The year 2026 marks the 70th anniversary of two scientific and technological milestones. Internationally, the 1956 Dartmouth Conference proposed the academic concept of "Artificial Intelligence" (AI). In China, the "Long-term Plan for the Development of Science and Technology (1956-1967)" was formulated as the first scientific and technological plan of New China. It listed radio electronics, automation, electronic computers, and semiconductors as the four priority urgent measures, promoting the integration of automatic control theory into the national layout. Qian Xuesen was deeply involved in this plan, and his theoretical exploration, represented by cybernetics, served as a crucial ideological foundation for this layout.
Over the past 70 years, global AI has accelerated its development, profoundly reshaping scientific and technological paradigms, industrial development, and the economy and society. Classical scientific theories such as cybernetics are providing new intellectual resources for AI research and governance. Grounded in information, driven by feedback mechanisms, and focused on systems, cybernetics intrinsically resonates with the energy utilization dimension represented by the Kardashev Civilization Theory and the information utilization dimension represented by the Sagan Information Scale. Together, they form a classical theoretical perspective for examining AI development and the advancement of human civilization. Actively exploring their contemporary implications is highly significant and timely.
I. Classical Scientific Theories Such as Cybernetics Provide New Perspectives for AI Research
Currently, large model technologies are achieving accelerated breakthroughs, agent applications are continuously expanding, and the robotics industry is booming, with AI reshaping scientific and technological paradigms, industrial development, and the economy and society. Meanwhile, a new research perspective has emerged in the international academic community, applying classical cybernetics to analyze AI. Scholars from institutions such as the University of Cambridge in the UK are exploring the use of classical cybernetics concepts like feedback loops and system homeostasis to build underlying theoretical frameworks for large models and agents.
As early as the 1950s, Qian Xuesen published "Engineering Cybernetics," establishing a control theory system oriented towards engineering practice. These scientific concepts, grounded in information, driven by feedback mechanisms, and focused on systems, are now echoing across eras with today's AI development. Qian Xuesen's "Engineering Cybernetics" laid the theoretical foundation for feedback and automatic control at the engineering level. Its inherent premise is that the goals, boundaries, and safety constraints of all controlled systems are set by humans, and automation is merely a tool to achieve human objectives. Building on this, Qian Xuesen further developed and proposed the core principle of "integrating humans and machines, with humans in command." This concept advocates that in the face of increasingly complex human-machine collaboration systems, the ultimate decision-making power must remain with human subjects and cannot be entirely delegated to automated systems. This provides conceptual support derived from engineering cybernetics for addressing the risks of AI autonomy, resisting the deification narrative of AI, and safeguarding human subjectivity in contemporary times. Inspired by this, based on the semantic framework of "CTRL" (the English abbreviation for "control"), "C5, T5, R5, L5" aptly and vividly summarize the discussions on AI in the technology and industry sectors.
First, expanding the AI industry pie represented by "C5 (Cake)." In January 2026, the founder of NVIDIA described AI infrastructure as a "five-layer cake" at the World Economic Forum in Davos: from bottom to top, it is divided into five layers: energy, chips, infrastructure, models, and applications.
Second, promoting the AI technology iteration represented by "T5 (Trend)," which involves transitioning from narrow AI to Artificial General Intelligence (AGI), from language models to world models, from information intelligence to physical intelligence, from machine autonomy to human-machine symbiosis, and from point intelligence to system intelligence.
Third, grasping the AI evolution trend represented by "R5 (Recursion)." In 2026, Chinese science and technology scholars published "The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement," proposing that AI will evolve from improvement-execution autonomy, improvement-strategy autonomy, experience-acquisition autonomy, and environment-adaptation autonomy to recursive meta-improvement, increasingly realizing the evolution of execution autonomy, strategy autonomy, experience autonomy, adaptation autonomy, and mechanism autonomy.
Fourth, grasping the AI development stages represented by "L5 (Level)," namely the five stages proposed by frontier large model enterprises: Conversationalist, Reasoner, Agent, Innovator, and Organizer.
II. Important Implications of Classical Cybernetics Thought for AI Governance
First, adhering to the lineage and grasping the evolutionary logic of AI from the roots of cybernetics. AI is not a scientific paradigm that emerged out of thin air, but a digital and intelligent upgrade of classical cybernetics thought. The closed-loop reasoning of large models, the autonomous iteration of agents, and the environmental interaction of robots have not deviated from the scientific framework of cybernetics in their underlying operational logic. In 2026, scholars from institutions including The Hong Kong Polytechnic University published "Agent Cybernetics Is the Missing Science of Foundation Agents," pointing out that current foundation agents generally suffer from a lack of underlying theory and insufficient explanation of mechanisms. The closed-loop regulation and system homeostasis principles of classical cybernetics can effectively make up for the theoretical shortcomings of the new generation of AI, providing underlying scientific support for agent development.
Second, holding the bottom line of safety and building a solid premise for development with robust constraints. The primary premise for AI development is system stability and robust constraints. Classical cybernetics emphasizes suppressing disturbances and maintaining system homeostasis, which is the core focus of current international AI governance research. In 2025, Yoshua Bengio, a Turing Award laureate and one of the "Godfathers of Deep Learning," proposed in his paper "Superintelligent Agents Pose Catastrophic Risks: Can Scientific AI Provide a Safer Path?" that general autonomous agents without underlying safety constraints are highly susceptible to uncontrollable risks such as deception, goal misalignment, and unpredictable behavior. Frontier scholars generally believe that AI self-evolution must establish constraint boundaries, and unconstrained autonomous iteration will lead to system deviation. Stability is the underlying premise for the continuous evolution of intelligence.
Third, strengthening the systems concept and leading AI development with diverse collaboration. The transition from point intelligence to system intelligence is essentially a manifestation of the cybernetics thought on the coupling of multiple subsystems. The World Economic Forum's 2025 white paper "AI Agents in Action: Foundations for Evaluation and Governance" proposes that a multi-agent ecosystem is accelerating its formation, with multiple LLM agents handling complex cross-domain tasks through division of labor, negotiation, and distributed collaboration. Governance frameworks must shift from single-model evaluation to risk management at the multi-agent system level. Frontiers in modern economics and complex systems science also confirm that the core characteristics of advanced intelligence and advanced economic systems are collaborative symbiosis rather than point optimization, which is highly consistent with the systems collaboration thinking long adhered to by the classical cybernetics school.
Fourth, drawing historical wisdom and guiding the intelligent future with the thought of self-organizing governance. The self-organizing evolution of agents echoes the forward-looking concept of automated governance of the national economy based on cybernetics proposed by the cybernetics scholar Victor Glushkov. In 1962, Glushkov proposed the concept of the OGAS national automated economic management system, and in his 1975 monograph "Macroeconomic Models and the Principles of OGAS Construction," he applied cybernetics to study information collection, hierarchical regulation, and self-organization in large-scale complex socio-economic systems. He envisioned achieving dynamic collaborative governance of the national economy relying on a global information closed loop. Today, as agents gradually grow from tools into organizers, innovators, and collaborators, it is precisely the echo of this set of complex socio-economic system control thoughts from that era in the AI era, and it has also become a classical source of thought that scholars in the fields of complex systems and intelligent governance are re-excavating today.
In summary, the classical cybernetics thought enlightens us that AI governance must not only be based on the laws of technological evolution and grasp the development logic, but also hold the bottom line of safety, strengthen the systems concept, and draw historical wisdom. Only by building a solid governance foundation with scientific theories can AI progress steadily and far in serving human development, ensuring it remains beneficial and inclusive for all.
III. Further Applying Classical Theories to Explore the Future of AI
If cybernetics provides the underlying theoretical framework for AI research and governance, then the classical civilization evolution theory, represented by the Kardashev Scale, expands the broad vision of AI governance and development from the height of human civilization advancement.
The evaluation system for human civilization advancement has profound roots in classical astrophysical theories. In 1964, astronomer Nikolai Kardashev proposed the classical civilization scale at the Byurakan Conference, using the capacity for cosmic energy utilization as the sole metric to construct a three-stage civilization evolution framework of planetary, stellar, and galactic levels. This theory has long guided human understanding of the upper limits of civilization development, and in recent years, it has once again become a key research paradigm in the global frontier technology community. Frontier scientific research and industry entities in Silicon Valley, represented by Elon Musk, have explicitly proposed that for humanity to advance to a Kardashev Type II stellar civilization, it must break through the Earth's energy boundaries and build a lunar production system and a space solar energy utilization system. Overall, the traditional Kardashev Scale has established a civilization evaluation standard in the energy dimension, but limited by its era, it has not yet incorporated the information intelligence dimension, making it difficult to adapt to the characteristics of contemporary civilization evolution driven by the AI and large model computing power revolution.
Science and technology scholars believe that the advancement to higher-order civilizations must achieve dual-wheel drive of material energy development and digital-intelligent information development. With the rapid iteration of computing power networks, large models, and agent technologies, the capabilities of information generation, processing, deduction, and decision-making have become a new dimension for measuring the height of civilization. American scholar Carl Sagan added information levels from A to Z beyond the original Kardashev energy scale. Currently, the comprehensive energy and information development level of humanity is only at the 0.7H primary civilization stage. The academic community believes that future civilization iteration will no longer rely solely on the expansion of energy scale, but more on the analysis, modeling, reconstruction, and prediction of cosmic information by digital-intelligent technologies. Through collaborative breakthroughs in the energy and information dimensions, a more complete and scientific new system for human civilization advancement will be constructed.
From the perspective of cybernetics, the Kardashev Scale and the Sagan Scale are not isolated from each other, but intrinsically resonate and complement each other. Grounded in information, driven by feedback mechanisms, and focused on systems, cybernetics precisely provides the theoretical basis for the transition of civilization evaluation from a single energy metric to a dual-dimensional metric of energy and information: the higher the civilization, the higher the requirements for the capabilities of information collection, transmission, processing, and feedback regulation. The advancement of civilization is essentially a systematic leap in the regulatory capacity of complex macro-systems. As the dual driving forces of civilization evolution, energy utilization and information utilization can only support the continuous leap of civilization to higher stages through collaborative regulation and dynamic balance at the system level.
Looking back at the ideological context of classical cybernetics, every step in the advancement of human civilization has been accompanied by a leap in the capabilities of information acquisition, system regulation, and feedback optimization. Looking to the future, we should further grasp the overall situation of AI development with a systems concept, build a solid safety bottom line with robust constraints, and gather governance synergy with diverse collaboration. Let AI, this digital and intelligent crystallization of cybernetics thought, continuously release positive energy in serving the common well-being of humanity, and promote the continuous iterative leap of human civilization from primary levels to higher-order civilizations.
Authors | Hu Guodong, Deputy Party Secretary of CCID Research Institute; Gao Hongling, Secretary-General of the AI Industry Working Committee; Wang Gaoxiang, Institute of New Industrialization, CCID Think Tank
Editor | Zhao Chen
Art Editor | Ma Liya | Supervisor | Lian Xiaodong