the potential sustainability benefits of a green digital transformation, we challenge perceptions of a virtual and dematerialised industry often reinforced through the cloud metaphor. Using the planetary boundary (PB) framework, we go beyond conventional assessments of carbon emissions during a data centre's operational state to train AI and explore their multiple impacts on different Earth systems throughout the entire hardware lifecycle. For each stage of AI's hardware lifecycle, it is necessary to ascertain whether the environmental impacts associated with that stage can be quantified in a manner that allows for comparison with the control variables defined in the PB framework. In a second phase of the study, we examine the geographical distribution of individual lifecycle stages in order to assess the relative vulnerability of countries to the environmental harms associated with AI development stages. By mapping the lifecycle stages around the world, it becomes evident that the extraction, manufacturing, and disposal of AI hardware have a significant detrimental impact on the environment and populations in the Global South (the Majority World), while the benefits of AI development and use are largely concentrated in the West (the Minority World). Our findings underscore the need to scrutinize the benefits and costs associated with AI. To address the issue, it is proposed that more robust environmental policy obligations be introduced for electronic products throughout their entire lifecycle.</p></div>","PeriodicalId":34794,"journal":{"name":"Environmental Challenges","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2024-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.sciencedirect.com/science/article/pii/S2667010024001690/pdfft?md5=776da69c7767463573112524f864dc11&pid=1-s2.0-S2667010024001690-main.pdf","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Environmental Challenges","FirstCategoryId":"1085","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S2667010024001690","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"Environmental Science","Score":null,"Total":0}
引用次数: 0
Abstract
We promote a more comprehensive understanding of the environmental costs associated with AI and “clouds” by raising awareness about the resource and emission-intensive nature of Artificial Intelligence through the use of data centres. While acknowledging the potential sustainability benefits of a green digital transformation, we challenge perceptions of a virtual and dematerialised industry often reinforced through the cloud metaphor. Using the planetary boundary (PB) framework, we go beyond conventional assessments of carbon emissions during a data centre's operational state to train AI and explore their multiple impacts on different Earth systems throughout the entire hardware lifecycle. For each stage of AI's hardware lifecycle, it is necessary to ascertain whether the environmental impacts associated with that stage can be quantified in a manner that allows for comparison with the control variables defined in the PB framework. In a second phase of the study, we examine the geographical distribution of individual lifecycle stages in order to assess the relative vulnerability of countries to the environmental harms associated with AI development stages. By mapping the lifecycle stages around the world, it becomes evident that the extraction, manufacturing, and disposal of AI hardware have a significant detrimental impact on the environment and populations in the Global South (the Majority World), while the benefits of AI development and use are largely concentrated in the West (the Minority World). Our findings underscore the need to scrutinize the benefits and costs associated with AI. To address the issue, it is proposed that more robust environmental policy obligations be introduced for electronic products throughout their entire lifecycle.