从GPT迈向AGI:人工智能治理的未来路径探索

Moving from GPT to AGI:Exploring the Future Pathways of Artificial Intelligence Governance

[研究目的]GPT(生成式预训练模型)已经展现出未来AGI(通用人工智能)的技术端倪,AGI将成为未来社会的基础设施,应用于各行各业。我国目前已经形成了“算法—深度合成—生成式人工智能”的治理推进脉络,但从GPT迈向AGI的进程中,数据资源匮乏、算力供给短缺、模型开源建设落后、责任承担不明确等问题仍是其发展的掣肘,因此亟待探索适应人工智能时代发展的治理路径。[研究方法]立足于国内人工智能发展与治理的现状,深入分析目前人工智能发展所面临的问题,结合美国的灵活监管和欧盟的统一治理等国外人工智能治理经验,从逻辑更新和政策优化两方面破解人工智能目前面临的发展困境。[研究结论]在逻辑更新方面,我国应在明晰AGI为可自主无限处理复杂任务的通用性人工智能系统的情况下,制定面向AGI的包容审慎人工智能治理新模式,重视伦理建设从而规避安全风险。在政策完善方面,应鼓励数据共享流通,加大对算力基础设施的投入,积极推动开源建设,完善人工智能生成物的责任承担制度,以期促进我国人工智能的健康发展。

[Research purpose]GPT(Generative Pre-training Model)has already shown the technology prototype of AGI(General Artificial Intelligence)in the future,and AGI will become the infrastructure of the future society,and be applied to all wakls of life.China has already formed the governance of“algorithm-deep synthesis-generative artificial intelligence”,but in the process of moving from GPT to AGI,the lack of data resources,shortage o farithmetic supply,backward model opensource construction,and unclear responsibility are still constraints to its development.There is an urgent need to explore the governance pathways adapted to the development of AI era.[Research method]Based on the status quo of AI development and governance in China,we analyze in depth the problems it faced,and combine foreign AI governance experiences such as the flexible regulation of the United States and the unified governance of the European Union,so as to crack the development dilemmas faced by AI in terms of logic updating and policy optimization.[Research conclusion]In terms of logical updating,China should formulate a new model of inclusive and prudent AI governance for AGI and pay attention to ethical construction so as to avoid safety risks,while making it clear that AGI is a general-purpose AI system that can autonomously and infinitely handle complex tasks.In terms of policy improvement,we should encourage data sharing and circulation,increase investment in computing power infrastructure,actively promote open-source construction,and improve the system of assuming responsibility for AI-generated objects,with a view to promoting the healthy development of AI in China.