findings are as follows: (1) AI significantly reduces household energy consumption. Controlling other factors constant, for a one-standard-deviation increase in the impact of AI, household energy consumption drops by an average of 10.751%. Robustness and endogeneity tests, including dealing with missing values, using different energy consumption and AI indicators, as well as applying instrumental variable method, placebo test and penalized regressions, confirm this conclusion. (2) Mechanism analysis shows that AI reduces energy consumption by lowering household income and increasing their financial fragility. (3) AI’s impacts on different types of energyconsumption are heterogeneous. Its negative effects are mainly observed in the significant reduction of electricity and gas consumption. Furthermore, it increases the probability of using solid fuels such as honeycomb coal, coal lumps, traditional biomass, etc., thereby increasing the reliance on low-grade energy resources and raising the risk of energy poverty. (4) AI has greater negative effects on those who do not have access to energy subsidies and households with poor energy security and stability, lower income and inadequate social security. Besides, its impacts on regions with higher levels of technological development are more prominent. (5) Feasible pathways to mitigate the adverse effects of AI are explored. It is found that improving labor protection can help alleviate its adverse consequences on energy consumption. This paper provides evidence on the impacts of technological disruption from a demand-based perspective. It highlights the need for better policies on energy, social security, income distribution and labor protection to weaken AI’s effects on household energy consumption and prevent them from falling into energy poverty.

摘要中译：能源可得性是联合国17项发展目标之一，而关于人工智能在工作中的大规模应用如何影响居民能源消费尚待系统研究。针对这一问题本文研究发现：（1）人工智能应用显著降低了家庭能源消费。在控制其他因素不变的条件下，AI的影响每提高一个标准差，家庭能源消费平均降低10.751%。这一结论在使用能源净消费指标、不同人工智能指标、工具变量法、安慰剂检验、惩罚回归等多种方法进行稳健性与内生性检验时均稳健。（2）机制分析发现，人工智能通过降低家庭收入提高了家庭财务脆弱性，从而减少了能源消费。（3）人工智能对不同类型能源消费的影响存在异质性，它的负效应主要体现为显著减少了电以及气类能源消费。同时，人工智能显著增加了使用蜂窝煤、煤块、传统生物质能等固体燃料的消费的可能性，使人们更倾向于使用低级固体燃料，增加了能源贫困可能性。（4）人工智能对没有享受到能源补贴、能源安全性与稳定性更低的家庭影响更大，对收入较低、缺乏良好社会保障等风险承担能力较弱的家庭影响更大，对技术水平更高的地区影响更突出。（5）本文从劳动保护角度考察了弱化人工智能对家庭能源消费负效应的可行路径。研究结果表明，来自劳动合同、工会等方面的劳动者权益保护有助于弱化人工智能对能源消费的不利影响。本文创新性得基于需求视角揭示出数字化变革对微观家庭的影响，具有重要的政策价值。应通过制定更具有针对性的能源补贴、社会保障、收入分配以及劳动保护政策，减少人工智能应用对家庭收入、财务平衡的冲击，弱化其对家庭能源消费的负效应，避免家庭陷入能源贫困。

作者：李超，张雨涵，李香，郝延伟

文章来源：Resources Policy 88 (2024) 104469