The Rationale and Path of AI Innovation and Development

Editor’s Note

编者按

In August 2025, the State Council issued the “Opinions on Deeply Implementing the ‘AI+’ Action,” which states: “Accelerate the formation of a new form of intelligent economy and intelligent society characterized by human-machine collaboration, cross-border integration, and co-creation and sharing.” In September, the “Tianjin Declaration of the Council of Heads of State of the Shanghai Cooperation Organisation Member States” called for “adhering to the concept of people-centeredness and intelligence for good, and jointly preventing the risks of artificial intelligence technology.” The rapid development of artificial intelligence technology is deeply intervening in and even reshaping the operational logic of human society, bringing both enormous economic, livelihood, and governance dividends, while also harboring risks and challenges such as value misalignment and ethical imbalance. Against this backdrop, how to achieve the integration of technology and the humanities, and to promote AI to act for good and benefit humanity with scientific value criteria and effective governance approaches, is a matter of widespread social concern. This edition specially invites scholars from relevant fields to discuss the above topics.

2025年8月，国务院印发《关于深入实施“人工智能+”行动的意见》提出，“加快形成人机协同、跨界融合、共创分享的智能经济和智能社会新形态”。9月，《上海合作组织成员国元首理事会天津宣言》呼吁，“将坚持以人为本、智能向善理念，共同防范人工智能技术的风险”。人工智能技术的快速发展，深度介入乃至重新塑造着人类社会运行逻辑，既带来了巨大的经济、民生和治理红利，也隐匿着价值失准、伦理失衡等风险挑战。在此背景下，如何实现科技与人文相互融合，以科学的价值准则、有效的治理思路推动人工智能向善而行、造福人类，为社会所普遍关注。本版特邀相关领域学者，围绕上述话题开展讨论。

Participants

与谈人

Li Ling, Associate Researcher, Institute of Marxist Studies, Fudan University

李凌 复旦大学马克思主义研究院副研究员

Lin Jialin, Assistant Professor, School of Journalism and Communication, Sun Yat-sen University

林嘉琳 中山大学新闻传播学院助理教授

Lu Liqiang, Researcher, Hundred Talents Program, School of Computer Science, Zhejiang University

卢丽强 浙江大学计算机学院百人计划研究员

Moderators: Ji Wenya, Wang Jin

主持人 冀文亚 王琎

1. The Multidimensional Impact of AI Technology Development on Human Society

1.人工智能技术发展给人类社会带来多维影响

Moderator: In the digital-intelligent era, artificial intelligence is deeply integrated into human learning, life, and work scenarios, becoming a key force shaping the paradigm of human production and life, profoundly changing the world. At the same time, it is like a “double-edged sword,” bringing both enormous opportunities and triggering many challenges. Please discuss what dividends the current innovation and development of AI has brought, and what risks it may encounter.

主持人：数智时代，人工智能深度融入人类学习、生活、工作场景，成为塑造人类生产生活范式的关键力量，深刻改变着世界。同时，它也犹如一把“双刃剑”，既带来巨大机遇，也引发诸多挑战。请谈谈当前人工智能的创新发展带来了哪些红利，又可能遭遇哪些风险？

Li Ling: Currently, the global wave of intelligentization is surging, and China’s AI technology and applications are accelerating breakthroughs, bringing huge development dividends. For example, it significantly improves labor productivity, bringing tangible economic dividends. AI drives technological iteration and upgrading, industrial structure upgrading, and the conversion of old and new driving forces. While transforming and replacing traditional industries, it may lead to the reduction and disappearance of old jobs, but it simultaneously spawns new business forms and new employment. By constructing more high-end and intelligent industrial and value chains, it greatly enhances total social welfare. Another example is promoting the equalization, personalization, and precision of public services, further expanding livelihood dividends. The deep integration of AI with livelihood sectors such as education, healthcare, elderly care, and cultural tourism helps make these livelihood services personalized, inclusive, and intelligent, achieving accessible and shared high-quality public services, thereby improving the quality of life and happiness index of the people. Yet another example is enhancing the scientific, forward-looking, and effective nature of public decision-making, forming public governance dividends. AI possesses powerful capabilities in perception, mining, analysis, prediction, and collaboration. It can simulate governance actions such as policy formulation, scenario deduction, risk early warning, and decision support, providing solid guarantees for public safety, public policy issuance, grassroots grid management, and people’s city construction. The in-depth implementation of the “AI+” action has opened a new stage of deep integration and widespread application of AI with thousands of industries. AI has leaped to become the infrastructure of economic and social development, and its application prospects are promising.

李凌：当前，全球智能化浪潮风起云涌，我国人工智能技术和应用加速突破，带来巨大发展红利。例如，显著提升劳动生产效率，带来现实的经济红利。人工智能推动技术更新迭代、产业结构升级、新旧动能转换，在改造与替代传统产业时尽管会带来旧岗位的缩减与消失，但同时会催生新业态、新就业，通过构造更加高端、智能的产业链和价值链，大幅提升社会总福利。再如，推动公共服务均等化、个性化与精准化，进一步拓展民生红利。人工智能与教育、医疗、养老、文旅等民生行业的深度融合，有助于促进这些民生服务的个性化、普惠化与智能化，实现优质公共服务触手可及、人人共享，从而提升人民群众的生活品质与幸福指数。还如，提升公共决策的科学性、前瞻性与有效性，形成公共治理红利。人工智能具有强大的感知、挖掘、分析、预测与协同能力，能够模拟政策制定、情景推演、风险预警、决策辅助等治理行动，为公共安全保障、公共政策出台、基层网格化管理、人民城市建设等提供坚实保障。“人工智能+”行动的深入实施，开启了我国人工智能与千行百业深度融合、普遍应用的新阶段，人工智能跃升为经济社会发展的基础设施，其应用前景大有可为。

Lin Jialin: The rapid development of AI technology has, in just a few years, brought about major changes in the production and consumption patterns of human society. In the media field, the deep development of intelligent algorithms and the widespread application of big data have changed the modes of information production, dissemination, and consumption. Automated machine writing has accelerated the speed of online data collection, analysis, and integration. The gradual maturity of full VR scenarios and simulated AI digital human technology points towards more realistic content presentation. Personalized distribution enables platforms to more accurately focus on user “interests.” Whether media industry practitioners or the general public, everyone can clearly perceive that the development of AI technology is leading us towards a more efficient and colorful future.

林嘉琳：人工智能技术的迅速发展，在短短数年间使得人类社会的生产与消费模式发生重大变革。在传媒领域，智能算法的深度开发和大数据的广泛应用，改变了信息生产、传播、消费的模式，自动化机器写作加快了在线数据收集、分析与整合的速度，全VR场景与仿真人工智能数字人技术的逐渐成熟指向更加逼真的内容呈现，个性化分发促使平台能够更加精准地聚焦用户“兴趣”。不论是传媒行业的从业者还是普罗大众，每个人都能明显感知到人工智能技术发展正在带领我们走向一个更加高效、多彩的未来。

At the same time, the problems inherent in AI itself and its impact on people are also projected into real production and life. Intelligent algorithms greatly influence people’s information behavior and content experience, even manipulating daily life, giving rise to numerous algorithmic ethical issues. For ordinary users, the first AI ethical risks they perceive include information cocoons, algorithmic bias, and algorithmic discrimination. Many studies have shown that users are increasingly aware that intelligent algorithms are acquiring their preferences and interests and “calculating” them based on this data. In recent years, as AI technology has become more deeply embedded in people’s lives, discussions on deeper issues surrounding the algorithm black box and the digital divide have also increased. Under the multiple interactions of technology with users, platforms, and society, humanity may be pushed into a survival predicament under “algorithmic tyranny” or “algorithmic hegemony.”

与此同时，人工智能本身存在的问题及其对人的影响也投射到现实生产与生活之中。智能算法极大影响着人们的信息行为与内容体验，甚至操控了日常生活，带来诸多算法伦理问题。对于普通用户而言，最先感知到的人工智能伦理风险包括信息茧房、算法偏向、算法歧视等。很多研究都表明，用户越来越明显地感知到智能算法正在获取自己的喜好、兴趣，并依据这些数据“算计”自己。近年来，伴随人工智能技术越来越深入人们的生活，围绕算法黑箱、数字鸿沟等产生的更深层次问题的探讨也越来越多，技术在与用户、平台、社会的多重交互之下，可能将人类推向“算法暴政”“算法霸权”之下的生存困境。

Li Ling: In summary, according to the differences in AI application scenarios, the ethical risks triggered by AI can be divided into three major areas: First, in the intelligent economy field, we must be vigilant against the structural unemployment risks and the risk of widening the wealth gap brought about by AI-driven industrial upgrading and “machines replacing people,” as well as guard against risks such as big data price discrimination, financial fraud, and malicious attacks caused by data leakage or data abuse. Second, in the intelligent life field, with the large-scale application of generative AI, problems such as deepfakes, false information, data inaccuracy, and large model hallucinations are becoming increasingly prominent, greatly impacting the social credit system and seriously affecting people’s normal communication, learning, and life. In addition, risks such as group polarization caused by information cocoons, social discrimination caused by algorithmic bias, and cyber violence caused by privacy leakage must also be prevented. Third, in the intelligent governance field, the focus should be on preventing risks such as loss of technological control due to the governance “technology black box,” the digital divide risk caused by the network Matthew effect, and ideological risks brought about by foreign infiltration.

李凌：概括来说，根据人工智能应用场景的差别，可以将人工智能引发的伦理风险划分为三大领域：一是在智能经济领域，既要警惕人工智能推动产业升级、“机器换人”带来的结构性失业风险和贫富差距拉大风险等，还要防范数据泄漏或数据滥用带来的大数据杀熟、金融欺诈、恶意攻击等风险。二是在智能生活领域，随着生成式人工智能大规模应用，深度伪造、虚假信息、数据失实、大模型幻觉等问题愈发凸显，极大地冲击社会诚信体系，严重影响人们正常的交往、学习与生活；除此之外，还要防范信息茧房导致的群体极化、算法偏见导致的社会歧视以及隐私泄漏带来的网络暴力等风险。三是在智能治理领域，重点要防范治理“技术黑箱”带来的技术失控风险、网络马太效应导致的数字鸿沟风险、境外渗透带来的意识形态风险等。

Lu Liqiang: Behind the development dividends and ethical risks of AI lies a profound transformation of the mode of production. From the perspective of historical materialism, the dividends and risks brought by AI development can be grasped through the production logic of social cooperation, the collective worker, and general intellect. The unmanned and intelligent characteristics exhibited by AI are the result of the socialized production of the entire chain of AI technology, and moreover, an abstract manifestation of technological power in the process of social cooperation. Behind the development of AI technology, some people are specifically responsible for data collection and annotation, others for algorithm design and model training, and still others for computing power supply. It can be seen that behind the so-called unmanned nature, there is a large number of practitioners in the AI industry, forming a collective worker through social cooperation.

卢丽强：人工智能发展红利和伦理风险的背后，是一场深刻的生产方式变革。在历史唯物主义视域下，可以从社会协作、总体工人、一般智力等生产逻辑把握人工智能发展带来的红利与风险。人工智能所表现出的无人化、智能化特征，是人工智能技术链条全环节社会化生产的结果，更是科技力量在社会协作过程中的抽象化体现。在人工智能技术发展的背后，有的人专门负责数据采集、标注，有的人负责算法设计、模型训练，有的人负责算力供给等。可见，所谓无人化背后，有着人工智能行业的大量从业者，是社会协作形成的总体工人。

This transformation of the mode of production means that the ethical norms and value propositions originally based on the traditional industrial production mode are facing challenges from the new mode of production. Just as “the hand-mill gives you society with the feudal lord; the steam-mill, society with the industrial capitalist,” the emergence and development of AI technology, bringing about a digital-intelligent transformation of the mode of production, also signifies the arrival of an intelligent society. On this basis, the emergence of new relationships such as new human-machine relations, data ownership, and algorithm interpretability means that the relationships between people based on this mode of production need to be reshaped, which in turn requires the construction of new ethics. In this process, various development opportunities and ethical risks are historically “produced.” Thus, whether it is the development dividends of AI or the new ethical risks, both need to be grasped by delving into the actual practical production relationships.

这一生产方式的变革，意味着原来建立在传统工业生产方式基础上的伦理规范、价值主张都面临着新的生产方式挑战。正如“手推磨产生的是封建主的社会，蒸汽磨产生的是工业资本家的社会”，人工智能技术的出现和发展带来的数智化生产方式变革，也意味着智能社会的到来。在此基础上，新型人机关系、数据所有权、算法解释权等新型关系的出现，意味着在此生产方式基础上人与人的关系要被重塑，进而也就要有新伦理的建构。在此过程中，形形色色的新发展机遇和伦理风险也就被历史地“生产”出来。如此，不论是人工智能发展红利，还是新的伦理风险，都需要深入其现实实践生产的关系中去把握。

2. Clarifying the “Rationale for Good” in a Scientific Grasp of Human-Machine Relations

2.在对人机关系的科学把握中明确“从善之理”

Moderator: Science and technology are essentially created by humans and serve humans. Promoting scientific and technological innovation and development requires consideration not only from the dimensions of technological rationality and instrumental rationality but also from the dimensions of value rationality and humanistic rationality. How should the development of AI balance the relationship between technological rationality and value rationality, and what value principles should it follow?

主持人：科学技术在本质上是由人创造且为人服务的，推动科技创新发展，既要从技术理性、工具理性维度考量，也要从价值理性、人文理性维度分析。发展人工智能应该如何平衡技术理性与价值理性的关系，遵照哪些价值原则？

Li Ling: To answer this question, we must first scientifically grasp the relationship between humans and AI technology. Although AI technology, especially intelligent agents, increasingly exhibits autonomy and agency, and some countries have even granted legal personhood to intelligent agents or robots, AI technology and the entities it creates are still not independent subjects; they are artificial products formed when humans, as subjects, objectify their essential powers in the process of practice. AI is merely a technological extension of human consciousness and practice; it does not itself possess autonomous value claims or proactive goal orientation, which determines its instrumental character. In other words, the relationship between humans and AI technology remains one between a subject and an objectified tool, not a relationship between equal subjects.

李凌：对这一问题作出回答，首先需要科学把握人与人工智能技术之间的关系。尽管人工智能技术特别是智能体表现出越来越强的自主性与主体性，甚至有些国家赋予智能体或机器人法律主体资格，但人工智能技术及其形成的实体仍然不是独立的主体存在，而只是人作为主体在实践过程中将本质力量对象化而形成的人工创制物。人工智能只是人类意识和实践延伸的技术产物，自身并不具备自主价值诉求或主动目标导向，这决定了人工智能的工具性特征。换言之，人与人工智能技术的关系，仍然是主体与对象化工具之间的关系，而非对等的主体之间的关系。

Therefore, the “principle of AI pursuing good” does not mean that AI seeks good as a human subject does, but rather that in the processes of cognition and practice, humans integrate the concept and value of good into the development, innovation, and application of AI technology, making it conform to humanity’s pursuit of truth, goodness, and beauty. The ethical practice activities that humans carry out with the aid of AI are, in essence, an extension of the subject’s actions toward good and a concrete manifestation of people’s free and comprehensive development. In this process, the morality, excellence, and authority of humans as subjects are confirmed, and subjectivity is returned and highlighted. Thus, the reason AI must develop toward good lies fundamentally in humanity’s own pursuit of good and well-being; AI technology, as a confirmation and manifestation of humanity’s essential powers, is merely an instrumental means to achieve this value goal.

因此，人工智能的“从善之理”，并不是像人作为主体那样去追求善，而是在认识与实践活动中，人类将善的理念与价值融入人工智能技术的开发、创新与应用过程中，使之符合人类对真善美的追求。人类借助人工智能开展的伦理实践活动，实质上是主体向善行动的拓展，是人自由而全面发展的现实表现。在此过程中，人类作为主体的道德性、卓越性与权威性得以确证，并实现了主体性的回归与彰显。由此可见，人工智能之所以要向善发展，其根源与基础仍在于人类自身要追求善、追求福祉，而作为人类本质力量确证与表现的人工智能技术，只是实现这一价值目标的工具手段。

Lu Liqiang: In fact, to grasp “AI for good,” the fundamental value foundation lies in taking people’s comprehensive development as the starting point and anchoring it in the common values of all humanity. This can be analyzed from two dimensions: the logic of science and technology and the logic of application.

卢丽强：事实上，对智能向善的把握，根本价值根基在于以人的全面发展为原点，并锚定全人类共同价值。对此，可以从科技逻辑和应用逻辑两个维度来分析。

On the one hand, in the logic of science and technology, “AI for good” requires advancing value alignment by starting from each link of AI technological innovation. Every step in AI research and development implies value tendencies, and it should be ensured that the principles of good are embedded in the technical design itself. For example, using high-quality annotated data to train large models, avoiding data poisoning and data contamination, and enabling large models to learn mainstream human values and ethical norms. In this dimension, the “principle of AI pursuing good” is to innovate and develop AI technology that is inclusive, fair, and people-centered.

一方面，在科技逻辑上，智能向善需要从人工智能科技创新的各个环节入手，推进价值对齐。人们在进行人工智能研发的每一个环节都隐含价值倾向，应确保技术设计本身就嵌入善的准则。例如，利用高质量标注的数据训练大模型，避免数据投毒、数据污染，让大模型学习人类主流价值观和伦理规范。在这个维度，人工智能的“从善之理”就是创新发展普惠、公平、人本的人工智能技术。

On the other hand, in the logic of application, “AI for good” requires, on the basis of grasping the various application laws of AI, applications that conform to ethical norms and mainstream values. The application scenarios of AI are directly linked to individual rights and social order. AI development must be made to conform to value goals such as inclusiveness, fairness, and human well-being, and all applications of AI must meet the value standards of good, with the concept of people-centeredness and “AI for good” running through the entire application process.

另一方面，在应用逻辑上，智能向善需要在把握人工智能各种应用规律基础上，进行合乎伦理规范和主流价值观的应用。人工智能的应用场景直接关联个体权益与社会秩序，要让人工智能的发展合乎普惠公平、人类福祉等价值目标，让人工智能的各种应用符合善的价值标准，将以人为本、智能向善的理念贯穿应用全流程。

Lin Jialin: As Professor Lu said, the key to “AI for good” lies in value alignment. So, what values should be aligned with, how to achieve alignment, and the self-correction of AI moral models in this process are the issues we currently face. From the National New Generation AI Governance Expert Committee issuing the *Governance Principles for a New Generation of AI – Developing Responsible AI* in 2019, which proposed a framework and action guide for AI governance, to the release of the *Ethical Norms for a New Generation of AI* in 2021, which called for integrating ethics into the entire AI lifecycle, to the 2024 World AI Conference and High-Level Meeting on Global AI Governance issuing the *Shanghai Declaration on Global AI Governance*, which proposed “jointly promoting breakthroughs in AI technology and its development for good,” and then to the 2025 legislative work plan of the Standing Committee of the National People’s Congress, which proposed advancing legislation for the healthy development of AI, from theory to practice, our expectations for human-machine relations and the consensus on AI values are becoming increasingly clear.

林嘉琳：正如卢老师所说，智能向善的关键在于价值对齐。那么，向什么样的价值对齐、如何实现对齐，以及这一过程中人工智能道德模型的自我纠偏，是当下我们需要面对的问题。从国家新一代人工智能治理专业委员会2019年出台《新一代人工智能治理原则——发展负责任的人工智能》提出人工智能治理的框架和行动指南、2021年发布《新一代人工智能伦理规范》提出将伦理道德融入人工智能全生命周期，到2024世界人工智能大会暨人工智能全球治理高级别会议发表《人工智能全球治理上海宣言》提出“共同推动人工智能技术突破与向善发展”，再到全国人大常委会2025年度立法工作计划提出推进人工智能健康发展立法工作，从理论到实践，我们对于人机关系的期待以及关于人工智能价值的共识正在变得越来越清晰。

“AI for good” is conceptually highlighted by respecting human dignity and subjectivity. First, it ensures that AI innovation is guided by human values, meaning that human ethical values serve as the standard and goal, driving AI behavior to align with human values, moral norms, and social expectations. Through methods such as value-sensitive design, ethical values are embedded into intelligent algorithms. Second, it always ensures human “presence.” Although AI is in fact driving automation and unmanned operations, AI lacking human “presence” can neither achieve good nor sustain itself. Only by combining AI’s rational capabilities with human multiple intelligences can the role of AI be maximized and the goal of “AI for good” be realized. Third, it fully unleashes human agency and autonomy, constructing a new model of human-machine collaboration where “machines excel and humans excel even more.” The higher the performance of AI, the higher the demands it places on people’s comprehensive literacy and capabilities. Only by fully exerting their own initiative can humans better control AI systems and, with the help of AI, promote their own free and comprehensive development.

智能向善在理念上突出表现为尊重人类尊严和主体性。一是确保以人的价值引领人工智能创新，即以人的伦理价值为标准和目标，推动人工智能的行为表现与人类价值观、道德规范、社会期望对齐，通过价值敏感设计等方法，将伦理价值植入智能算法。二是始终确保人类的“在场”。尽管人工智能事实上在推动自动化与无人化，但缺乏人类“在场”的人工智能，既无法向善也难以为继。只有将人工智能的理性能力与人的多元智能结合起来，才能最大限度发挥人工智能的作用，实现智能向善的目标。三是充分发挥人的主体性与自主性，构建“机优人更优”的人机协同新模式。人工智能性能越高端，对人提出的综合素养与能力要求也就越高。人类只有充分发挥自身的能动性，才能更好地驾驭人工智能系统，借助人工智能促进自身自由全面发展。

3. Exploring the “Path to Good” in Balancing Innovation Development and Risk Prevention

3.在平衡创新发展与风险防控中探索“向善之路”

Moderator: The “Collingridge dilemma” is a core concept in the field of technology governance, revealing the contradiction between “cognitive lag” and “control lag” in the process of technological development. That is, people often find it difficult to fully understand the risks of a technology before its impacts become apparent, and by the time the risks are clear, the technology has already become deeply embedded in society, making control extremely costly or even infeasible. In AI governance, this “dilemma” is particularly pronounced. How should it be resolved?

主持人：“科林格里奇困境”是技术治理领域的核心概念，揭示了技术发展过程中“认知滞后”与“控制滞后”间的矛盾，即人们往往难以在技术影响显现前充分理解其风险，而当风险明确时，技术已深度嵌入社会，控制成本极高甚至不可行。在人工智能治理中，这一“困境”极为显著，应如何破解？

Lu Liqiang: Promoting the innovative development of AI requires both preventing premature regulation from stifling innovation and avoiding the neglect of risks due to regulatory absence. The key to resolving the “Collingridge dilemma” lies in breaking free from the binary thinking of “regulation versus innovation” and exploring a synergistic path of “using technological innovation to feed governance innovation.” Current achievements in AI technology itself, such as explainable AI, real-time risk monitoring algorithms, and data security technologies, provide technical support for this collaborative governance model. This means it is entirely possible to use technological tools to precisely identify risks, flexibly adapt to the pace of innovation, and allow governance to resonate with technological development.

卢丽强：推动人工智能创新发展，既要防止因过早规制而抑制创新活力，也要避免因监管缺位导致放任风险。破解“科林格里奇困境”，核心在于跳出“规制与创新非此即彼”的二元思维，探索“以技术创新反哺治理创新”的协同路径。当前人工智能技术自身的发展成果，如可解释人工智能、实时风险监控算法、数据安全技术等，为这种协同治理模式提供了技术支撑，这意味着我们完全有可能借助技术工具精准识别风险、灵活适配创新节奏，让治理与技术发展同频共振。

Specifically, achieving full-lifecycle dynamic governance requires implementing measures at each stage based on the characteristics of the technology: In the pre-event stage, relying on technical tools such as algorithm fairness testing and data compliance review, establish an access mechanism for high-risk applications, screen for data bias and algorithm vulnerabilities in advance, while preserving room for iteration for low-risk technologies such as entertainment-oriented generative AI. In the mid-event stage, combine regulatory sandboxes with real-time monitoring technologies, use edge computing to identify the tendency of recommendation algorithms to create information cocoons, and rely on federated learning to monitor cross-border AI security risks, achieving “trial-and-error while optimizing.” In the post-event stage, use blockchain to clarify responsibility boundaries, establish a tiered punishment mechanism for risks such as data leaks and algorithmic discrimination, ensuring both accountability with evidence and avoiding excessive punishment that stifles innovation. Of course, technological tools are not a panacea; they must be combined with sound laws and regulations, broad ethical consensus, and diverse social participation to jointly build a solid framework for AI governance.

具体而言，实现全生命周期动态治理需要紧扣技术特性分环节落地：事前环节，依托算法公平性检测、数据合规审查等技术工具，建立高风险应用准入机制，提前筛查数据偏见与算法漏洞，同时为娱乐类生成式人工智能等低风险技术保留迭代空间；事中环节，结合监管沙盒与实时监测技术，通过边缘计算识别推荐算法信息茧房倾向，依托联邦学习监测跨境人工智能安全隐患，实现“边试错边优化”；事后环节，借助区块链明确责任边界，针对数据泄露、算法歧视等风险建立分级处罚机制，既确保追责有据，又避免过度惩戒抑制创新。当然，技术工具并非万能钥匙，它必须与健全的法律法规、广泛的伦理共识以及多元的社会参与相结合，才能共同构建起坚实的人工智能治理框架。

Li Ling: The “road to goodness” for artificial intelligence is no smooth thoroughfare; it is fraught with all kinds of unknowns and uncertainties. The top priority is to grasp the right timing and pace, seeking a balance between innovation and safety. To achieve this “balance,” efforts must be concentrated in the following areas: First, enhance agility and flexibility. For emerging technological innovations, do not rush to enact rigid laws and regulations; instead, adopt means such as public policy, industry standards, ethical review, and risk early warning, so that the pace and intensity of governance can adapt to the requirements of fault tolerance and trial-and-error in AI innovation, ensuring that AI innovation complies with laws and regulations. Second, adhere to full-cycle and dynamic governance. Leverage the governance advantages of technology and institutions, establish a full life-cycle dynamic governance mechanism, set up an access evaluation mechanism for high-risk applications in the pre-event stage, pilot regulatory sandboxes in the mid-event stage, and explore tiered punishment mechanisms in the post-event stage, so as to avoid both “one-size-fits-all” regulation and unchecked technological development. Third, achieve tiered classification and precision. Implement tiered, classified, and precise governance mechanisms for different application scenarios and different technologies. In particular, for AI technology applications closely related to human welfare, social equity, ideological security, and national security, strict supervision and enhanced review should be applied, forming “circuit-breaker mechanisms” and “firewalls” for high risks.

李凌：人工智能的“向善之路”并非一片坦途，而是充满各种未知和不确定因素，重中之重是把握好时机与节奏，在创新与安全之间寻求平衡。要实现这一“平衡”，需要在这几方面着力：一是增强敏捷性和灵活性。针对新兴技术创新，不要急于出台刚性法规，而是采取公共政策、行业标准、伦理审查、风险预警等手段，使得治理节奏与强度能够适应人工智能创新在容错、试错方面的要求，确保人工智能创新合规合法。二是坚持全周期与动态性。发挥技术和制度的治理优势，建立全生命周期动态治理机制，通过事前环节对高风险应用建立准入评估机制，事中环节试行监管沙盒以及事后环节探索分级处罚机制，既要避免“一刀切”式监管，也要防止技术无序发展。三是实现分级分类与精准化。针对不同应用场景与不同技术，实行分级分类精准治理机制，特别是对于那些与人类福祉、社会公平、意识形态安全、国家安全紧密相关的人工智能技术应用，要从严监管、加强审查，形成针对高风险的“熔断机制”与“防火墙”。

Lin Jialin: Human autonomy is not only the intellectual prerequisite for humans to understand, use, and develop technology, but also the fundamental guarantee for steering AI technology. To achieve a balance between technological innovation and risk prevention in AI governance, it is necessary to give full play to the unique agency of humans, using human subjectivity to counter the ever-expanding “autonomy” of algorithms. Taking the process of news content writing and publishing as an example, in the era of intelligent communication, professional media and content platforms must still uphold the important role of human subjects such as editors and reviewers as information gatekeepers, effectively filtering out illegal or low-quality content and avoiding the risks accompanying automated generation and personalized recommendation. As content consumers, audiences need to cultivate critical thinking based on a full understanding of AI, enhance their discernment, information integration, and emotional management capabilities in an intelligent society, so as to cope with the negative effects brought by intelligent algorithms.

林嘉琳：人的自主性既是人类认识技术、使用技术、发展技术的智力前提，也是驾驭人工智能技术的根本保障。人工智能治理要实现技术创新与风险防控的平衡，就要坚持发挥人类独特的能动性，以人之主体性对抗算法日益膨胀的“自主性”。以新闻内容写作与发布的流程为例，专业媒体和内容平台在智能传播时代仍要坚持编辑、审核员等人类主体作为信息把关人的重要角色，有效排除不法或低质内容，规避伴随自动化生成与个性化推送带来的风险；作为内容消费者的受众需要建立基于对人工智能充分认知的批判性思维，提升应对智能社会的辨别能力、信息整合能力与情绪管理能力，以应对智能算法带来的负面效果。

No matter how AI technology develops, we should always start from the essential needs of human beings—that is, the needs for authenticity, certainty, and subjectivity—and seek a balance point between technological progress and human needs. In this process, the deeper connotation of autonomy—moral autonomy—must still be controlled by humans. Only then can it be ensured that the development and application of AI technology always serve human welfare.

无论人工智能技术如何发展，我们都应该从人的本质需求出发，即人对真实性、确定性、主体性的需求出发，寻找技术进步与人类需求的发展平衡点。在这一过程中，自主性的深层内涵——道德自主，仍需由人类掌控，这样才可能保证人工智能技术的开发与应用始终服务于人类福祉。

Moderator: Forming a joint force to promote AI for good is an important task in current AI governance. The convergence of this joint force is not only an inevitable choice for addressing technological challenges but also a crucial link in building a modern governance system. How should multiple subjects coordinate and cooperate to better consolidate the social synergy for AI for good?

主持人：形成推动人工智能向善的合力，是当下人工智能治理的重要课题。这种合力的凝聚，不仅是应对技术挑战的必然选择，更是构建现代化治理体系的重要一环。多元主体应该如何联动配合，才能更好凝聚智能向善的社会合力？

Lu Liqiang: The government plays the role of top-level designer and overall coordinator in AI governance. From a practical perspective, the government needs to take institutional supply and overall coordination as its core, building a basic framework for AI ethical governance. In recent years, China has issued policy documents such as the *Ethical Norms for the New Generation of Artificial Intelligence* and the *Interim Measures for the Management of Generative Artificial Intelligence Services*, clarifying the ethical bottom lines and compliance requirements for AI R&D and application. These measures have laid a solid foundation for the whole society to form the concept of AI for good. Enterprises, as the core subjects of AI technology R&D and application, need to integrate laws, regulations, and ethical requirements into the entire process of production and operation, implementing their main responsibilities. The deep involvement of the public is a key driving force for AI for good. At the level of participation, the public can provide feedback on their demands for AI ethical governance through policy-opinion solicitation platforms built by the government, or propose improvement suggestions through user feedback channels set up by enterprises. Social organizations can guide public participation in supervision by issuing risk warnings, conducting service experience surveys, and other means. At the level of literacy improvement, some primary and secondary schools in China have incorporated AI ethical knowledge into information technology courses, using case-based teaching to analyze the harms of algorithmic discrimination and information cocoons.

卢丽强：政府在人工智能治理中扮演着顶层设计者和统筹引导者的角色，从实践维度出发，政府需以制度供给与统筹引导为核心，构建人工智能伦理治理的基础框架。近年来，我国已出台《新一代人工智能伦理规范》《生成式人工智能服务管理暂行办法》等政策文件，明确了人工智能研发、应用的伦理底线与合规要求。这些举措为全社会形成人工智能向善理念奠定了坚实基础。企业作为人工智能技术研发与应用的核心主体，需将法律法规与伦理要求融入生产经营全流程，落实主体责任。公众的深度融入是人工智能向善的关键驱动力。在参与层面，公众可通过政府搭建的政策征求意见平台反馈对人工智能伦理治理的诉求，也可借助企业设立的用户反馈渠道提出改进建议。社会组织则可通过发布风险提示、开展服务体验调查等方式，引导公众参与监督。在素养提升层面，我国部分中小学已将人工智能伦理知识纳入信息科技课程，通过案例教学解析算法歧视、信息茧房的危害。

Lin Jialin: AI for good requires not only “hard law” enforced by state coercive power, but also “soft law” formulated and actively implemented by a wide range of stakeholders. Technology platforms, as the controllers of technology and content, are the direct subjects responsible for maintaining information diversity and optimizing the information environment. The main responsibilities and scope of obligations of enterprise platforms in building a diverse information environment should be clearly defined. Controlling information concentration and diversity from the content publishing end can have a positive effect on countering the homogenized information environment brought about by AI content production and dissemination. The cultivation and improvement of users’ digital literacy is a prerequisite for consolidating the social synergy for AI for good. Only when users possess digital literacy encompassing multiple dimensions such as computational thinking, data literacy, and risk awareness can they better safeguard their rights as digital citizens in a future society.

林嘉琳：智能向善既需要国家强制力保证实施的“硬法”，也需要有广泛的利益相关者参与制定并主动落实的“软法”。科技平台作为技术与内容的支配者，是维护信息多样性、优化信息环境的直接主体。应当明确企业平台在构建多元信息环境中的主体责任和义务范围，从内容发布端控制信息浓度和信息的多样性，能够对抗衡人工智能内容生产与传播所带来的同质化信息环境起到正向效果。而用户数字素养的培养和提升，是凝聚智能向善社会合力的前提。用户只有具备包含计算思维、数据素养、风险意识等多重维度在内的数字素养，才能更好地保障作为未来社会数字公民的权利。

Li Ling: From an international comparative perspective, China’s AI ethical governance differs from both the EU’s model of strong regulation and weak development and the US’s laissez-faire model of development first, regulation later. Instead, it is gradually exploring a unique model of multi-subject participation, innovation alongside regulation, and dynamic, agile governance. The important reason this governance model can build an efficient governance pattern of government guidance, corporate self-discipline, and public participation is that it properly coordinates the relationship between an effective market and a competent government. It both unleashes the initiative and agency of market subjects, stimulates the vitality of factors, allocates resources efficiently, and balances innovation and safety to the greatest extent, while also emphasizing the government’s guiding and leading functions. By strengthening top-level design and strategic planning, it promotes the formation of an AI ethical governance system, guides enterprises and society to form the concept of AI for good, achieves high-level and efficient coordination and interaction among the government, enterprises, and the public, and thus forms the “Chinese governance” of AI innovation and development.

李凌：从国际比较视野来看，中国的人工智能伦理治理，既不同于欧盟强规范、弱发展的立法模式，也不同于美国先发展、后规制的放任模式，而是逐步探索多元主体参与、边创新边规范、动态调整敏捷治理的独特模式。这一治理模式之所以能够构建政府引导、企业自律、公众参与的高效治理格局，很重要的原因就是统筹好有效市场和有为政府的关系，既发挥市场主体的主动性与能动性，激发要素活力，高效配置资源，最大限度平衡创新与安全，又强调政府的引导与主导功能，通过加强顶层设计、战略规划，推动形成人工智能伦理治理体系，引导企业、社会形成智能向善理念，实现政府、企业、公众高水平高效率的协同互动，形成了人工智能创新发展的“中国之治”。

Expert Commentary

专家点评

Chen Changfeng, Professor at the School of Journalism and Communication, Tsinghua University: In the *Opinions on Deeply Implementing the “AI+” Action* recently issued by the State Council, we are pleased to see that philosophy and social sciences have received due attention. The document clearly states, “Conduct in-depth research on the deep-level impacts and mechanisms of AI on human cognitive judgment, ethical norms, etc., explore and form a theoretical system of AI for good, and promote AI to better benefit humanity.” The discussion by the three young scholars is not only a rapid response to the national top-level design but also a vivid exploration of advancing “AI + philosophy and social sciences” and building a theoretical system of AI for good.

清华大学新闻与传播学院教授陈昌凤：在国务院近期印发的《关于深入实施“人工智能+”行动的意见》中，我们欣喜看到，哲学社会科学得到了应有的重视。该意见明确提出，“深入研究人工智能对人类认知判断、伦理规范等方面的深层次影响和作用机理，探索形成智能向善理论体系，促进人工智能更好造福人类”。三位青年学者的讨论，既是对国家顶层设计的快速响应，也是推进“人工智能+哲学社会科学”、构建智能向善理论体系的生动探索。

In recent years, theoretical research on AI has gradually integrated with various disciplines such as philosophy, Marxist theory, law, journalism and communication, and economics, and has increasingly built a new research paradigm of human-machine collaboration. This has significantly enhanced the scale and quality of research in philosophy and social sciences, forming a vibrant landscape of a hundred schools of thought contending.

近年来，关于人工智能的理论研究逐渐与哲学、马克思主义理论、法学、新闻与传播、经济学等各个学科门类结合，并日益构建起新的人机协同的研究范式，显著提升了哲学社会科学的研究规模与研究质量，形成了百花齐放的争鸣格局。

Just as AI for good requires the synergistic force of multiple subjects, theoretical research on AI also needs to adopt a multidisciplinary perspective. We not only need to promote exchange and dialogue among various branches of philosophy and social sciences, but also actively encourage dialogue and cooperation between humanities and social science scholars and scientists and engineers, jointly participating in the design and application of AI innovation, and collaboratively developing AI products with a human touch.

正如智能向善需要凝聚多元主体的协同合力，人工智能理论研究也要形成多学科交叉的视野。我们不仅需要促进哲学社会科学各门类之间的交流对话，而且要主动推动人文社科学者与科学家、工程师的对话与合作，共同参与到人工智能创新设计与应用过程中，一起研发出具有人性温度的人工智能产品。

I fully concur with the three scholars’ exploration of the “principle of goodness” and the “path toward goodness” in artificial intelligence from multiple perspectives. Artificial intelligence is a creation of humanity, an entity intimately bound up with human existence and development, and a manifestation and expression of human subjectivity. It should therefore, in its coexistence with humans, safeguard human subjectivity, rather than seek to replace, control, or even destroy humanity. In recent years, owing to the immense successes of AI technology in economic, social, and other fields, currents of thought such as dataism and AI worship have emerged in society, spreading ideas such as the supremacy of AI and the primacy of data, and even advocating the use of AI to control everything and replace humanity, attracting considerable attention. But these ideas or arguments invert the relationship between humans and technology, and between value rationality and technical rationality; they are fundamentally mistaken. AI does not possess a status higher than or equal to that of humans, nor can it cast off the value rationality that humans have infused into their creations and inventions; it must therefore necessarily submit to and serve human value rationality and goal-oriented demands.

我非常赞同三位学者从多重视角探寻人工智能的“从善之理”与“向善之路”。人工智能是人类创造出来的同人的存在与发展休戚相关的存在物，是人的主体性的彰显与表征，因而应该在与人的共存中维护人的主体性，而不是寻求对人的替代、控制甚至毁灭。近年来，由于人工智能技术在经济社会等多方面取得的巨大成功，社会上涌现出诸如数据主义、人工智能崇拜等思潮，传播人工智能至上、数据为王等理念，甚至鼓吹用人工智能控制一切、取代人类的论调，获得不少关注。但这些理念或论调颠倒了人与技术、价值理性与技术理性的关系，从根本上就是错误的。人工智能并不拥有高于或等同于人的地位，也不可能摆脱人类在创造发明它们时所注入的价值理性，因而必然要服从服务于人的价值理性与目标诉求。

Promoting the development of AI should adhere to the core concept of people-centeredness and intelligence for good. Throughout the entire life cycle of AI innovation, human life, emotions, dignity, and will must be respected, with the promotion of the free and comprehensive development of the individual and the enhancement of human well-being as the most fundamental value pursuit of AI. Specifically, value principles of people-centeredness and intelligence for good can be constructed from dimensions such as respect for human dignity, respect for human autonomy, fairness, transparency, protection of personal information, safety, responsibility, and sustainable development. First, respect for human dignity: ensure that AI serves humanity and conforms to human values and overall interests. Second, respect for human autonomy: humans should always exist as ends, and must not be alienated into tools or means by AI technology. Third, fairness: the development of AI should be committed to promoting procedural fairness, equality of opportunity, and fairness of outcomes. Fourth, transparency: ensure that AI systems are traceable, accessible, and understandable. Fifth, protection of personal information, especially individuals’ sensitive information and private data. Sixth, safety: ensure that AI does not endanger human survival safety, nor infringe upon individual physical and mental health and life safety. Seventh, responsibility: bear legal and moral responsibility for adverse consequences caused by AI. Eighth, sustainable development: coordinate the relationship among humans, AI technology, and nature, achieving the synergistic development of humanity, nature, society, and AI, and enhancing overall human well-being.

推动人工智能发展应坚持以人为本、智能向善的核心理念。在人工智能创新的全生命周期中都要尊重人的生命、情感、尊严、意志等，将促进人的自由全面发展和增进人类福祉作为人工智能最根本的价值追求。具体来说，可以从尊重人格尊严、尊重人类自主、公平、透明、保护个人信息、安全、责任和可持续发展等维度构建以人为本、智能向善的价值原则。一是尊重人格尊严，确保人工智能为人类服务，符合人类价值观与整体利益。二是尊重人类自主，人应该始终作为目的而存在，而不能被人工智能技术异化为工具或手段。三是公平，人工智能的发展要致力于促进程序公平、机会公平与结果公平。四是透明，保证人工智能系统具有可追溯性、可访问性与可理解性。五是保护个人信息，特别是个体的敏感信息、隐私信息。六是安全，确保人工智能不危及人类生存安全，不侵犯个体身心健康与生命安全。七是责任，对人工智能造成的不良后果承担法律与道德责任。八是可持续发展，协调好人、人工智能技术与自然的关系，实现人类、自然、社会与人工智能的协同发展，增进人类整体福祉。

AI is not a heartless thing, nor is it value-free or value-neutral. From the moment it was invented and created, AI should embody value aspirations such as humanity’s pursuit of truth, goodness, and beauty, the expansion of freedom, the enhancement of efficiency, and the manifestation of subjectivity. Integrating these value concepts into the entire life cycle of AI is the fundamental reason why we adhere to people-centeredness and intelligence for good.

人工智能并非无情物，也非价值无涉或价值中立物。从其被发明创造出来开始，人工智能就应该蕴含人类追求真善美、拓展自由、提高效率、彰显主体性等价值诉求，将这些价值理念融入人工智能全生命周期，这便是我们坚持以人为本、智能向善的根本缘由。

Online editor: Huihui

网络编辑：慧慧