关键词: AI安全, 控制范式, 共生范式, 人机协作, 数字治理

Abstract: Artificial intelligence technology is currently developing at an unprecedented pace, with safety concerns becoming a global focal point. Traditional AI safety research has predominantly relied on a “control paradigm”, emphasizing limitations, regulations, and value alignment to control AI behavior and prevent potential risks. However, as AI capabilities continue to strengthen, unidirectional control strategies are revealing increasingly significant limitations, with issues such as transparency illusions, adversarial evolution, and innovation suppression gradually emerging. Industry leaders like Sam Altman and Dario Amodei predict that AI may comprehensively surpass human capabilities in multiple fields within the next 23 years, making the reconstruction of AI governance paradigms particularly urgent. This paper proposes a new perspective—the “symbiotic paradigm”—emphasizing humanmachine collaboration as the core and understanding and trust as the foundation. Through establishing four pillars: transparent communication, bidirectional understanding, creative resonance, and dynamic boundaries, it promotes AI safety’s transition from control to cocreation, serving as one of the foundational paths for digital governance transformation. This paper systematically demonstrates the feasibility and necessity of the symbiotic paradigm through four dimensions: theoretical analysis, technological paths, practical cases, and governance recommendations, aiming to provide a sustainable alternative for future AI safety research and digital governance practices.

Key words: AI safety, control paradigm, symbiotic paradigm, humanmachine collaboration, digital governance