当前位置:首页 > 报告详情

人工智能增加温室气体排放的风险.pdf

上传人: 分** 编号:930190 2025-10-02 24页 1.72MB

1、Risks of AI Increasing Greenhouse Gas EmissionsDavid RolnickMila Quebec AI InstituteMcGill UniversityClimate Change AI“AI can help solve climate change,but it uses a lot of energy”?23Estimated total energy consumption of AIComparisonTotal electricity consumption in 2023:Global:29,479.05 TWhLow-incom

2、e countries:126.35 TWhUnited States:4,249.05 TWhSource:Our World in DataAdapted by Lynn Kaack from RoegenCentre for Sustainability and INFRAS with data from Schneider Electric,2023;Schneider Electric,2024;De Vries,2023;IEA,2024;Morgan Stanley,2024;Goldman Sachs,2024;Semianalysis,2024 4Embodied emiss

3、ionsfrom production and end-of-life of hardwareMining and productionTransportationDisposal and recyclingOperational emissionsfrom energy consumed during computationComputational costs of AIKaack,Donti,Strubell,Kamiya,Creutzig,Rolnick,“Aligning AI with Climate Change Mitigation,”Nature Climate Change

4、 2022.“AI can help solve climate change,but it uses a lot of energy”?56A menagerie of AI algorithmsHuge models/high energy usageSmaller models/low energy usageGenerate text/images/videoClassify/optimize/predict/etc.Often very broad goals and dataSomewhat narrower scope and dataA few uses in climate

5、actionWidely used across sectorsChatGPT DeepSeekMidjourney U-Nets,ResNets,LSTMsPhysics-informed NNs Interpretable ML 7A menagerie of AI algorithmsChatGPT DeepSeekMidjourney High-computation AI is only useful sometimesU-Nets,ResNets,LSTMsPhysics-informed NNs Interpretable ML 8AI use cases across soci

6、etyWhat algorithms do we need to be effective?9Right approachAI use cases across societySome big generative AI Lots of small,focused AI10Wrong approachAI use cases across societyLLMs for everything!11Wrong approachAI use cases across societySmaller LLMs for everything!12Wrong approachAI use cases ac

word格式文档无特别注明外均可编辑修改,预览文件经过压缩,下载原文更清晰!
三个皮匠报告文库所有资源均是客户上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作商用。
根据标记内容,全文主要探讨了人工智能(AI)在解决气候变化中的作用及其潜在的环境风险。以下是核心数据和关键点: 1. **AI能源消耗**:2023年全球AI总能耗约为29,479.05 TWh,其中美国占比4,249.05 TWh。 2. **AI算法多样性**:包括ChatGPT、DeepSeek、Midjourney等,其中大型模型能耗高,小型模型能耗低。 3. **AI应用**:AI在气候变化中的应用包括但不限于: - 优化能源使用(如预测需求) - 改变消费模式(如AI广告影响消费) - 提高能源生产效率(如AI在化石燃料勘探中的应用) 4. **AI与气候变化的结合**: - AI可以针对特定问题进行设计,以减少能源消耗。 - 所有AI应用都应考虑其对气候的影响。 5. **政策建议**: - 改善数据标准和数据共享。 - 为AI研究和创新提供有影响力的资金。 - 建立跨部门创新中心。 - 增强气候相关行业、政府和民间社会的AI能力和素养。 - 建立负责任和参与式的设计最佳实践。 - 在塑造新技术发展时考虑潜在的气候影响。
"AI助力气候变暖吗?" "小AI模型能否解决大问题?" "AI如何影响快时尚消费?"
客服
商务合作
小程序
服务号
折叠