大科技生成式 AI 拯救地球說法缺乏證據

• Generative AI systems (ChatGPT, Gemini, Copilot) require substantially higher energy consumption than traditional machine learning applications • Data center electricity demand projected to grow from 1% of global consumption to at least 20% of rich world's electricity demand growth through end of decade[4] • Traditional AI applications demonstrate measurable climate benefits (e.g., optimizing wind pattern forecasting, battery chemistry testing for solar technology), but these benefits are conflated with generative AI in industry claims[3] • Google's widely-cited claim of 5-10% global greenhouse gas emissions reduction by 2030 traces back to a 2021 Boston Consulting Group blog post based on client experience rather than peer-reviewed research[3][4] • No single documented case where generative AI systems achieved 'material, verifiable and substantial level of emissions reductions'[3]

The report signals growing regulatory and stakeholder pressure on Big Tech to mandate transparency in energy consumption and emissions reporting. Industry faces credibility crisis as environmental claims face scientific scrutiny, potentially triggering mandatory disclosure requirements and stricter environmental impact assessments before data center expansion. The conflation of traditional and generative AI in climate narratives may lead to policy frameworks that differentiate between application types. Continued unrestricted data center expansion risks prolonging fossil fuel dependence and straining power grids and water supplies, potentially triggering infrastructure and environmental justice concerns that could drive legislative action.