Artificial intelligence capital stock in Europe, UK, USA and Japan 1995 - 2020

Following (Rimmington et al., 2018), we target the EUKLEMS & INTANProd database data on capital stocks, categorized by SIC code and country, including software, databases, computer hardware, and machinery. From their definition of AI, we measure four distinctive AI capital stock categories:
1. Data and equipment: computing equipment, communication equipment and compute software and databases, chained linked volumes (2015), millions of national currencies.
2. Research and development: chained linked volumes (2015), millions of national currency.
3. Intangible assets: organizational capital, brand, industrial design, chained linked volumes (2015), millions of national currency.
4. Skills: training, labor compensation for NACE_R2 Computer programming, consultancy, and information service activities and NACE_R2 Education, chained linked volumes (2015), millions of national currency.
Building on our methodology, we get a closer proxy for AI stock using the data from (Bontandini et al., 2023) and (Rimmington et al., 2018) approach. We improve them to get more reliable and exact AI stock measurements using four AI dimensional coefficients. The AI dimensional coefficients approach we develop here builds on AI intensify coefficient applied to calculate AI investments (Evas et al., 2022).
We derive four AI dimensional coefficients following (Evas et al., 2022) analysis of AI investments in the EU. These coefficients will enable us to calculate the AI net stock per country accurately.
1. AI dimensional coefficient 1 (AI1): number of AI economic players / GDP in billion € (ratio). For details, see G2: AI player intensity (Righi et al., 2021),
2. AI dimensional coefficient 2 (AI2): number of AI players in AI R&D / GDP in billion € (ratio). For details, see R1: AI players in AI R&D (Righi et al., 2021),
3. AI dimensional coefficient 3(AI): University places with AI content in the EU(Bachelor and master level) / GDP in millions € (ratio). For details, see S6: University places with AI content in the EU (Righi et al., 2021). Data for the USA, UK, and Japan for AI dimensional coefficients AI3 were derived using (OECD.AI, 2023),
4. AI dimensional coefficient 4 (AI4): AI investments in the EU in million € / GDP in billion €, (ratio). For details, see G6: AI investments in the EU (Righi et al., 2022). Data for the USA, UK, and Japan for AI dimensional coefficients AI4 were derived using - venture capital investments in AI start-ups (OECD.AI, 2023).
We have developed and implemented AI dimensional coefficients (AI1 - AI4) to improve the accuracy of data analysis. Our revised data is presented as follows: Total revised AI net capital stock = AI data and equipment net capital stock x (AI1) + AI research and development net capital stock x (AI2) + AI skills stock net x (AI3) + AI net intangible assets stock x (AI4).

参照（Rimmington等，2018）的研究框架，本数据集聚焦于EUKLEMS & INTANProd数据库中按标准产业分类（Standard Industrial Classification, SIC）和国别划分的资本存量数据，涵盖软件、数据库、计算机硬件与机械设备。基于该研究对人工智能的定义，我们划定了四类差异化的人工智能资本存量范畴：
1. 数据与设备：包含计算设备、通信设备、计算软件及数据库，采用2015年链式加权不变价，单位为百万本国货币。
2. 研发投入：采用2015年链式加权不变价，单位为百万本国货币。
3. 无形资产：涵盖组织资本、品牌、工业设计，采用2015年链式加权不变价，单位为百万本国货币。
4. 技能维度：包含培训支出、针对欧盟产业分类修订版2（NACE_R2）下计算机编程、咨询及信息服务活动的劳动报酬，以及NACE_R2分类下教育活动的相关投入，采用2015年链式加权不变价，单位为百万本国货币。

基于本研究的方法体系，结合（Bontandini等，2023）与（Rimmington等，2018）的数据集与研究思路，我们得到了更贴合实际的人工智能资本存量代理指标。通过引入四类人工智能维度系数，我们进一步优化了原有测算方法，获得了可靠性与精准性更强的人工智能资本存量测算结果。

本次提出的人工智能维度系数测算框架，源自（Evas等，2022）用于测算人工智能投资的人工智能强度系数方法。我们参照（Evas等，2022）对欧盟人工智能投资的分析框架，推导得到四类人工智能维度系数，借此可精准测算各国人工智能净资本存量：
1. 人工智能维度系数1（AI1）：人工智能经济主体数量与以十亿欧元计的国内生产总值（GDP）之比（比值）。详细说明参见指标G2：人工智能主体强度（Righi等，2021）。
2. 人工智能维度系数2（AI2）：人工智能研发领域的人工智能主体数量与以十亿欧元计的国内生产总值（GDP）之比（比值）。详细说明参见指标R1：人工智能研发领域主体数量（Righi等，2021）。
3. 人工智能维度系数3（AI3）：欧盟境内开设人工智能相关课程的本科及硕士学位席位数量与以百万欧元计的国内生产总值（GDP）之比（比值）。详细说明参见指标S6：欧盟境内人工智能相关课程学位席位（Righi等，2021）。针对美国、英国与日本的AI3系数数据，我们通过经合组织人工智能数据库（OECD.AI, 2023）获取。
4. 人工智能维度系数4（AI4）：欧盟境内人工智能投资规模（以百万欧元计）与以十亿欧元计的国内生产总值（GDP）之比（比值）。详细说明参见指标G6：欧盟境内人工智能投资规模（Righi等，2022）。针对美国、英国与日本的AI4系数数据，我们通过人工智能初创企业风险投资数据（OECD.AI, 2023）获取。

我们构建并应用了AI1至AI4四类人工智能维度系数，以提升数据分析的精准度。经优化后的人工智能总净资本存量测算公式如下：
修正后人工智能总净资本存量 = 人工智能数据与设备净资本存量 × AI1 + 人工智能研发净资本存量 × AI2 + 人工智能技能净资本存量 × AI3 + 人工智能无形资产净资本存量 × AI4