綠色算力與氣候變遷:AI 永續轉型與決策模型的挑戰
Green Computing and Climate Change: AI Sustainable Transformation and the Challenges of Decision Models
聯合國 COP31 期間,專家探討 AI 算力能源消耗與氣候變遷的關聯。Climate Change AI 共同創辦人 David Rolnick 強調,AI 永續關鍵在於選擇合適工具,而非盲目追求更大模型,並需透過統計思維應對氣候預測的不確定性。
During the UN COP31, experts explored the link between AI computing energy consumption and climate change. David Rolnick, co-founder of Climate Change AI, emphasized that the key to AI sustainability lies in selecting appropriate tools rather than blindly pursuing larger models, and highlighted the need for statistical thinking to address the uncertainties in climate forecasting.
聯合國氣候變遷大會(COP31)近期於全球關注下召開,與此同時,臺灣環境部舉辦「綠色算力與氣候數位科技」論壇,聚焦人工智慧(AI)發展與氣候變遷之間的複雜關係。隨著生成式 AI 技術的快速擴張,其背後龐大的算力需求與能源消耗,已成為全球氣候政策中的核心議題。
The United Nations Climate Change Conference (COP31) was recently held under global attention. Concurrently, Taiwan’s Ministry of Environment hosted the "Green Computing and Climate Digital Technology" forum, focusing on the complex relationship between the development of artificial intelligence (AI) and climate change. With the rapid expansion of generative AI technology, the massive computing power requirements and energy consumption behind it have become a core issue in global climate policy.
加拿大麥吉爾大學電腦科學學院副教授、同時身為 Climate Change AI 共同創辦人的 David Rolnick 在論壇中指出,AI 與氣候變遷並非零和博弈,AI 技術在協助解決氣候問題上具有潛力,但關鍵在於必須重新評估 AI 的應用模式,而非僅是追求模型規模的擴大。
David Rolnick, an associate professor at the School of Computer Science at McGill University in Canada and co-founder of Climate Change AI, pointed out at the forum that AI and climate change are not a zero-sum game. While AI technology has the potential to help solve climate issues, the key lies in re-evaluating AI application models rather than merely pursuing the expansion of model scale.
多個專業來源共同確認,當前科技界與氣候政策制定者正面臨「AI 算力能源消耗」與「氣候目標達成」的雙重挑戰。Rolnick 在論壇中明確提出,AI 永續發展的關鍵不在於單純追求更大型的模型,而是應採取「選對 AI 工具」的策略。這意味著在氣候數位科技的應用上,必須考量能源效率與實際解決問題的效能,而非陷入盲目擴充算力的競爭。
Multiple professional sources have confirmed that the technology sector and climate policymakers are currently facing the dual challenge of "AI computing energy consumption" and "achieving climate goals." Rolnick clearly stated at the forum that the key to sustainable AI development is not simply pursuing larger models, but adopting a strategy of "choosing the right AI tools." This means that in the application of climate digital technology, energy efficiency and actual problem-solving effectiveness must be considered, rather than falling into a competition of blindly expanding computing power.
此觀點與學界對於氣候模型預測的討論相互呼應,即在面對氣候變遷這種高度複雜的系統時,模型與資料的精確度及適用性,遠比模型本身的規模更為重要。
This view echoes discussions in academia regarding climate model forecasting, namely that when facing a highly complex system such as climate change, the accuracy and applicability of models and data are far more important than the scale of the models themselves.
然而,針對 AI 如何有效預測與應對氣候風險,各界仍存在不同的解讀與尚待釐清之處。劍橋大學統計學名譽教授大衛.史畢格霍特(David Spiegelhalter)從統計學角度提醒,氣候變遷的預測本質上充滿了「深層不確定性」。在處理氣候資料時,機率、風險與未知因素交織,導致單一模型的預測結果往往無法完全涵蓋所有變數。目前論壇中對於「綠色算力」的定義與實踐路徑,與統計學界對於「模型預測極限」的警告之間,仍存在需要整合的空間。
However, there are still different interpretations and areas requiring clarification regarding how AI can effectively predict and respond to climate risks. David Spiegelhalter, Emeritus Professor of Statistics at the University of Cambridge, warned from a statistical perspective that climate change predictions are inherently full of "deep uncertainty." When processing climate data, probabilities, risks, and unknown factors are intertwined, often causing the predictions of a single model to fail to fully cover all variables. Currently, there is still a need to integrate the definitions and practical paths of "green computing" discussed in the forum with the warnings from the statistical community regarding the "limits of model prediction."
例如,AI 在最佳化能源分配上的具體成效,是否足以抵銷其訓練過程所產生的碳足跡,目前尚無定論,仍需更多實證資料支援。
For example, whether the specific effectiveness of AI in optimizing energy distribution is sufficient to offset the carbon footprint generated during its training process remains inconclusive and requires more empirical data to support.
對於南美等氣候變遷高敏感地區而言,AI 算力與氣候政策的連結具有深遠影響。南美洲擁有廣大的安地斯山脈與亞馬遜雨林,這些地區的氣候穩定性直接影響全球生態。若能透過合適的 AI 工具進行精準的氣候監測與資源分配,將有助於該地區推動綠色轉型。然而,若過度依賴大型算力模型而忽略了能源基礎設施的在地化限制,可能會對南美地區的能源供應造成額外負擔。
For highly climate-sensitive regions such as South America, the link between AI computing power and climate policy has profound implications. South America possesses the vast Andes Mountains and the Amazon rainforest, and the climate stability of these regions directly affects the global ecosystem. If appropriate AI tools can be used for precise climate monitoring and resource allocation, it will help the region promote a green transition. However, over-reliance on large-scale computing models while ignoring the localized limitations of energy infrastructure could place an additional burden on energy supplies in South America.
此外,教育界如 Chesapeake Math and IT Academy South 等機構,也正透過強調科學、數學與技術的嚴謹教育,培養下一代應對此類複雜科技議題的能力,顯示出從基礎教育到高階政策,對於科技與環境共存的重視。
Furthermore, educational institutions such as the Chesapeake Math and IT Academy South are cultivating the next generation's ability to address such complex technological issues by emphasizing rigorous education in science, mathematics, and technology, demonstrating the importance placed on the coexistence of technology and the environment from basic education to high-level policy.
儘管 AI 被視為氣候變遷的潛在解方,但各方觀點亦提醒不應過度解讀其功能。Rolnick 的觀點強調的是「工具選擇」,而非「科技萬能論」。史畢格霍特則從統計學視角提醒,專家模型與預測並非絕對真理,它們往往會忽略統計上的隨機性與未知風險。因此,在推動氣候數位科技時,必須保持批判性思考,避免將 AI 視為解決氣候危機的唯一途徑。
Although AI is seen as a potential solution to climate change, various perspectives also warn against over-interpreting its functions. Rolnick's view emphasizes "tool selection" rather than "technological panacea." Spiegelhalter reminds us from a statistical perspective that expert models and predictions are not absolute truths; they often ignore statistical randomness and unknown risks. Therefore, when promoting climate digital technology, one must maintain critical thinking and avoid viewing AI as the only way to solve the climate crisis.
限制在於,目前的算力效率提升速度,是否能跟上氣候變遷惡化的速度,以及在資源有限的情況下,如何平衡 AI 技術開發與基礎氣候適應設施的投資,仍是政策制定者必須面對的難題。
The limitation remains that whether the current rate of improvement in computing efficiency can keep up with the rate of climate change deterioration, and how to balance the investment in AI technology development with basic climate adaptation infrastructure under limited resources, remain difficult problems that policymakers must face.
後續觀察重點將聚焦於 COP31 會議中,各國對於「綠色算力」是否能達成具體的國際標準與合作協議。此外,學界與產業界如何將「深層不確定性」納入 AI 氣候模型的開發流程中,將是未來氣候預測準確度提升的關鍵。我們將持續追蹤 AI 技術在氣候變遷應用中的能源效率指標,以及南美等地區在匯入相關數位科技後的實質氣候韌性變化。同時,教育體系如何因應此類跨領域趨勢,將科學、數學與環境素養結合,亦是長期觀察的重點。
Subsequent observations will focus on whether countries can reach specific international standards and cooperation agreements regarding "green computing" during the COP31 conference. In addition, how academia and industry incorporate "deep uncertainty" into the development process of AI climate models will be the key to improving the accuracy of future climate forecasting. We will continue to track energy efficiency indicators of AI technology in climate change applications, as well as the substantial changes in climate resilience in regions like South America after the introduction of relevant digital technologies. At the same time, how the education system responds to such interdisciplinary trends by combining science, mathematics, and environmental literacy is also a key long-term observation point.