文 | 多大NewMedia，作者 | 陈婷、Ryan Li
人工智能正在改变一个长期存在的经济学问题：重大技术创新出现之后，为什么整个经济体系往往需要很长时间才能完成相应的重构？
过去，这个问题主要表现为技术扩散与组织变革之间的时间差。今天，AI可能使问题更加复杂。
因为AI不仅能够改变生产方式，还可能参与生产流程、组织系统乃至部分产业基础设施的设计与改造。它既可能加速原有结构的变化，也可能帮助企业提高自身的适配能力。
与此同时，金融AI Agent的兴起又带来了新的问题：当大量自动化系统开始独立或协同决策，原有的责任边界、治理方式和风险传导机制是否仍然有效？
这意味着，我们需要关注的可能不只是技术发展与制度调整之间的速度差，而是技术变革、系统适配和风险传导三者之间正在形成的新关系。
一、电气化的历史：为什么技术革命不等于生产率革命？
2026年9月，多伦多大学Rotman商学院教授Kevin Bryan发表了一篇题为《AI是一项技术，而推动它扩散是管理问题》的文章。
他以电气化和集装箱运输为例，说明技术创新与经济价值充分实现之间，存在复杂的组织与产业重构过程。
在蒸汽动力时代，工厂通常围绕中央动力装置组织生产。机器通过传动轴和皮带连接，厂房结构、设备布局和生产流程都受到动力系统的约束。
电动机出现后，企业可以用电力替代蒸汽动力，但这种替代本身并不能释放电气化的全部价值。
真正的生产率改善，需要重新安排机器位置、改变工厂布局、调整生产顺序和劳动分工。独立的电力驱动使工厂能够按照工作流程而非动力传输方式组织生产，也为新的大规模生产方式创造了条件。
Bryan指出，从爱迪生第一座发电站投入运行，到美国约一半制造业实现电气化，经历了大约40年。
集装箱运输也是类似的例子。
标准化集装箱本身并不是特别复杂的技术，但要真正改变全球物流体系，需要港口、铁路、仓储、船舶和相关企业共同调整。技术发明只是起点，围绕技术重新组织产业活动才是更艰难的过程。
这两个历史案例揭示了一个重要规律：
重大技术的经济价值，不仅来自技术能力本身，还来自围绕技术展开的互补性创新、组织调整和产业生态重构。
今天的AI同样面临这个问题。
企业部署先进模型，不意味着原有工作流程、岗位配置、决策机制和业务模式就能自动转化为更高生产率。
但是，历史经验也引出了一个新的问题：AI是否会改变这种重构过程本身的速度？
二、AI的特殊性：它可能加速产业重构本身
与电力不同，AI不仅是一种被应用于生产的技术，还能够参与某些生产活动和组织流程的重新设计。
AI可以辅助编写代码、测试软件、分析业务流程、整理知识、配置系统，并在明确授权范围内执行部分跨系统任务。
随着AI Agent能力的发展，一些过去需要多人、多部门协作的数字化工作，可能获得更高程度的自动化支持。
这意味着，AI可能同时影响两个过程：
一是新技术在企业中的应用速度；二是企业围绕新技术重新设计生产与组织方式的速度。
后者尤其值得关注。
过去，一家企业进行复杂的业务转型，需要大量人力完成流程分析、系统开发、测试、培训和组织协调。AI如果能够承担其中一部分工作，就可能降低重构成本，缩短某些环节的实施周期。
这种能力甚至可能跨越企业边界，影响软件供应商、服务提供商、平台和客户之间的分工。
因此，电气化历史揭示的组织重构要求仍然成立，但其历史时间尺度未必可以直接套用于AI。
AI可能不仅加速技术扩散，还可能加速技术扩散所依赖的互补性重构。
不过，这仍然是需要验证的判断。
数字化程度高的行业可能变化更快，而实体基础设施、监管许可、跨机构协作以及劳动者技能转换，仍然受到不同约束。
产业结构被改变的速度，也不等于新的经济社会秩序能够以相同速度稳定形成。
三、产业生态真的正在加速重构吗？
要判断AI是否正在加快产业重构，不能只看模型能力或单项任务效率。
更有意义的证据，应当来自组织结构、产业分工、市场入口和价值分配关系的实际变化。
在信息服务领域，这种变化已经开始显现。
美国皮尤研究中心2025年对Google搜索行为的研究发现，出现AI生成摘要的搜索中，用户点击传统搜索结果的比例为8%；没有AI摘要时，这一比例为15%。
这项研究并不能单独证明AI摘要导致了全部点击差异，但它表明，AI摘要出现时，用户访问外部网站的行为存在显著变化。
当用户越来越多地在搜索平台内直接获得答案，传统内容提供者依赖搜索引擎获取访问量的方式可能受到影响。
这不仅是搜索效率问题，还涉及信息分发入口、用户关系以及平台与内容生产者之间的价值分配。
在企业组织层面，金融科技公司Klarna提供了另一个观察窗口。
根据其公开披露，2025年AI助手处理了约80%的客户服务聊天，公司估计其承担了相当于700多名全职客服的工作量。
这意味着AI应用已经影响客户服务的人员配置和运营方式，而不只是缩短单次交互时间。
但这种变化仍主要发生在企业内部，不能直接证明整个金融服务产业生态已经完成重构。
软件开发领域的研究则提供了更有价值的比较。
2026年，NBER发布的一项研究分析了超过50万名GitHub开发者及其AI工具使用数据。
研究发现，自主编码Agent使代码提交活动累计增加约240%，但这种增益在后续生产环节明显减弱：项目数量增加约80%，实际软件发布量增加约30%。
研究还发现，在所观察的主要软件市场中，新应用数量显著增加，但总体使用量并未同步增长。
这些结果说明，AI确实可能大幅提高某些生产环节的效率，但代码生成、产品发布和市场采用之间仍然存在不同约束。
AI加速了某些环节，并不意味着整条产业价值链能够同比例加速。
这恰好连接了电气化历史与今天的AI实践。
一方面，AI正在改变部分数字产业的组织分工和市场关系，说明重构过程可能已经开始加快。
另一方面，后续环节的瓶颈仍然存在，说明技术能力与最终经济价值之间的距离并未消失。
因此，目前更审慎的判断是：AI正在加速部分组织和产业环节的调整，但它能否系统性缩短整个产业生态的重构周期，仍需要更长时间和跨行业证据。
四、金融AI Agent：当组织重构触及责任与控制边界
金融业正在成为观察AI动态适配问题的重要领域。
传统银行数字化主要围绕金融机构自身的业务系统和服务渠道展开。
但随着第三方AI Agent的发展，客户可能越来越多地通过个人智能体获取金融信息、比较产品、管理资金，甚至授权执行交易。
银行内部也可能使用多个智能体分别完成客户识别、风险分析、产品匹配和交易处理。
这种变化不只是增加一种服务工具，而可能改变金融机构与客户之间的交互关系。
当一个完整业务流程由多个Agent共同完成时，谁对最终结果负责？
当模型能力升级、工具权限扩大或Agent之间的协作方式发生变化时，原有授权和风险评估是否仍然适用？
如果客户智能体与银行系统之间发生错误交易，银行、技术提供商和客户之间应如何分配责任？
这些问题已经成为金融业正在面对的现实治理挑战。
2026年10月19日，在外滩年会金融AI治理圆桌上，中国工商银行行长刘珺指出，商业银行从计算机系统建设进入数字化转型，而数字化转型尚未完全完成，AI转型已经到来。
这揭示了另一种时间错位：上一轮组织和技术转型尚未结束，下一轮转型要求已经开始叠加。
香港金融管理局副总裁阮国恒则强调，随着Agent应用扩大，治理需要覆盖模型选择、部署、更新、事故处理和变更管理等完整生命周期。
当不同智能体只负责局部任务，人工监督也仅覆盖部分环节时，整体责任可能变得模糊。
由此可见，金融AI治理面临的挑战并不只是为新技术增加审批规则。
更重要的是，当技术能力持续变化时，金融机构能否同步调整业务流程、决策权限、责任机制和风险控制方式。
五、从企业风险到系统性风险：AI可能改变风险传导方式
金融AI治理还有一个不能忽视的层面：不同机构的智能体之间可能形成新的相互依赖。
在外滩年会圆桌中，欧洲50集团主席Lorenzo Bini Smaghi提出，金融AI讨论不能只停留在单个机构的局部视角，还需要考虑大量智能体相互交互之后的市场整体行为。
假设多个金融机构的智能体使用相似的数据、模型和风险判断标准。
当市场出现不利信号时，它们可能在较短时间内形成相似判断，并建议或执行类似交易。
如果这些行动进一步影响市场价格和流动性，又触发其他智能体的风险控制措施，就可能形成反馈循环。
金融市场中的羊群行为和顺周期性并不是新现象。
但AI可能改变相关行为的反应速度、自动化程度和系统之间的相互作用。
这并不意味着AI必然引发市场危机。
不同模型、交易限制、人工干预和流动性缓冲，都可能降低同步行动带来的风险。
真正值得研究的是：在什么条件下，多个原本独立运行的AI系统，会通过共同数据、共同模型、交易行为和市场反馈形成新的风险传导路径？
这也说明，单个企业通过AI提高运营效率，并不意味着整个金融系统的韧性必然同步提高。
企业层面的价值实现，与金融系统层面的风险控制，必须分别观察。
六、AI既可能扩大适配缺口，也可能帮助缩小缺口
讨论AI快速迭代，很容易形成一种单向判断：技术不断加速，组织和制度只能被动追赶。
但现实可能更加复杂。
AI可以加快企业改变业务流程的速度，也可以帮助企业发现流程问题、识别风险异常、分析客户反馈和辅助调整资源配置。
监管机构同样可能借助AI提升数据分析、风险监测和监督效率。
因此，AI既可能创造新的适配压力，也可能提高系统自身的适配能力。
2026年10月15日，金融稳定理事会发布《负责任采用人工智能的稳健实践》最终报告。
报告提出12项稳健实践，其中明确包括组织适配能力、持续风险评估、绩效监测、有效人工监督以及第三方AI依赖风险管理。
这说明，AI治理正在从一次性审批转向覆盖应用生命周期的持续管理。
但制度文件的存在，并不等于治理已经有效。
如果模型发生重大变化后，企业不能及时重新评估风险；如果异常已经被发现，却没有明确的决策和执行责任；如果控制措施已经实施，却无法验证其效果，那么治理仍可能存在实际缺口。
因此，组织适配能力需要落实为一套可以运行和检验的机制：
变化识别 → 影响评估 → 决策调整 → 行动执行 → 结果验证。
它不仅要回答谁负责，还要回答谁有权调整、何时必须行动、采取什么控制，以及如何判断问题已经得到解决。
这种能力也需要持续更新。
因为AI系统的风险并不只由最初的部署方式决定，还可能随着模型升级、业务范围扩大、工具权限变化和外部环境改变而演化。
七、重新理解AI时代的动态适配与系统性风险
通用技术扩散研究早已指出，技术创新与经济价值实现之间存在互补性组织变革。
组织动态能力研究也长期关注企业如何感知环境变化、重构资源并调整自身。
金融稳定研究则已经建立了关于共同风险暴露、顺周期性和系统性风险传导的分析基础。
这些研究并未因为AI出现而失效。
但AI快速迭代、辅助组织重构以及Agent之间的自主交互，可能改变上述机制的运行条件。
因此，我们认为，理解AI时代的经济社会转型，需要将三个相互关联的过程放在一起观察。
第一，结构变化。
AI以什么速度改变生产流程、组织分工、产业关系和劳动需求？它是否正在加速互补性创新与产业生态重构？
第二，系统适配。
企业、监管机构和社会制度能否及时调整流程、资源、能力和责任机制？AI又能在多大程度上帮助这些系统提高自身的适配能力？
第三，风险传导。
当不同系统的变化速度不一致时，局部失衡如何被缓冲、吸收，或者通过金融、产业和社会关系进一步扩散？
这三个过程之间并不存在简单的线性关系。
技术变化加速，不一定意味着适配缺口必然扩大。
企业适配能力提高，也不一定意味着系统性风险同步下降。
产业生产率上升，更不意味着劳动者、消费者和社会整体必然以相同速度分享收益。
例如，一家银行可能迅速完成客户服务自动化，但客户在复杂决策中的申诉和纠错机制仍然不足。
一家金融机构可能通过AI提高自身风险管理效率，却仍然受到外部模型集中依赖和市场同步行为的影响。
而在社会层面，企业可能较快完成岗位调整，但劳动者重新学习、转换职业和恢复收入稳定，可能需要更长时间。
因此，AI时代的动态适配问题，不能仅仅理解为技术发展速度与监管立法速度之间的差距。
更重要的是理解：技术如何改变结构，系统如何调整自身，以及这些变化之间的相互作用如何影响价值实现与风险演化。
结语：真正需要加速的，不只是技术
电气化历史告诉我们，重大技术的经济价值需要围绕技术展开的生产体系和组织重构。
AI可能改变这一过程的速度，因为它不仅能够应用于生产，也可能参与部分生产流程和组织系统的重新设计。
但目前的证据也表明，单项任务效率的提高，并不必然转化为最终产品、市场需求和产业整体价值的同比例增长。
更值得关注的是，AI可能同时改变三个方面：经济结构的重构速度、组织与制度的适配能力，以及风险在不同系统之间的传导机制。
它既可能加快某些调整，也可能使不同系统之间的变化更加不均衡。
因此，AI时代真正重要的问题，不只是技术还能进步多快，也不只是监管能否追上技术。
而是：
当AI持续改变经济运行方式时，我们能否建立一种持续学习、及时调整、有效协调并验证实际结果的能力，使技术进步真正转化为可持续的经济与社会价值？
这或许是AI时代比单纯追求技术领先更为深远的管理与治理挑战。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Man, how big is NewMedia?
Artificial intelligence is transforming a long-standing economic problem: why does it take a long time for the entire economic system to complete the corresponding restructuring after major technological innovations have emerged?
In the past, the problem was mainly characterized by the time lag between technology diffusion and organizational change. Today, AI may further complicate the issue.
Because AI can not only change production patterns, it may also be involved in the design and adaptation of production processes, organizational systems and even some industrial infrastructure. It has the potential to accelerate changes in pre-existing structures or to help enterprises improve their own adaptive capacity.
At the same time, the rise of financial AI Agent has created new problems: Are the original lines of responsibility, governance and risk transfer mechanisms still in effect when a large number of automated systems begin to make independent or synergistic decisions?
This means that the possibility that we need to focus on is not just the speed gap between technological development and institutional adjustment, but the emerging relationship between technological change, system adaptation and risk transfer.
I. History of electrification: Why is the technological revolution not equal to the productivity revolution?
In September 2026, Kevin Bryan, Professor at the Rotman Business School of the University of Toronto, published an article entitled " AI is a technology and promoting proliferation is a management issue ".
He cited electrification and container transport as examples of complex organizational and industrial restructuring processes between technological innovation and the full realization of economic values.
In the age of steam power, plants are usually organized around central power units. Machines are connected through axles and belts, and plant structures, equipment layouts and production processes are bound by power systems.
After the emergence of an electric motor, an enterprise can replace steam power with electricity, but this substitution alone does not release the full value of electrification.
Real productivity improvement requires repositioning machines, changing factory layouts, reordering production and division of labour. Independent power-driven plants are able to organize their production according to workflow, not by power transmission, and create conditions for new large-scale production.
Bryan noted that it had been about 40 years since Edison ' s first power station had been operational and about half of the manufacturing industry in the United States had been electrified.
Container transport is a similar example.
Standardized containers are not by themselves particularly complex technologies, but a real change in the global logistics system requires a combination of port, rail, warehousing, ship and related enterprises. Technological innovation is only the starting point, and it is more difficult to reorganize industrial activities around technology.
These two historical cases reveal an important pattern:
The economic value of major technologies derives not only from technological capabilities themselves, but also from complementary innovations, organizational adjustments and industrial ecological restructuring around technology.
Today's AI faces the same problem.
The deployment of advanced models by enterprises does not mean that the original workflow process, staffing, decision-making mechanisms and business models automatically translate into higher productivity.
However, history has also raised a new problem: Will AI change the speed of the process itself?
Specificity of AI: It may accelerate the restructuring of industry itself
Unlike electricity, AI is not only a technology that is applied to production, but can also participate in the redesign of certain production activities and organizational processes.
AI can support the development of codes, test software, analyse business processes, collate knowledge, configure systems and carry out some of the cross-system tasks within a clear delegation of authority.
With the development of AI Agent ' s capacity, some of the digitalization that used to require multi-person, multi-sectoral collaboration may receive a higher degree of automated support.
This means that AI may influence both processes:
One is the speed with which new technologies are applied in enterprises; the second is the speed with which enterprises redesign production and organization around new technologies.
The latter is of particular concern.
In the past, a complex business transformation of an enterprise required significant human resources to complete process analysis, system development, testing, training and organizational coordination. If AI is able to take on some of these tasks, it may reduce its composition and reduce the implementation cycle of certain segments.
Such capacity may even cut across business boundaries and affect the division of labour among software providers, service providers, platforms and clients.
Therefore, the organizational restructuring requirements revealed by the history of electrification remain valid, but their historical timescale may not be directly applicable to AI.
AI may not only accelerate the spread of technology, but may also accelerate the complementary re-establishment on which it depends.
However, this is still a judgement that needs to be validated.
Industries with a high degree of digitization may change faster, while physical infrastructure, regulatory licensing, cross-institutional collaboration and the transfer of skills among workers are still subject to different constraints.
The rate at which the industrial structure has been transformed is not the same as the rate at which the new economic and social order can be stabilized.
iii. Are industrial ecology really accelerating the reconstruction?
To judge whether AI is expediting the restructuring of industry, it should not be limited to modelling capacity or individual task efficiency.
More meaningful evidence should come from actual changes in organizational structures, industrial division of labour, market entry and value distribution relationships.
In the area of information services, this change has begun to manifest itself.
A 2025 study by the Pew Research Centre of the United States of America of Google search behaviour found that in AI-generated summary searches, users clicked 8 per cent of traditional search results; in the absence of Asummary, the proportion was 15 per cent.
The study did not prove alone that AIsummary caused all hits to vary, but it showed that there had been a significant change in the behaviour of users in accessing external websites when AIsummary appeared.
As users increasingly gain direct access to answers within the search platform, the way traditional content providers rely on search engines for access may be affected.
This is not only a matter of search efficiency, but also of information distribution portals, user relations and the distribution of value between the platform and content producers.
At the corporate level, Klarna, a financial technology company, provides another observation window.
According to its public disclosure, in 2025 AI ' s assistant handled approximately 80 per cent of customer service chats, and the company estimated that it performed the equivalent of 700 full-time visitors.
This means that the AI application has affected the staffing and mode of operation of client services, rather than just shortening the single interactive time.
However, this change continues to occur mainly within enterprises and cannot be a direct proof that the entire financial services industry has been restructured.
Research in the area of software development provides a more valuable comparison.
In 2026, a study by NBERpublish analysed data on the use of more than 500,000 GitHub developers and their AI tools.
The study found that the autonomous code Agent had resulted in a cumulative increase of about 240 per cent in code submission activities, but that this gain had been significantly reduced in the subsequent production chain: the number of projects had increased by about 80 per cent and the volume of the actual software publish had increased by about 30 per cent.
The study also found that there had been a significant increase in the number of new applications in the major software markets observed, but that overall usage had not increased simultaneously.
These results suggest that AI may have significantly improved the efficiency of certain production chains, but there are still different constraints between code generation, product publish and market adoption.
AI has accelerated some of the links, which does not mean that the whole industry value chain can accelerate in proportion.
This coincides with the history of electrification and today's AI practice.
On the one hand, AI is changing the organizational division of labour and market relations in part of the digital industry, suggesting that the restructuring process may have begun to accelerate.
On the other hand, bottlenecks in the follow-up chain remain, suggesting that the distance between technological capabilities and the ultimate economic value has not disappeared.
Thus, it is now more prudent to judge that AI is accelerating the restructuring of some of its organizational and industrial components, but that its ability to systematically shorten the entire industrial ecological reconstruction cycle still requires longer time and cross-industry evidence.
IV. Financial AI Agent: When the organization is restructured to reach the boundaries of responsibility and control
The financial sector is becoming an important area for observing the dynamic adaptation of AI.
The digitalization of traditional banks is mainly centred on the financial institutions ' own business systems and service channels.
However, as third-party AI Agent develops, clients may increasingly access financial information, compare products, manage funds and even authorize transactions through individual AI agent.
Multiple AI agents may also be used within banks to complete customer identification, risk analysis, product matching and transaction processing.
This change is not merely an addition of a service tool, but may change the interaction between financial institutions and customers.
Who is responsible for the final outcome when a complete business process is being completed by multiple Agents?
Are the original authorizations and risk assessments still applicable when model capabilities are upgraded, tools permission are expanded or the modalities of collaboration between Agent change?
How should responsibility be allocated between banks, technology providers and customers in the event of an erroneous transaction between the customer AI agent and the banking system?
These issues have become real governance challenges facing the financial sector.
On 19 October 2026, at the Financial AI Governance Round Table of the Annual Conference on the Outer Beach, the President of the Bank of Commerce and Industry of China, Liu Xin, stated that commercial banks had moved from computer system construction to digital transformation, which had not yet been fully completed and had already arrived.
This reveals another time lag: the last round of organizational and technological transformation has not yet been completed and the next round of transformation requirements has begun to overlap.
The Vice-President of the Hong Kong Finance Authority, Nguyen Guo-chung, stressed that with the expansion of the Agent application, governance needed to cover the full life cycle of model selection, deployment, updating, incident management and change management.
The overall responsibility may become blurred when different AI agents are responsible only for local tasks and manual supervision covers only part of the chain.
Thus, the challenge for financial AI governance is not just to increase approval rules for new technologies.
More importantly, the ability of financial institutions to synchronize business processes, decision-making permission, liability mechanisms and risk control approaches as technological capacity continues to change.
From enterprise risk to systemic risk: AI may change risk transfer
Financial AI governance also has a dimension that cannot be ignored: there may be new interdependence between the different institutions AI agent.
At the round table of the annual meeting on the outer beach, Lorenzo Bini Smaghi, President of the European 50 Group, suggested that the financial AI discussion should not be confined to the local perspective of individual institutions, but also take into account the overall behaviour of markets after a large number of AI agent interactions.
It is assumed that AI agent in multiple financial institutions uses similar data, models and risk judgement criteria.
When there are adverse signals in the market, they may form similar judgements in a relatively short period of time and recommend or implement similar transactions.
If these actions further affect market prices and liquidity and trigger other AI agent risk control measures, a feedback cycle may emerge.
Sheep behaviour and procyclicality in financial markets are not new.
However, AI may change the speed of reaction, automation and interaction between the system.
This does not mean that AI is bound to trigger a market crisis.
Different models, trade restrictions, manual interventions and liquidity buffers can all reduce the risks associated with simultaneous operations.
What is really worth studying is the conditions under which multiple AI systems, which were originally operating independently, can create new risk transfer pathways through common data, common models, trading behaviour and market feedback.
This also suggests that the increased operational efficiency of individual enterprises through AI does not mean that the resilience of the financial system as a whole necessarily increases simultaneously.
The realization of value at the enterprise level and risk control at the financial system level must be observed separately.
VI. AI may either expand or help to close the gap
The discussion of AI ' s rapid overlap can easily lead to a one-way judgement that technology is accelerating and that organizations and systems can only catch up passively.
But reality may be more complex.
AI can speed up business process changes and help businesses to identify process problems, identify risk anomalies, analyse client feedback and assist in adjusting resource allocation.
Regulators may also use AI to improve the efficiency of data analysis, risk monitoring and oversight.
As a result, AI may create new adaptive pressures or improve the system ' s own adaptive capacity.
Final report of the Financial Stability Council, publish, 15 October 2026, " The sound practice of responsible use of artificial intelligence ".
The report presents 12 robust practices that explicitly include organizational adaptability, continuous risk assessment, performance monitoring, effective manual oversight and third-party AI reliance on risk management.
This indicates that AI governance is moving from one-time approval to continuous management covering the application life cycle.
But the existence of institutional documents does not mean that governance is effective.
There may still be a real gap in governance if, following significant changes in the model, the enterprise is unable to reassess risks in a timely manner; if anomalies are discovered without clear decision-making and enforcement responsibilities; and if control measures are implemented but their effectiveness cannot be validated.
Therefore, organizational adaptability needs to be translated into a set of mechanisms that can be operated and tested:
Change Identification Impact Assessment Decision Adjustment Action Implementation Results Validation.
It answers not only who is responsible, but also who has the power to adjust, when to act, what controls to take, and how to judge that the problem has been resolved.
This capacity also needs to be continuously updated.
Because the risks of AI are not determined only by the initial mode of deployment, they may also evolve as the model is upgraded, the scope of operations expands, the tool permission changes and the external environment changes.
VII. Reunderstanding the dynamics and systemic risks of the AI era
Research on the diffusion of generic technologies has long pointed to complementary organizational changes between technological innovation and the realization of economic values.
Organizational dynamic capacity studies also focus long-term attention on how enterprises perceive environmental change, reshape resources and adjust themselves.
For its part, the Financial Stability Study has established an analytical basis for common risk exposure, procyclicality and systemic risk transfer.
These studies were not invalidated by the emergence of AI.
However, the fast-tracking of AI, the restructuring of supporting organizations and the autonomous interaction between Agent may change the conditions under which these mechanisms operate.
We therefore believe that understanding the economic and social transformation of the AI era requires that three interrelated processes be seen together.
First, structural changes.
How fast has AI changed production processes, organizational division of labour, industrial relations and labour needs? Is it accelerating complementary innovation and industrial ecological reconstruction?
Second, systems fit.
Can enterprises, regulators and social systems adapt processes, resources, capacities and accountability mechanisms in a timely manner? To what extent can AI help these systems improve their adaptive capacity?
Third, risk transfer.
How can local imbalances be buffered, absorbed or further diffused through financial, industrial and social relations when the speed of change varies across systems?
There is no simple linear relationship between the three processes.
The acceleration of technological change does not necessarily mean that the adaptation gap will necessarily widen.
Improved enterprise adaptability does not necessarily mean that systemic risks decline simultaneously.
The rise in industrial productivity means that workers, consumers and society as a whole are bound to share the benefits at the same rate.
For example, one bank may quickly automate customer services, but complaints and error mechanisms for clients in complex decision-making remain inadequate.
A financial institution may improve its own risk management efficiency through AI, but is still affected by the concentration of external models and market synchronization.
At the social level, enterprises may be able to complete their job adjustment relatively quickly, but re-learning, transition to a profession and return to income stability may take longer.
Therefore, the dynamic adaptation of the AI era cannot be understood simply as a gap between the pace of technological development and that of regulatory legislation.
More important is to understand how technology changes its structure, how the system adjusts itself, and how the interaction between these changes affects value realization and risk evolution.
Conclusion: What really needs to accelerate is not just technology.
The history of electrification has taught us that the economic value of major technologies requires the re-engineering of production systems and organizations around technologies.
AI may change the pace of the process, as it can be applied not only to production but also to the redesign of some production processes and organizational systems.
However, current evidence also suggests that efficiency gains in individual assignments do not necessarily translate into a proportional increase in final products, market demand and the overall value of the industry.
Of even greater concern is the fact that AI may change three aspects simultaneously: the speed of the restructuring of the economy, the adequacy of the organization and the system, and the transmission mechanisms between different systems of risk.
It could accelerate certain adjustments or make changes between systems more uneven.
Thus, the real importance of the AI era is not only how fast technology can progress, but also whether regulation can catch up with technology.
They are:
When AI continues to change the way the economy works, can we build a capacity for continuous learning, timely adjustment, effective coordination and validation of results on the ground so that technological advances can truly translate into sustainable economic and social values?
This is perhaps the far-reaching management and governance challenge of the AI era more than simply pursuing technology.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.