和ChatGPT聊天费水？到2027年，全球AI竟然要用这么多水

标题摘要内容

来源: | 作者:国际交流部 | 发布时间: 2024-01-15 | 414 次浏览 | 分享到:

今年以来，人工智能（AI）的热潮席卷全球。不过，你知道和AI对话还会消耗水资源吗？

央视财经视频截图

综合美联社、欧洲新闻电视台报道，加州大学河滨分校电气与计算机工程系研究人员任绍磊的一项研究发现，每当人们问人工智能大模型ChatGPT 5到50个提示词或问题时，它就会“吞下”500毫升的水（接近一小瓶矿泉水的水）。而每个月都有十数亿用户在这个平台提问，这一数字增长的很快。

“It’s fair to say the majority of the growth is due to AI,” including “its heavy investment in generative AI and partnership with OpenAI,” said Shaolei Ren, a researcher at the University of California, Riverside who has been trying to calculate the environmental impact of generative AI products such as ChatGPT.

In a paper due to be published later this year, Ren’s team estimates ChatGPT gulps up 500 milliliters of water every time you ask it a series of between 5 to 50 prompts or questions. With nearly 1.5 billion users a month, that adds up fast.

央视财经视频截图

为什么AI会消耗水资源？

据央视财经报道，训练和运行AI大模型，通常需要数以万计的服务器来提供计算资源、存储资源以及低延迟的网络连接，这些服务器会以集群的方式部署在名为“数据中心（data center）”的物理场所，当它们一起工作时，会在短时间内产生高度集中的热量，所以需要大量的水资源进行冷却。

以微软和谷歌为例，2022年，微软一共用掉了约17亿加仑（约64亿升）水，相当于可以填满约2500个奥运会规格的泳池；而谷歌的数据中心和办公室则用掉了总计56亿加仑（约212亿升）水，相当于8500个奥运会规格的游泳池。

央视财经视频截图

In its latest environmental report, Microsoft disclosed that its global water consumption spiked 34% from 2021 to 2022 (to nearly 1.7 billion gallons, or more than 2,500 Olympic-sized swimming pools), a sharp increase compared to previous years that outside researchers tie to its AI research.

Google’s water consumption has also increased after growing AI efforts. As per Google’s 2023 Environmental Report, the tech giant used 5.6 billion gallons of water last year.

AI耗水量增长迅速

数据显示，现在人工智能所需要的算力预计每100天就会翻一番，并且在未来五年内可能会增长超过一百万倍。研究人员预计，到2027年，全球范围内的AI需求可能会需要消耗掉66亿立方米的水资源，相当于丹麦年总用水量的4至6倍或英国年总用水量的一半。

央视财经视频截图

According to recent studies, “operational water withdrawal of global AI may reach 4.2 – 6.6 billion cubic meters in 2027”, representing “4 – 6 times more than the total annual water withdrawal of Denmark or half that of the United Kingdom”.

除了耗水，还费电

除了耗水外，人工智能还消耗大量的电。

据《纽约时报》此前报道，一项研究分析称，到2027年，人工智能服务器每年的用电量将在85到134太瓦时（Twh）之间。这相当于阿根廷、荷兰和瑞典一年的用电量，约占目前全球用电量的0.5%。

In a middle-ground scenario, by 2027 A.I. servers could use between 85 to 134 terawatt hours (Twh) annually. That’s similar to what Argentina, the Netherlands and Sweden each use in a year, and is about 0.5 percent of the world's current electricity use.

《纽约时报》：“人工智能可能很快需要相当于整个国家的电力

这项技术背后，依赖于成千上万的专用计算机芯片。”

另据欧洲电视台，对训练人工智能至关重要的数据中心，其能源消耗占全球能源消耗的近1%。这个数字在未来几年还会上升。

Data centers, essential for AI training, account for almost 1% of the world's energy consumption. This figure is set to rise over the next few years.

然而，这些数据中心排放的二氧化碳量取决于它们所在国家是用煤炭、天然气还是可再生能源发电。在微软和艾伦人工智能研究所资助的一篇论文中，研究人员表明，通过变更人工智能的训练地点，有可能将操作的二氧化碳排放量减少75%。

But these centres emit varying amounts of CO2 depending on whether their base country produces its electricity from coal or gas, for example, or from renewable energies. In a paper funded by Microsoft and the Allen Institute for Artificial Intelligence, researchers showed that by modifying the training location of an AI, it was possible to reduce the operation's CO2 emissions by 75%.

不过，报道指出，尽管人工智能会对环境产生影响，但它在对抗全球变暖方面也发挥着正面作用。例如，人工智能可以帮助气象学家预测极端天气事件，或优化工业流程以减少二氧化碳排放。

Despite their environmental impact, AIs have a role to play in the fight against global warming. For example, they can help meteorologists predict extreme weather events, or optimise industrial processes to reduce CO2 emissions.

如今，开发者面临的挑战是透明度。大多数公开的人工智能大模型，都没有透露这些模型是在哪里训练的，或者使用它们的碳成本。

The particular challenge for developers today is transparency. Most publicly available AI models do not reveal where the models were trained, or the carbon cost of using them.

报道认为，当用户了解到这些信息时，他们可以做出更明智的选择。例如，在使用ChatGPT解答一些基础问题前，可以先在浏览器搜索下问题答案，因为浏览器搜索所消耗的能源更少。

欧洲新闻电视台：“ChatGPT比传统的互联网搜索耗能更多”

When users do have this information, they can make their own informed choices. For example, before bombarding ChatGPT with basic questions, we can just do a simple browser search, which is less power-hungry.

综合来源：央视财经，欧洲新闻电视台，纽约时报，美联社