英國綠色工作研究看AI任務曝險，培訓決策仍須回到現場證據

11月2日，Trellis刊出Lucy Shea對英國綠色工作與AI任務曝險研究的評論。研究將工作所需能力與AI能力作比較，指出部分綠色職業較偏向低曝險或人機互補。這是職業任務模型的分析，不是已觀察到的失業率，也不是工作永遠不受AI影響的保證。

On November 2, Lucy Shea discussed a new UK study of green occupations and AI exposure in Trellis. Its task-based analysis separates work with limited AI exposure from work where AI may complement people. These are modelled occupational characteristics, not observed redundancy rates or guarantees of permanent job security.

《Green Advantage》由CBI Economics製作，Futerra Solutions Union與ECIU參與委託。Trellis作者Shea與委託方具關係，讀者應將其評論視為有明示立場的研究解讀。不同機構的文件提供不同範圍的資訊，不能因刊登於媒體就將委託方觀點當成獨立驗證。

CBI Economics produced Green Advantage with commissioning involvement from Futerra Solutions Union and ECIU. Shea is connected to a commissioner. Her commentary therefore carries a disclosed perspective; publication by a media outlet does not independently validate the commissioned findings. Each supporting document has a distinct evidential scope.

研究的關鍵區分在於「AI較少觸及核心任務」與「AI可協助工作但仍需要人的判斷」。SNN編輯判讀是：兩者對培訓的意義不同，前者可能更重現場技能，後者更需界定人機分工；兩者都無法單憑職稱推算個人未來是否會失業。

The distinction between limited exposure and complementarity matters for training. SNN interprets the former as a reason to examine practical skills and the latter as a reason to test responsibility at the human-machine boundary. Neither category determines an individual worker’s employment outcome from a job title alone.

SNN建議把研究用作工作設計的起點。先列出一個職務實際處理的任務，再區分遠端資訊處理、現場安裝及安全判斷，並用實際測試確認AI能支援到哪一步。調整培訓或工作配置前，仍需比較服務品質、錯誤後果與人力需求，不能只根據「綠色」標籤決策。

SNN recommends starting with the tasks inside a role. Separate information processing, physical installation and safety-critical judgement, then test where an AI tool can assist. Before redesigning training or staffing, compare output quality, error consequences and demand. A green label alone offers little guidance about the appropriate division of work.

研究引用英國DESNZ於2025年公布的《Clean Energy Jobs Plan》：該政策文件估算潔淨能源支持的就業可由2023年約44萬個增至2030年約86萬個。這是帶有不確定性的需求情境，並非已實現或淨新增的職缺。政府資料可核對這項需求輸入，沒有獨立重現新研究的AI曝險計算。

The study uses the UK Department for Energy Security and Net Zero’s 2025 Clean Energy Jobs Plan, which projects supported clean-energy employment rising from roughly 440,000 in 2023 to 860,000 by 2030. These uncertain demand estimates are neither realised vacancies nor net additional jobs. The government data checks this input, not the new study’s AI calculations.

SNN編輯解讀是，能源建設的人力需求與AI能力進展應分別追蹤。工程專案延後會改變招募時點，工具改善則可能重排工作內容，兩種變化不必然相抵。企業若同時保留需求情境與任務測試紀錄，較能解釋培訓預算為何調整，而非以研究標題替代用人判斷。

SNN’s interpretation is that project demand and AI capability require separate monitoring. A delayed construction project can shift recruitment timing while a better tool changes task allocation. Keeping both a demand scenario and a test history makes a revised training budget easier to explain than relying on a research headline.

以下為SNN編輯分析與建議（editorial analysis）。揭露前證據基礎設施在本題的作用，是讓職務判斷能回溯到任務與測試；英國研究不代表台灣已出現相同替代率。英國潔淨能源工程團隊應在核定下一年度培訓預算前，比對職務任務清冊、現場作業工單與模型版本紀錄，標出仍需人工負責的安全與施工任務。 若國際工程客戶因英國能源轉型政策而調整技能標準，這類變動可能透過跨境工程採購合約中的人員資格與驗收要求，傳導為台灣風電工程承攬商的培訓成本與投標條件；這是條件式台灣傳導，而非已發生的本地市場結果。台灣風電與電網工程團隊應在新標案投標前，以AI任務測試紀錄和職能評量紀錄核對團隊能否完成實際施工要求，並逐項保存人員簽核。 若國際客戶更新設備維護與訓練的市場要求，變動也可能經由供應商資格條款傳導至台灣設備供應鏈，增加培訓成果的查驗需求。台灣設備供應商人資團隊應在每季檢視招募與薪資情境表，連結實際專案排程及現場作業工單，將需求推估與已核准職缺分開。職業訓練機構應在每期課程結訓驗收前，將職能評量紀錄與培訓驗收紀錄交叉核對，保存學員在AI輔助與未使用工具兩種情境下的任務結果。

SNN editorial analysis and recommendations: Pre-Disclosure Evidence Infrastructure should connect occupational judgements to tasks and tests. The UK study does not establish equivalent substitution rates in Taiwan. UK clean-energy engineering teams should compare the job-task inventory, site work-order records and model-version record before approving the next annual training budget, identifying safety and construction tasks that retain human accountability. If international engineering customers update skills standards in response to UK energy-transition policy, the changes could flow through international engineering procurement contracts into Taiwan wind contractors’ qualification requirements, training costs and tender conditions. This Taiwan transmission is conditional. Taiwan wind and grid engineering teams should reconcile an AI task-test log with competency assessment records before a new tender, testing actual installation requirements and recording individual approvals. If overseas customers revise equipment-maintenance and training requirements, supplier qualification clauses could transmit those changes into Taiwan equipment suppliers’ training and verification costs. Taiwan equipment supplier HR teams should reconcile hiring and pay-band scenarios with project schedules and site work-order records each quarter, distinguishing modelled demand from approved vacancies. Vocational training providers should compare competency assessment records and training acceptance records before each cohort’s completion review, preserving task results with and without AI assistance.

下一步值得追蹤的是研究方法能否隨AI能力與工作內容更新，以及企業採用後是否產生可比較的任務成果。SNN將把後續跨市場應用視為待核驗進展；台灣的職業分類、施工責任與訓練制度需另行對照，不能直接移植英國模型的比例。

Future checks should examine how occupational assessments change as tools and tasks evolve, and whether employers can document comparable task outcomes after adoption. Cross-market use remains a development to verify. Taiwan’s occupational categories, construction responsibilities and training arrangements require their own mapping before any UK percentage is applied locally.