AI精读|OECD报告-《人工智能在企业中的应用:为政策制定提供新证据》
作者：微信文章English Content Follows Below加入“知识星球”,跟“禅与Moto”一起学、用AI
企业在人工智能应用中的新洞察：为政策制定者和企业提供参考
人工智能（AI）具有显著潜力，能够提升经济生产力，应对OECD国家经济增长停滞等挑战。它还能带来其他积极成果，例如降低生产中的缺陷率（减少材料投入）以及提高能源效率（例如优化物流）。
由OECD、波士顿咨询集团（BCG）和INSEAD商学院联合开展的最新研究《人工智能在企业中的应用：为政策制定提供新证据》深入探讨了企业如何采用AI以及政府如何促进这一过程。该研究的核心是一项面向政策制定的调查，涵盖七国集团（G7）的840家企业以及巴西的167家企业，辅以对技术扩散相关公共部门机构的研究和对企业的访谈。G7国家的调查聚焦于制造业和信息通信技术（ICT）服务行业的中小型及大型企业。
人工智能应用的当前格局
在不同国家和行业中，AI应用的普及率仍然相对较低，通常被视为例外而非常态，尤其是在生成式AI出现之前。许多国家整个行业的采用率仅为个位数。然而，企业规模与AI采用之间存在强烈的正相关，规模较大的企业更有可能采用AI及多种AI技术。某些行业，如ICT和金融/保险，AI采用率高于平均水平。
积极使用AI的企业以多种方式应用它。最常见的应用包括面向客户的服务、流程控制、自动化以及生产优化。
企业面临的主要障碍
尽管AI具有潜在优势，但企业在采用AI时面临若干重大障碍：
技能稀缺：可用技能尤其是专业人才的缺乏是一个主要障碍，即使对许多大型企业也是如此。中小型企业尤其需要与大型跨国公司竞争有限的AI专家。数据成熟度不足：技术扩散机构强调，企业往往缺乏必要的数据成熟度——即收集、管理和处理正确数据的能力，这是实施AI的根本障碍。管理层理解不足：管理者常常难以理解AI是什么、采用AI涉及哪些内容、如何解决实际业务问题，并且他们往往低估了组织文化和流程所需的变革。许多人期望AI是一种简单的“即插即用”技术。投资回报率估算困难：投资回报率（ROI）的不确定性被认为是关键障碍，特别是对中小企业而言。定义和界定商业案例可能具有挑战性。监管不确定性：关于监管的不确定性可能阻止企业采取行动。企业特别希望明确AI安全使用的责任，尤其是在自主系统方面。IT基础设施不足：缺乏高速宽带等问题也可能成为障碍。
政府如何支持AI采用
研究显示，相当比例的企业使用并积极评价了旨在帮助AI采用的各种公共服务。使用更多AI应用的企业通常更有可能使用一系列公共支持服务。
最受重视和使用的支持类型包括：
信息和建议：即使是高级AI用户也寻求有关AI各个领域的额外信息。大多数企业认为公共部门提供的信息服务（如商业用例示例、预期投资回报率、供应商信息和监管更新）“有用”或“非常有用”。获取此类信息对于做出明智决策至关重要。信息匮乏似乎是采用早期阶段的主要障碍。人力资本发展：开发技能的政策和计划，尤其是专业人才，受到高度评价和广泛使用。促进再培训和终身学习投资以及大学教育和职业培训的举措被认为非常有用。融资渠道：大约42%的受访企业使用了促进融资的计划，如税收抵免或补助。公共数据和IT基础设施：收集和发布行政公共数据集的举措被认为是有用的，强调了使公共数据对企业开放的潜在好处。升级IT基础设施（如高速宽带）也得到企业的支持。
技术扩散机构的作用
致力于促进技术采用的公共或准公共机构发挥了重要作用。这些机构与企业日常互动，提供了关于挑战和有效支持机制的独特视角。它们提供的服务包括：
技术扩展服务：帮助企业定义业务问题并开发概念验证。商业咨询服务：为管理者提供非技术指导。补助：支持研发和应用公共研究，以降低AI投资风险。网络和协作平台：构建公共和私人参与者的生态系统。在职培训：解决AI技能瓶颈。信息服务和开源代码：提供资源以增强AI能力。
这些机构建议从使用现成数据的简单问题入手进行概念验证，并让经济学家与技术专家一起帮助估算投资回报率。与企业员工共同制定实施计划对于合作至关重要。
对政策制定者的启示
该研究提供了几个与政策相关的关键结论：
优先发展人力资本的举措。加强信息服务和建议，作为一种潜在高影响、低成本的措施。审查和简化访问公共数据的程序，并确保其质量。解决IT基础设施不足的问题。提供监管清晰度，特别是在AI使用的责任方面。考虑支持机制，帮助企业（尤其是中小企业）估算AI项目的投资回报率。评估技术扩散机构相对于国家广泛采用AI目标的影响和规模。评估当前资助工具，以鼓励内部AI开发和伙伴关系。
展望未来
本研究基于2022-23年收集的数据提供了宝贵见解。然而，需要注意的是，生成式AI的广泛公共使用在此之后才兴起，其对未来企业AI采用模式的影响尚待观察。随着更多企业，尤其是中小型企业，旨在成为活跃的AI用户，了解它们的需求并有效响应将变得越来越关键。
点击“阅读原文”，下载报告全文
English
The Adoption of Artificial Intelligence in Firms-New Evidence for Policymaking
AI Adoption in Firms: New Insights for Policymakers and Businesses
Artificial intelligence (AI) holds significant potential toboost economic productivity and address challenges like stagnating growth in OECD countries. It can also lead to other positive outcomes, such as lower defect rates in production (reducing material inputs) and making processes more energy efficient (e.g., optimising logistics).
A recent study, "The Adoption of Artificial Intelligence in Firms: New Evidence for Policymaking," conducted by the OECD, the Boston Consulting Group (BCG), and INSEAD Business School, delves into how companies are adopting AI and how governments can facilitate this process. The core of this study is a policy-oriented survey of840 enterprises in the Group of Seven (G7) countries and 167 in Brazil, complemented by studies of public sector institutions that help technology diffusion and interviews with firms. The survey in G7 countries focused on medium-sized and large enterprises in the manufacturing and information and communication technology (ICT) services sectors.
The Current Landscape of AI Adoption
Across various countries and sectors, the adoption of AI applications is stillrelatively low, often seen as the exception rather than the norm, particularly prior to the advent of generative AI. Single-digit adoption rates for entire sectors are common in many countries. However, there is a strong positive correlation between firm size and AI adoption, with larger firms more likely to adopt AI and multiple AI technologies. Certain sectors, such as ICT and finance/insurance, tend to have above-average AI adoption rates.
Firms actively using AI employ it in various ways. The most frequent applications includecustomer-oriented servicesand process control, automation, and optimisation of production.
Key Obstacles Facing Firms
Despite the potential benefits, enterprises face several significant barriers to adopting AI:
Scarcity of Skills: A lack of available skills, especially specialized talent, is a major hindrance, even for many large firms. Smaller firms, in particular, compete with large multinationals for limited AI specialists.Lack of Data Maturity: Diffusion institutions emphasize that enterprises often lack the necessary data maturity – the ability to gather, manage, and process the correct data – which is a fundamental barrier to implementing AI.Insufficient Management Understanding: Managers often struggle to understand what AI is, what adoption entails, how it can address real business problems, and they tend to underestimate the required changes in business culture and processes across the organisation. Many expect AI to be a simple "plug-and-play" technology.Difficulty Estimating ROI: Uncertainty about the financial return on investment (ROI) is highlighted as a critical obstacle, particularly for SMEs. Defining and delimiting the business case can be challenging.Regulatory Uncertainty: Uncertainties about regulations can prevent firms from taking steps towards adoption. Enterprises specifically seek clarity regarding accountability for the safe use of AI, especially with autonomous systems.IT Infrastructure Deficits: Issues such as a lack of high-speed broadband can also be obstacles.
How Governments Can Support AI Adoption
The study reveals that asignificant share of enterprises has used and positively values various public servicesdesigned to aid the adoption of AI. Enterprises that use more AI applications are generally more likely to use a range of public support services.
The most valued and used types of support include:
Information and Advice: Even advanced AI users seek additional information on various AI domains. A large majority of enterprises perceive information services from the public sector, such as examples of business use cases, expected ROI, vendor information, and regulatory updates, as "helpful" or "very helpful". Access to such information is crucial for making informed decisions. Information scarcity appears to be a primary barrier in the early stages of adoption.Human Capital Development: Policies and programs to develop skills, particularly specialized talent, are highly valued and used. Initiatives fostering investments in retraining and lifelong learning, as well as investing in university education and vocational training, are perceived as very helpful.Access to Finance: Approximately 42% of surveyed enterprises use programs that promote access to finance, such as tax credits or grants.Public Data and IT Infrastructure: Initiatives to gather and publish administrative public datasets are seen as helpful, emphasizing the potential benefits of making public data accessible to firms. Upgrading IT infrastructure, such as high-speed broadband, is also supported by firms.
The Role of Diffusion Institutions
Public or quasi-public institutions dedicated to facilitating technology adoption play an important role. These institutions interact daily with firms and offer a unique perspective on the challenges and effective support mechanisms. They offer services such as:
Technology Extension Services: Helping firms define business problems and develop proofs-of-concept.Business Advisory Services: Providing non-technical guidance to managers.Grants: Supporting R&D and applied public research to reduce the risk of AI investments.Networking and Collaborative Platforms: Building ecosystems of public and private actors.On-the-Job Training: Addressing AI skill bottlenecks.Information Services and Open-Source Code: Providing resources to strengthen AI capabilities.
These institutions recommend starting with straightforward problems using readily available data for proofs-of-concept and involving economists alongside technical experts to help estimate ROI. Co-developing implementation plans with firm staff is crucial for cooperation.
Implications for Policymakers
The study offers several key policy-relevant takeaways:
Prioritize human capital development initiatives.Enhance information services and advice as a potentially high-impact, low-cost measure.Review and simplify procedures for accessing public data and ensure its quality.Address IT infrastructure deficits.Provide regulatory clarity, especially regarding accountability for AI use.Consider support mechanisms that help firms, especially SMEs, estimate the ROI of AI projects.Evaluate the impact and scale of diffusion institutions relative to the national goal of widespread AI adoption.Assess current funding instruments to encourage internal AI development and partnerships.
Looking Ahead
This study provides valuable insights based on data collected in 2022-23. However, it is important to note that the widespread public use of generative AI emerged after this period, and its impact on future AI adoption patterns in firms remains to be seen. As more enterprises, particularly smaller ones, aim to become active AI users, understanding their needs and responding effectively will become increasingly critical.
页:
[1]