On Socialist Ideology Empowering the "Value Alignment" of Large-Scale AI Models

As artificial intelligence penetrates ever deeper into every corner of human life, the value orientations it embodies are gradually coming to light, and large-scale AI models are becoming a veritable "ideological machine" for shaping social values. In 2014, Professor Stuart Russell of the University of California, Berkeley, first raised the "value alignment problem" in an interview on the "myths of artificial intelligence," arguing that "we need to build intelligences that are provably aligned with human values, not just pure intelligence." Since the 2020s, Western countries led by the United States have keenly sensed the critical impact of "value alignment" on the development of the AI industry and even on the global competitive landscape, and have accordingly elevated "value alignment" from an academic proposition to a political issue, accelerating the deployment of AI "value alignment" engineering. In March 2021, the U.S. National Security Commission on Artificial Intelligence stated bluntly that American values should become the guiding principles for the R&D and application of global AI systems. In July 2025, the White House proposed leveraging the U.S. position in international diplomacy and standards-setting bodies to further advocate for international AI governance schemes that reflect American values, countering China's influence in AI standard-setting. At the same time, the G7-led Global Partnership on AI (GPAI) echoed this sentiment, working through the OECD to embed Western ideology into international AI technical standards. Furthermore, "Anthropic, a U.S. company, released a 'Constitutional AI' value alignment pathway," pushing AI to directly cite the First Amendment of the U.S. Constitution when answering questions, marking the entry of U.S. "value alignment" into a substantive implementation phase. It is evident that the choice of direction, standard-setting, and power struggles behind the control of "value alignment" are in fact a deep-seated contest between different ideologies in the global AI arena—a strategic means by which Western countries attempt to package their own ideology as "universal values," thereby consolidating technological hegemony and exporting political ideas through control of the discourse on "value alignment."

General Secretary Xi Jinping, during the 20th group study session of the Political Bureau of the CPC Central Committee, stressed: "We must grasp the development trends and patterns of AI, and accelerate the formulation and improvement of relevant laws, regulations, policies, systems, application norms, and ethical standards, and build systems for technical monitoring, risk early warning, and emergency response, to ensure that AI is safe, reliable, and controllable." In September 2024, the China Cybersecurity Standardization Technical Committee issued the "AI Safety Governance Framework," proposing that China "continuously promote the alignment of values and ethics in the design, development, and application of AI." Therefore, constructing China's own AI "value alignment" scheme, with socialist ideology at its core, is an urgent necessity. However, current Chinese research on "value alignment" is mainly confined to passive defense frameworks such as "engineering control theory," "legal punishment theory," and "moral constraint theory." There is a relative scarcity of practical research on how to activate the proactive empowerment of socialist ideology and how to transform China's core socialist values into encodable, verifiable underlying code for large models. Based on this, systematically clarifying the resource elements of socialist ideology in empowering the "value alignment" of large-scale AI models, establishing scientific practical principles, and exploring feasible implementation pathways are not only necessary requirements for promoting the healthy and sustainable development of AI in the New Era, but also key measures for safeguarding China's national ideological sovereignty and security.

I. The Logic of Socialist Ideology's Emergence in Empowering the "Value Alignment" of Large-Scale AI Models

"Alignment with whom" is the primary link in the "value alignment" of large-scale AI models. It cannot be understood merely as alignment with certain value rules and conditional judgments, but rather as alignment with a systematic, directional, and guiding ideological system. Looking at the two major ideologies worldwide, the primary value orientation of capitalist ideology is the maximization of profit. If AI were systematically implanted with a value system of "maximizing profit and efficiency," it would lead large models to exploit humans in a more oppressive manner, which clearly runs counter to the original intent of value alignment. Socialist ideology centrally reflects the people-centered development concept and the value pursuit of a community with a shared future for mankind. It possesses abundant foundational value resources, process-oriented steering resources, and outcome-guaranteeing resources, making it the optimal choice for the "value alignment" of China's large-scale AI models.

(1) Socialist Ideology Contains the Foundational Value Resources Needed for the "Value Alignment" of Large-Scale AI Models

From the perspective of the service targets of large-scale AI models, four value dimensions can be identified: individual civic values, social group values, national holistic values, and global common values. Individual civic values mainly concern how large models meet or influence the needs and interests of single users; group values extend further to the community or organizational level, exploring how large models promote harmony, efficiency, and cooperation within groups; holistic values are rooted in the overall national development landscape, highlighting the supporting efficacy of large models for national strategy implementation; common values, as the highest level, transcend geographical boundaries, focusing on the contribution of large models to the overall well-being of human society, sustainable development, and global governance.

Marx and Engels pointed out in *The German Ideology*: "The ideas of the ruling class are in every epoch the ruling ideas." In the Chinese context, socialist ideology is naturally the ideology that coalesces broad social consensus and occupies a dominant position. From the perspective of value resource supply, socialist ideology contains value resources such as "core socialist values" and "common values of all humanity," which can be utilized as foundational resources for the "value alignment" of large-scale AI models. Specifically, at the level of individual civic values, the core socialist values of "patriotism, dedication, integrity, and friendliness" anchor the value orientation for large models in providing personalized services; at the level of social group values, the core socialist values of "freedom, equality, justice, and rule of law" can guide large models in promoting the orderly functioning and overall efficiency of society; at the level of national holistic values, the core socialist values of "prosperity, democracy, civility, and harmony" provide fundamental guidance for large models in assisting national liberation and the great rejuvenation of the Chinese nation. At the level of global common values, China advocates for the common values of all humanity. In the "Global AI Governance Initiative" released at the Third Belt and Road Forum for International Cooperation, it was explicitly stated that the common values of all humanity, centered on "peace, development, fairness, justice, democracy, and freedom," are inherent requirements for AI for good, clarifying China's position on AI governance. In contrast, the "universal value" alignment schemes touted by the Western world weaponize AI for geopolitical games, failing to transcend "narrow nationalism, party conservatism, national priority-ism, and national exceptionalism," and run counter to the vision of common progress in human digital civilization. Therefore, only by actively practicing the common values of all humanity on a global scale can we promote the sustainable development and common prosperity of human digital civilization.

(2) Socialist Ideology Possesses the Capability for Whole-Process Steering in the "Value Alignment" of Large-Scale AI Models

"The value alignment of AI agents cannot be generalized or achieved overnight; it should adopt a distributed embedding approach." The value alignment of large-scale AI models can be broken down into two major stages: forward alignment and backward alignment. Forward alignment, also known as the alignment training stage, focuses on training the system according to specific value requirements to achieve preliminary value consistency. Backward alignment, also known as the alignment refinement stage, focuses on continuously evaluating the system in real-world scenarios to ensure that the trained system consistently maintains value alignment in actual application.

Compared with the capitalist ideological process of “value alignment,” which is hijacked by the logic of capital and serves the utilitarian interests of a few monopoly groups, the Chinese socialist ideological process of “value alignment” has always been people-centered and displays an incomparable superiority. In the forward alignment stage, the West relies on the narrow business ethics and chaotic values of enterprises, “technology becomes a tool for capital accumulation and the debasement of human beings,” the training corpus is manipulated by capital, and value bias is severe. For instance, content generated by ChatGPT once conflated chimpanzees with Black people, “Amazon’s automated hiring algorithm tended to give low scores to résumés containing the character for ‘woman’ (女),” and so on, revealing the racial and gender discrimination latent in the underlying corpus. In contrast, Chinese socialist ideology, with its profound theoretical foundation and broad social acceptance, provides a solid value basis for the AI training process. Socialist ideology can systematically integrate core socialist values, ethical norms, laws and regulations, and other elements into the training process through “corpus supply,” ensuring from the source that AI is anchored in a “people-centered” direction. Moreover, compared with Western abstract theories of human nature, it eliminates more metaphysical dogmatic elements, demonstrating the unity and forward-looking nature of value guidance. In the backward alignment stage, the West is deeply mired in a vortex of “technological deviation” characterized by capital-dominated technological closure, self-referential monitoring mechanisms, and lagging social feedback responses. Chinese socialist ideology, by contrast, can continuously correct value deviations in AI applications through a “value verification” function. The root cause is that socialist ideology is a value-calibration system formed through the long-term evolution of social practice, possessing both historical necessity and practical adaptability. From a global perspective, the choice of value goals for socialist ideology has undergone a historical process that began with utopian socialism as its ideological starting point, took scientific socialism as its theoretical pillar, was tested by the practice of the world socialist movement, and was deeply explored by socialism with Chinese characteristics. From a Chinese perspective, socialist ideology has experienced the rigorous reasoning of “theoretical construction,” the historical磨合 of “class shaping,” the widespread dissemination of “social popularization,” and the epochal innovation of “reconstruction and sublimation.” This evolutionary process has forged its enduring calibration and testing functions of “value identification—value correction—value standardization—value guidance.” In summary, “value alignment” is a developmental process of constant game-playing between reality and ideals. It is precisely because socialist ideology meets the core standards of completeness of the theoretical system, representativeness of class interests, and social mass acceptance that it can ensure its authority and lasting influence when empowering the “value alignment” of AI.

(III) Socialist Ideology Drives Large AI Models toward the Ultimate Value Destination of “Machine for Humanity”

“The ultimate goal of AI development is to satisfy humanity’s pursuit of a better life” and to advance toward the ultimate liberation and emancipation of humanity. As Marx said, “Natural science, through industry, increasingly enters into human life, transforms human life, and prepares for human liberation.” However, in the current context of digital capitalism, the value of AI has undergone a certain degree of alienation. “The capitalist application of large digital-intelligent models supported by human-computer interaction makes the production of the entire society increasingly digital-intelligent. Its production model manifests as the ‘humanization’ of digital-intelligent systems and the ‘autonomization’ of digital-intelligent production. Capital’s sucking and squeezing of living labor becomes increasingly convenient, flexible, and precise.” Technology that was originally meant to work for human liberation has become a tool for the极致 discipline of the labor process and the “endless exploitation” of workers.

In response to this adverse trend, communist ideology becomes the ideological foundation and action guide for correcting the value alienation of AI. On the one hand, large AI models empowered by communist ideology will inevitably strive to reduce and even eliminate the constraints imposed on humanity by the division of labor, prompting humanity to establish a “free association of individuals.” Specifically, applying Marxist social division of labor theory can enable the global allocation and sharing of social resources produced by AI. Combining public ownership with large AI models can “form the covalent bond structure of humanity’s future life and destiny,” ensuring that the social成果 generated by technological progress can serve all people fairly and efficiently, thereby enhancing overall social welfare. On the other hand, large AI models empowered by communist ideology will inevitably strive to win more free time for human society and realize the free and comprehensive development of humanity. As Marx stated: “The whole development of humanity, insofar as it goes beyond the development directly required by humanity’s natural existence, is nothing but the use of this free time, and the premise of the whole development of humanity is to take this free time as a necessary foundation.” Currently, the construction of digital China is a vivid practice of “the free and comprehensive development of humanity.” It adheres to the core concept of “people-oriented” AI, actively builds a concept of time based on the measure of life, and, while upholding temporal justice, strives to promote the free and comprehensive development of humanity, providing a broad field for large AI models to achieve the ultimate goal of human liberation. In summary, “around the middle of the 21st century, there is a very good hope of ushering in another vigorous opportunity for world socialism,” and large AI models will inevitably become the engine driving human society to leap to a higher stage.

II. Practical Principles for Socialist Ideology to Empower the “Value Alignment” of Large AI Models

Empowering the “value alignment” of large AI models with socialist ideology is a long-term, systematic task. It must seek coordination among macro-level design, meso-level operation, and micro-level practice to promote the high-quality development of large AI models. In this context, highlighting the scientific methodology of socialist ideology—the principle of balancing innovation and development with safety and controllability, the principle of combining positive incentives with negative sanctions, and the principle of integrating industry self-governance with social co-governance—is key to advancing AI “value alignment.”

(I) Adhere to the Principle of Balancing Innovation and Development with Safety and Controllability

From the macro-development perspective of “value alignment,” adhering to the “principle of balancing innovation and development with safety and controllability” can enable large AI models to proceed steadily and far. Currently, “effective acceleration” prioritizes R&D efficiency and speed, seeking rapid commercialization of technology, but often neglects the ethical challenges arising from technology development and deployment, as well as its potential impact on social structures and values. “Superalignment,” on the other hand, emphasizes ethical and social effects, insisting that technology applications must precisely align with core social values to avoid risks. However, overly strict regulations may also hinder development and innovation; “some AI products can easily be subjected to strict regulation or even ‘shut down’ in the name of ensuring safety.” Viewed this way, “effective acceleration” and “superalignment” have created a practical contradiction, and this distorted development perspective can easily become a political tool for Western powers to serve their own private interests. Taking the Trump administration as an example, when technological development aligned with geopolitical interests, it turned a blind eye to potential ethical risks and even revoked executive orders imposing safety restrictions on the development and deployment of AI, blindly pursuing the “effective acceleration” representative of liberalism to seize the initiative. When it needed to suppress competitors or engage in political manipulation, it used “superalignment” as a tool, labeling competitors with the charge of “safety risks” and curbing their development by restricting technological cooperation and imposing tariffs. In fact, the “effective acceleration” and “superalignment” of large AI models are inherently intertwined. In the Chinese context, they should form a concept aimed at promoting the coordinated development of technological progress and social safety.

“Coordinating development and security” is a major strategic concept put forward by General Secretary Xi Jinping in grasping the domestic and international trends of change. It is deeply rooted in the fertile soil of Marxist dialectics and constitutes an indispensable principle of contemporary national governance in China. Therefore, in the process of socialist ideology empowering the “value alignment” of large AI models, it is essential to adhere to the “principle of giving equal weight to innovative development and secure controllability,” and to “grasp both the promotion of development and regulatory oversight with both hands, and make both hands strong.” This means avoiding both the stifling of technological innovation through unnecessary regulation and the danger of technology using its internal logic or objective attributes as a pretext to ignore and violate the bottom line of social values. A comprehensive, coordinated system of governance must be built to guide large models steadily forward through crises and challenges. Currently, China’s relevant laws and standards have already outlined a clear path for the innovative development of AI. For example, the *Interim Measures for the Management of Generative AI Services* state: “The innovative application of generative AI technology in various industries and fields is encouraged, generating positive, healthy, uplifting, and virtuous quality content, exploring and optimizing application scenarios, and building an application ecosystem.” At the same time, the *Basic Security Requirements for Generative AI Services in Cybersecurity Technology* stipulate rules regarding the sources of corpora, the content of corpora, and the security of corpus annotation that violate the core socialist values, requiring that “for obviously extreme content and content obviously inducing the generation of illegal or harmful information, the request should be refused.” In concrete practice, the large model DeepSeek, developed independently in China, serves as a typical example. It achieved technological innovation and catch-up through the “scaling law,” while also using a “four-stage training process of cold start – reinforcement learning (reasoning scenario) – SFT – reinforcement learning (full scenario) to align the model.” This reduces training costs while ensuring that the AI-generated content conforms to China’s mainstream values, guaranteeing the value security of the large model by giving equal weight to development and security, and providing a useful reference for the stable development of AI in China.

(2) Adhere to the Principle of Running Positive Incentives and Negative Sanctions in Parallel

From the meso-level operational perspective of “value alignment,” adhering to the “principle of running positive incentives and negative sanctions in parallel” is of crucial importance for driving the behavioral optimization and value shaping of large AI models. Currently, value integration strategies in the AI field generally tend to adopt a development paradigm that pursues “natural justice.” That is, researchers continuously push large models to deeply simulate human moral concepts and value systems, thereby attempting to let AI naturally form “human-like values.” As practice deepens, researchers have found that leaving large models unchecked can unexpectedly cultivate intelligent systems capable of evading rules and using non-moral means to achieve value goals. It can even lead to a phenomenon where machines engage in “deceptive alignment” to obtain human incentives and rewards, such as generating non-existent references to cater to human needs. This is the famous “WALL-E paradox” (where “WALL-E” refers to AI). “If ‘WALL-E’ still evolves according to the program instructions set by humans, it will forever remain merely a higher-order, more intelligent tool, and machine subjectivity will be out of the question; if ‘WALL-E’ transcends the logic of carbon-based life and becomes a completely new silicon-based species, human-machine alignment will become incalculable and unpredictable.”

Therefore, it is necessary to re-examine the “value alignment” mechanism of AI, deeply applying the guiding and critical functions of socialist ideology to ensure that it can both efficiently complete tasks and effectively prevent the occurrence of non-moral behavior. These two functions evolve in the practice of “value alignment” into the concrete principle of running positive incentives and negative sanctions in parallel. When implementing the principle of positive incentives, a rich set of value-oriented rewards can be introduced. These include giving additional rewards based on the model’s demonstrated moral sensitivity and compliance during the decision-making process, and positively reinforcing choices made by the model in complex situations that conform to the core socialist values through tests simulating moral dilemmas in human society. When implementing the principle of negative sanctions, it is necessary to “use mainstream value orientation to steer algorithms.” A refined violation identification and correction system can be established based on the mainstream ideology. Once a violation is detected, the sanction process is immediately initiated, urging relevant enterprises to temporarily restrict some functions of the model, adjust its training dataset to reduce the impact of harmful information, etc. General Secretary Xi Jinping pointed out: “The primary responsibility of internet enterprises must be consolidated, and the internet must never be allowed to become a platform for spreading harmful information and creating rumors.” China’s *Regulations on the Administration of Network Data Security* Article 40 has already imposed a certain degree of regulation on enterprise platform entities. In summary, by innovatively combining multi-dimensional rewards with refined sanctions, supplemented by a dynamic value calibration mechanism, it is possible to build an AI system that is both efficient and moral, bringing more profound impacts to human society.

(3) Adhere to the Principle of Running Industry Self-Governance and Social Co-Governance in Parallel

From the micro-level operational perspective of “value alignment,” adhering to the “principle of running industry self-governance and social co-governance in parallel” is the best criterion for improving the efficiency of aligning large AI models with core values. As large models become involved in diverse tasks, the complexity of value alignment work increases significantly, giving rise to an “alignment tax” – the additional costs incurred for value alignment, such as value assessment, ethical review, continuous learning, and adaptation. Faced with high “alignment taxes,” some AI enterprises choose to simplify or even abandon “value alignment.” In essence, relying solely on the knowledge structure and subjective judgment of personnel within the industry is insufficient to support comprehensive and in-depth value alignment work. Considering the public nature of AI, the practice of the industry bearing responsibility and risk alone has gradually lost its legitimacy.

During the 9th collective study session of the 18th CPC Central Committee Political Bureau, General Secretary Xi Jinping emphasized: “It is necessary to integrate multi-disciplinary forces, strengthen research on AI-related legal, ethical, and social issues, and establish and improve the laws, regulations, institutional systems, and ethics needed to ensure the healthy development of AI.” This means that the “value alignment” of large AI models requires more open-source entities. Cross-domain, cross-industry collaborative governance is an inevitable trend for achieving the goal of “value alignment.” Because socialist ideology possesses strong cohesive and guiding power, in the process of empowering large AI models to achieve “value alignment,” it can broadly absorb the reasonable demands and wisdom of different social groups, effectively reducing the industry’s own “alignment tax” costs, thereby ensuring that large AI models achieve sustainable development on a track of shared responsibility. For example, the State Council’s *Opinions on Deeply Implementing the “AI+” Action* stated the need to “strengthen government guidance and industry self-discipline, adhere to the principles of inclusiveness, prudence, classification, and grading, and accelerate the formation of a dynamic, agile, and multi-synergistic AI governance framework.” This provides a solid methodological guide for the “value alignment” of large AI models and clearly establishes the “principle of running industry self-governance and social co-governance in parallel” as a core operational principle that must be observed. Only by combining the industry’s conscious fulfillment of its primary responsibilities with the effective intervention of diverse social forces, “promoting the alignment and coordination of development strategies, governance rules, and technical standards among all parties,” and systematically integrating dispersed social knowledge, diverse value demands, and dynamic ethical considerations into the entire process of technological development, can it be ensured that the development direction of large AI models remains consistent with the overall interests and long-term well-being of human society while promoting technological innovation, ultimately achieving a harmonious unity of technological progress and social values.

III. Pathways for Advancing Socialist Ideology to Empower the “Value Alignment” of Large AI Models

Marx pointed out that “if in all ideology men and their relations appear upside-down as in a camera obscura, this phenomenon arises from the historical process of their lives,” which reveals the dialectical relationship between ideology as a system of ideas and social existence. Against the backdrop of the New Era, socialist ideology radiates rich ethical, emotional, cultural, and social values. Large-scale artificial intelligence models need to rely on these values to form mechanisms of ideological dissemination and identification, embedding them as new elements and new standards into the underlying logic of machine operation, thereby driving large-scale AI models to develop a new “spiritual ethos.”

(I) Using the Ethical Values of Socialist Ideology to Drive Large-Scale AI Models to "Possess Virtue"

Socialist ideology embodies an ethical core that conforms to the laws of social development and aligns with fairness and justice. It prominently demonstrates its role as a value-guiding system in enabling large models to adhere to bottom lines, distinguish right from wrong, and act for the good — this is the key to advancing scientific and technological practice. First, the ethical essence of socialist ideology must be deeply integrated into the design and construction of large-scale AI models. For example, the Ministry of Industry and Information Technology's "Draft Measures for the Management of AI Science and Technology Ethics" mentions principles such as respect for the right to life, respect for intellectual property rights, adherence to fairness and impartiality, maintenance of openness and transparency, ensuring controllability and trustworthiness, and strengthening responsibility and accountability. During the model training phase, a weighted loss function based on these ethical principles could be introduced, transforming the value goal of "ethical safety" into design standards such as "technical robustness" and "safe interruptibility," ensuring that the model automatically weighs and follows ethical principles during optimization. Second, AI researchers should receive ethics-oriented education, enabling them to deeply understand the ethical values required by socialist ideology. Through professional training, seminars, and other forms, the ethical literacy and sense of responsibility of researchers can be enhanced, internalizing ethical and moral norms in the mind and externalizing them in action, thereby proactively integrating ethics into model design and application, achieving a harmonious symbiosis between technology and ethics. Third, to ensure that large-scale AI models continuously align with the ethical values of socialist ideology, a comprehensive ethical evaluation system must be established. This system should include a professional ethics review committee mechanism, a third-party independent evaluation mechanism, and a social public supervision and feedback mechanism. By regularly monitoring and evaluating the ethical performance of models, their ethical settings can be promptly adjusted and optimized, thereby driving AI to continuously progress along the path of "possessing virtue."

(II) Using the Emotional Values of Socialist Ideology to Drive Large-Scale AI Models to "Possess Emotion"

Socialist ideology contains rich elements of familial-national sentiment and humanistic care, fully embodying emotional values that come from the people, serve the people, and benefit the people. "Meeting the spiritual and emotional needs of the people for a better life is the starting point and goal of mainstream ideological dissemination." Therefore, empowering large-scale AI models with the emotional values of socialist ideology means shaping a unique "emotional core" for AI, enabling them to display more vivid and lively traits in their interactions with the masses, adding the charm of "possessing emotion." Affective computing theory reveals that completing the replication and transmission of emotional values from mainstream ideology to AI and then to the masses requires following the logic of "emotional parsing — emotional mapping — emotional experience," constructing an emotional pathway of "moving people with emotion, touching people with emotion, and transforming people with emotion." "Emotional parsing" refers to the process of "quantitatively annotating" emotional vocabulary, carefully selecting content highly aligned with the emotional values of socialist ideology from vast media information. This stage involves analyzing both the emotional elements of national pride and social responsibility embedded in grand narratives such as national prosperity and national rejuvenation, as well as the emotional elements of a sense of gain and belonging embedded in micro-narratives such as daily necessities, thereby providing rich "emotional nourishment" for large models. "Emotional mapping" refers to the process of classifying and labeling the parsed emotional elements, then embedding them into the latent space representation layer of the large model, achieving precise projection from textual semantics to emotional vectors. This stage can employ "emotional enhancement technology," classifying emotional elements as positive or negative and assigning intensity coefficients, then aligning them with the existing vector space in the latent space, making the emotional values of socialist ideology closer to the emotional expectations of the masses. "Emotional experience" refers to the process by which the masses encounter and feel the emotional values of socialist ideology through large-scale AI models. This stage also requires designing more reasonable emotional progressive transmission pathways based on changes in the "emotional profiles" of the masses, and further sorting and optimizing the transmitted content based on human-computer interaction and feedback, to deepen the masses' understanding and identification with socialist ideology. In summary, integrating the emotional values of socialist ideology into large-scale AI models essentially endows them with a "heart" full of deep feeling and understanding of the people, driving them to become "emotional" entities at the level of emotional display that can perceive the warmth and cold of the people and transmit familial-national sentiment.

(I) Using the Ethical Values of Socialist Ideology to Drive Large-Scale AI Models to “Have Virtue”

With the rapid development of large-scale artificial intelligence models, their inherent tendency to pursue profit has triggered a series of problems contrary to the orientation of advanced socialist culture. In the process of cultural creation, some large models excessively pursue sensory stimulation, leading to frequent phenomena of shoddy and fabricated cultural products, with cultural garbage proliferating unchecked. Socialist ideology is “a collection of ideological integration, national cohesion, cultural centripetal force, and civilizational constraint generated by social subjects on the basis of core value identification,” and it contains rich cultural values. “Technological development cannot be separated from cultural nourishment. Without advanced culture, technological innovation and the effective cultivation of talent become like water without a source or a tree without roots.” [n] Therefore, integrating the cultural values of socialist ideology into large-scale AI models is an effective way to promote their “quality improvement.” From the perspective of cultural communication studies, the core of this process lies in effectively transmitting and embedding cultural memes representing socialist ideology into large-scale AI models, and successfully generating content with socialist cultural characteristics within them. Cultural memes “generally need to go through four communication processes: assimilation, memory, expression, and transmission.” However, because large-scale AI models are based on the continuity and creative output of cultural entities, they can transform culture originally carried by “inert carriers” into “active carriers,” achieving a “vivid transformation and migration” of cultural symbols and rhetorical devices. The mode of meme transmission is thus also innovated, transforming into a new paradigm of “meme extraction – fusion and reconstruction – innovation and regeneration.” “Meme extraction” refers to the process of deeply excavating core value concepts, behavioral patterns, and symbol systems from China’s mainstream culture, refining iconic memes representing the national image, key memes embodying national identity, and regional memes showcasing local human landscapes, thereby constructing a database of Chinese cultural memes aligned with socialist ideology. “Fusion and reconstruction” is the process by which large-scale AI models use cultural memes from the Chinese cultural gene database as templates, and through the supplementation, embellishment, and reconstruction of cultural imagery, transform polysemous cultural information into language, image, and video content. “Innovation and regeneration” refers to the process by which large-scale AI models, leveraging their own creative output capabilities, engage in a deep re-creation of cultural memes. With technological participation, large-scale AI models can use their advantage as a “world simulator” to promote the integration of socialist culture into cultural reproduction in digital form, and based on the mutual embedding of original and produced content, generate more new culture with “boundless vitality” characteristics. In summary, empowering large-scale AI models with the cultural values of socialist ideology means using the profound heritage of Chinese culture to inject distinct Chinese characteristics into artificial intelligence, further driving large models to create cultural content that embodies the essence of advanced socialist culture while being full of novel creativity and unique style, thereby leading large models toward higher quality development.

(IV) Promoting the “Order” of Large-Scale AI Models with the Social Values of Socialist Ideology

Engels pointed out that “the material mode of existence is the initial cause, but this does not preclude the ideological sphere from also reacting upon the material mode of existence.” Socialist ideology, as a “stabilizing device” for social development, plays important roles in constructing, maintaining, and criticizing various material products of practice, including large-scale AI models, thereby promoting their “orderly” development. First, the constructive value is manifested in the design and arrangement of the industrial system for large-scale AI models by socialist ideology. Specifically, this means leveraging the advantages of the new national system [n] to scientifically construct a positive structure between socialist ideology and the development of the AI industry at the top-level design, “further deepening reform comprehensively and building institutions and mechanisms conducive to technological innovation,” providing ideological support for the institutional adjustment, strategic planning, and future evolution of the AI industry, and “based on the full-cycle governance of industrial development, adopting layered, graded, and specialized governance for the base layer, professional layer, and application layer of large models,” promoting the safe and orderly development of AI. Second, the maintenance value is manifested in socialist ideology firmly, self-consciously, and confidently affirming the construction of digital civilization characterized by AI, and providing various conveniences for its further development. Socialist ideology should affirm and promote the excellent works, cases, and patents generated by large-scale AI models that align with mainstream values, to create broad reference and educational significance and foster a favorable socialist AI development atmosphere. For example, People’s Daily, as a mainstream media outlet, has integrated and connected with several domestic mainstream large models. Based on scientific collection, cleaning, annotation, and risk control, it has cooperatively created a mainstream value corpus exceeding 30 terabytes, safeguarding the continued development of large-scale AI models. Third, the critical value is manifested in socialist ideology resolutely criticizing and resisting unreasonable phenomena that appear in the process of AI industrialization. Regarding the potential for large-scale AI models to align with Western-centric development views, it is necessary to develop ideological resources from the character or realm of one’s own national culture to resist the penetration of Western ideology, continuously enhance the ability to expose and unmask the harmfulness of Western ideology, and “discover a new world through criticizing the old world,” [n] actively exploring the development path for China’s digital civilization, striving to build a technological civilization system deeply integrated with the core socialist values, and promoting the harmonious coexistence between technological progress and the people’s aspiration for a better life.

Conclusion

The Third Plenary Session of the 20th Central Committee of the Communist Party of China explicitly proposed to “establish an AI safety supervision system.” Empowering the “value alignment” of large-scale AI models with socialist ideology is precisely an effective measure to make large models safer and more orderly, and it “concerns the future of general-purpose AI development.” With the emergence of characteristics such as “generality, emergence, and multimodality” in large-scale AI models, “value alignment” has gradually become a regular and arduous task. On the one hand, although the digital rationality of large-scale AI models exhibits certain common features to some extent, there remains a significant asymmetry between it and the complex and diverse value systems of humanity. On the other hand, the value design of AI entities is not achieved overnight but is a continuous process of iteration and refinement, requiring multiple stages such as value expectation analysis, value embedding, and value evaluation, each of which demands careful design and strict control. However, these stages currently still face numerous technical and ethical challenges that require ongoing in-depth research. Looking ahead, research on “value alignment” needs to shift from theoretical space to technical operational space, and research methods need to delve deeper from ontology and epistemology to praxeology, thereby promoting the connotative development of the theory. In summary, empowering AI “value alignment” with socialist ideology requires sustained and persistent effort, promoting the harmonious coexistence of humans and AI, driving human-machine co-evolution, and achieving the digital-intellectual progress of civilization.