{"id":97125,"date":"2024-07-30T06:00:00","date_gmt":"2024-07-30T04:00:00","guid":{"rendered":"https:\/\/www.futuribles.com\/quel-impact-environnemental-de-lintelligence-artificielle\/"},"modified":"2024-10-31T11:34:56","modified_gmt":"2024-10-31T10:34:56","slug":"quel-impact-environnemental-de-lintelligence-artificielle","status":"publish","type":"post","link":"https:\/\/www.futuribles.com\/en\/quel-impact-environnemental-de-lintelligence-artificielle\/","title":{"rendered":"What is the environmental impact of artificial intelligence?"},"content":{"rendered":"
\n
The consumption of resources and energy required to run artificial intelligence (AI) systems is an increasingly high-stake issue. It is the subject of major controversy, as there is a lack of precise data to enable a proper forward-looking assessment. Operators do not communicate transparently about the energy and resource consumption of the AI systems they design and operate. Nevertheless, some studies and indirect data allow us to establish orders of magnitude.<\/h6>\n\n\n<\/div>\n
\n
\n
\n

In 2023, according to the International Energy Agency (IEA)<\/a>, data centre consumption accounted for 2% of global electricity demand, 80% of which was linked to computing operations and infrastructure cooling requirements. This figure could double by 2026. Water withdrawals from<\/a> Meta, Google and Microsoft data centres amounted to 2.2 billion cubic metres in 2022, double the annual withdrawals of a country like Denmark. Computers heat up and need to be cooled down to protect electronic components, which means that cold water circuits are still needed in most cases. Due to the dominance of GAFAM (Google, Apple, Facebook, Amazon), most of this resource consumption takes place in the United States (40% for electricity), although the geographical reach of these players is expanding.<\/p>\n

For the past five years or so, AI has been the main driver of growth in data centre resource requirements, alongside cryptocurrencies. This is because AI models have become more complex and their training is based on increasingly large datasets. Between Google\u2019s first AI model in 2017 (Transformer) and the latest version, Gemini Ultra, the computing power required has increased by a factor of 7.<\/p>\n\n <\/div>\n

\n
\n Voir aussi<\/span>\n
\n
\n \n \"\"<\/a>\n <\/span>\n \n The European Union agrees on AI Act<\/a>\n \n \n \n <\/use>\n<\/svg>\n <\/a>\n <\/span>\n
\n
23 January 2024<\/div>\n SOUPIZET Jean-Fran\u00e7ois<\/a>\n
\n 6 min.\n <\/div>\n <\/div>\n\n
\n \n \n <\/use>\n<\/svg>\n\n Newswatch\n <\/a>\n Geopolitics<\/a>\n Science and Technology<\/a>\n Society<\/a>\n <\/div>\n <\/div>\n<\/div>\n <\/div>\n <\/div>\n <\/div>\n<\/div>\n
\n\n\n
\n
\n
\n

\n \n <\/p>\n

\n Training compute of notable machine learning models by domain, 2012-2023 (in petaFLOPS*)\n <\/p>\n <\/div>\n

\n \"\"\n <\/div>\n <\/div>\n<\/div>\n
\n

*FLOPS: FLoating-point Operations Per <\/em>Second. PetaFLOPS = 1015<\/sup> FLOPS.
\nSource: <\/em>AI Index Report 2024<\/em><\/a>, <\/em>p. 51.<\/em><\/p>\n\n <\/div>\n\n\n<\/div>\n

\n
\n
\n

More computing means more electricity consumption. The training phase of ChatGPT-3 would have required 1.29 gigawatt hours[1]<\/a><\/sup> (equivalent to the average annual electricity consumption of almost 600 French people). This estimate relates to the model\u2019s training phase, but not to its use by the public. A query on ChatGPT-3 or Bloom would consume around 4 Wh, equivalent to the consumption of an LED bulb for 15 minutes. Generating images would consume 60 times more<\/a> energy than generating text. This means that training the model is a major fixed consumption item, but it is quickly overtaken by the variable consumption linked to usage, given the demand that can reach several million requests per day for models like ChatGPT-3.<\/p>\n

With regard to the water requirements of AI, a benchmark study<\/a> coordinated by Shaolei Ren (University of California) states that a typical exchange of around twenty to fifty questions and answers with ChatGPT would require half a litre of water, and estimates that by 2027, the various uses of AI would result in the consumption of between 4,200 and 6,600 billion litres of water (including that needed to cool power plants, which accounts for almost 90% of these volumes), equivalent to that of a developed country with a population of around 30 million. Google\u2019s 2024 Environmental Report<\/em><\/a> states that almost 80% of the water it uses to cool its data centres is consumed (in the form of evaporation), but this ratio falls sharply in urban areas, where the water leaving the data centres can be used to supply district heating networks (10%-20%). Google\u2019s water consumption has almost doubled since 2019, reaching 23 million litres in 2023, the equivalent of around forty golf courses. The Silicon Valley giant acknowledges that a significant proportion of its water consumption takes place in regions subject to moderate (15%) or high (15%) water stress.<\/p>\n\n <\/div>\n

\n
\n Voir aussi<\/span>\n
\n
\n \n \"\"<\/a>\n <\/span>\n \n Ecology, Civil Liberties, the Digital, Civilizations etc.<\/a>\n \n \n \n <\/use>\n<\/svg>\n <\/a>\n <\/span>\n
\n
26 June 2024<\/div>\n HA\u00cbNTJENS Jean<\/a>\n
\n 16 p.\n <\/div>\n <\/div>\n\n
\n \n \n <\/use>\n<\/svg>\n\n Journal Article\n <\/a>\n Environment<\/a>\n Society<\/a>\n <\/div>\n <\/div>\n<\/div>\n <\/div>\n <\/div>\n <\/div>\n<\/div>\n
\n\n\n<\/div>\n
\n
\n
\n

In addition to the direct resources consumed by data centres (scope 1) and the indirect resources consumed in the production of the energy used (scope 2), there are also resources consumed by companies operating data centres, both upstream and downstream in the value chain (scope 3). In particular, the production of semiconductors and computer processors (GPUs) is particularly electricity and water intensive, both during the component production and assembly phase, and during the extraction and refining of ores. On this point, the rapid obsolescence of processors undeniably contributes to a waste of resources, but the digital giants only communicate on the aggregate greenhouse gas (GHG) emissions for this scope 3 (50% of their total emissions), without any further detail on the items responsible.<\/p>\n

For companies developing AI systems, limiting the consumption of data centre resources is also a financial issue: the training costs of cutting-edge models such as ChatGPT-4 and Google Gemini run into the hundreds of millions of euros. The Web giants generally promise efficiency and sobriety gains to bring this consumption under control. In fact, Google claims that its data centres are now capable of delivering the same computing power for five times less electricity than they did five years ago.<\/p>\n

But the Silicon Valley giant is seeing its goal of carbon neutrality by 2030 undermined by developments in AI. While it is due to halve its greenhouse gas emissions compared with 2019, these have risen by 50% over this period<\/a>, a growth driven mainly by the doubling of its electricity consumption to 25 TWh in 2023, coinciding with the development of its AI-related activities.<\/p>\n\n <\/div>\n

\n
\n Voir aussi<\/span>\n
\n
\n \n \"\"<\/a>\n <\/span>\n \n The Five Barriers to Artificial Intelligence<\/a>\n \n \n \n <\/use>\n<\/svg>\n <\/a>\n <\/span>\n
\n
27 February 2023<\/div>\n BRAUNSCHWEIG Bertrand<\/a>\n
\n 20 p.\n <\/div>\n <\/div>\n\n
\n \n \n <\/use>\n<\/svg>\n\n Journal Article\n <\/a>\n Science and Technology<\/a>\n Society<\/a>\n <\/div>\n <\/div>\n<\/div>\n <\/div>\n <\/div>\n <\/div>\n<\/div>\n
\n\n\n\n

Admittedly, Google is able, through the purchase of renewable energy certificates, to communicate its annual consumption of 100% renewable electricity on a global scale. But this does not correspond to a physical reality: a third of the electricity consumed on the electricity networks by Google\u2019s thirty or so data centres around the world is of fossil origin (particularly in Asia and certain American states).[2]<\/a><\/sup> However, Alphabet\u2019s subsidiary has set itself the ambitious target of consuming low-carbon energy 24 hours a day. To achieve this, it intends to continue investing in the purchase and production of renewable electricity (equivalent to 20 GW of power production capacity since 2010). Google also wants to develop more flexible uses for its data centres, in particular by transferring computing needs from one region or time of day to another, depending on the demand for electricity on the power grids. AI could help solve these optimisation problems. As part of a collaboration with Microsoft, Google is<\/a> also investing in small nuclear reactor (SMR) technologies.<\/p>\n\n\n\n

From these scattered and essentially indirect data, a major trend emerges: the consumption of resources linked to AI systems is growing rapidly. While the level of pressure on the planet\u2019s resources caused by the growth of AI seems to be limited, there are serious problems at a local level<\/a> (water stress and stress on the electricity network). Some data centre projects are the subject of growing protests from civil society. It is difficult to assess the pressure on resources that the development of AI will represent, given the persistence of the Jevons paradox (\u2018rebound effect\u2019). The environmental impact of AI systems could nevertheless influence, if not condition, future uses and developments towards more frugal systems, and lead public authorities, companies or individual users to make choices, whether voluntary or forced.<\/p>\n\n\n\n

\u2014\u2014\u2014\u2014\u2014\u2014\u2013
N.B.: this article has been translated from French by DeepL, and revised by the author and Futuribles.<\/span><\/em><\/p>\n\n\n

    \n
  1. \n

    De Vries<\/span> Alex, \u201cThe Growing Energy Footprint of Artificial Intelligence\u201d, Joule<\/em>, vol. 7, no. 10, 18 October 2023, p. 2191-2194. \u2191<\/a><\/p>\n<\/li>\n

  2. \n

    When we consider the electricity consumed on the geographical scale of the electricity network and at an hourly time step. \u2191<\/a><\/p>\n<\/li>\n<\/ol><\/div>","protected":false},"excerpt":{"rendered":"

    The consumption of resources and energy required to run artificial intelligence (AI) systems is an increasingly high-stake issue. It is the subject of major controversy, as there is a lack of precise data to enable a proper forward-looking assessment. Operators do not communicate transparently about the energy and resource consumption of the AI systems they … Continued<\/a><\/p>\n","protected":false},"author":25,"featured_media":92629,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_seopress_titles_title":"","_seopress_titles_desc":"","_seopress_robots_index":"","_seopress_robots_follow":"","_seopress_robots_imageindex":"","_seopress_robots_snippet":"","_seopress_robots_primary_cat":"none","_seopress_robots_breadcrumbs":"","_seopress_robots_freeze_modified_date":"","_seopress_robots_custom_modified_date":"","_seopress_robots_canonical":"","_seopress_social_fb_title":"","_seopress_social_fb_desc":"","_seopress_social_fb_img":"","_seopress_social_fb_img_attachment_id":0,"_seopress_social_fb_img_width":0,"_seopress_social_fb_img_height":0,"_seopress_social_twitter_title":"","_seopress_social_twitter_desc":"","_seopress_social_twitter_img":"","_seopress_social_twitter_img_attachment_id":0,"_seopress_social_twitter_img_width":0,"_seopress_social_twitter_img_height":0,"_seopress_redirections_value":"","_seopress_redirections_enabled":"","_seopress_redirections_enabled_regex":"","_seopress_redirections_logged_status":"both","_seopress_redirections_param":"","_seopress_redirections_type":301,"_seopress_analysis_target_kw":"","footnotes":""},"categories":[140],"tags":[202,210],"keyword":[988,138,1029,1886,267],"class_list":["post-97125","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-newswatch","tag-environnement-en","tag-science-et-technique-en","keyword-intelligence-artificielle-en","keyword-environnement-en","keyword-ressources-naturelles-en","keyword-sobriete-en","keyword-developpement-durable-en"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.futuribles.com\/en\/wp-json\/wp\/v2\/posts\/97125","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.futuribles.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.futuribles.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.futuribles.com\/en\/wp-json\/wp\/v2\/users\/25"}],"replies":[{"embeddable":true,"href":"https:\/\/www.futuribles.com\/en\/wp-json\/wp\/v2\/comments?post=97125"}],"version-history":[{"count":5,"href":"https:\/\/www.futuribles.com\/en\/wp-json\/wp\/v2\/posts\/97125\/revisions"}],"predecessor-version":[{"id":99185,"href":"https:\/\/www.futuribles.com\/en\/wp-json\/wp\/v2\/posts\/97125\/revisions\/99185"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.futuribles.com\/en\/wp-json\/wp\/v2\/media\/92629"}],"wp:attachment":[{"href":"https:\/\/www.futuribles.com\/en\/wp-json\/wp\/v2\/media?parent=97125"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.futuribles.com\/en\/wp-json\/wp\/v2\/categories?post=97125"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.futuribles.com\/en\/wp-json\/wp\/v2\/tags?post=97125"},{"taxonomy":"keyword","embeddable":true,"href":"https:\/\/www.futuribles.com\/en\/wp-json\/wp\/v2\/keyword?post=97125"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}