全国政协委员贺晗：建议构筑人工智能时代核心竞争力

Core Viewpoint - The article emphasizes the need for China to strengthen its foundational capabilities in artificial intelligence (AI) to build core competitiveness and avoid "dimensionality reduction attacks" in the face of global technological competition [2]. Group 1: Strengthening Original Capabilities - China has made significant progress in AI application and scenario empowerment, achieving a transformation from "1 to 100" in recent years [2]. - The U.S. maintains an absolute advantage in AI due to continuous breakthroughs in foundational architectures, while China's reliance on overseas open-source ecosystems poses risks [4]. - The domestic AI hardware landscape is fragmented, lacking a unified software stack, which increases development costs and reduces efficiency [4]. - Recommendations include reforming research evaluation mechanisms, fostering "patient capital," and creating a unified domestic computing software ecosystem [5]. Group 2: Building a Self-Controlled Intelligent Ecosystem - Domestic foundational models primarily focus on natural language processing and lack capabilities for complex reasoning and decision-making, leading to inefficiencies in real-world applications [7]. - The current ecosystem is fragmented, with a lack of standardized interfaces and components, resulting in low replicability and high migration costs [7]. - There is a need for a strategic design to encourage the establishment of a "trusted intelligent agent operating base" and to promote standardization in tool interfaces [8]. Group 3: Accelerating Embodied Intelligence Development - Embodied intelligence requires high-quality, task-level interaction data, but the current data collection landscape is fragmented, leading to "data islands" [10]. - Many companies still use customized algorithms for specific tasks, lacking generalizable foundational models, which hampers performance in real-world scenarios [10]. - Recommendations include establishing a national "embodied data factor project" to create a public data foundation and promote the development of foundational models [11]. Group 4: Activating Computing Power - There are significant challenges in computing power distribution between eastern and western regions of China, leading to inefficiencies in data transmission and operational costs [13]. - The demand for electricity in computing centers is projected to grow exponentially, necessitating a coordinated development system for computing and electricity [14]. - Suggestions include designating central provinces as national computing hubs and constructing direct connections between computing nodes to enhance efficiency [14].