AI对齐的关键阻滞、路线探新及重要意义——基于迈向通用人工智能的现实呼唤  

Key obstacles in AI alignment,exploration of new approaches,and their significance:Towards the realistic call for general artificial intelligence

作　　者：王明 WANG Ming(School of Marxism,Dongbei University of Finance and Economics,Dalian 116025,China)

机构地区：[1]东北财经大学马克思主义学院,辽宁大连116025

出　　处：《重庆邮电大学学报(社会科学版)》2025年第2期104-115,共12页Journal of Chongqing University of Posts and Telecommunications(Social Science Edition)

基　　金：国家社科基金项目(中央马工程委托项目):我国科技创新体制机制研究(2020MYB038)。

摘　　要：从ChatGPT到DeepSeek,生成式人工智能的智能化程度不断提升,为人类社会创造福祉的巨大潜力也逐渐凸显。生成式人工智能技术所创造的巨大商业价值已经促使人工智能领域的巨头快速推动这项技术的迭代,从而使生成式人工智能迈向通用人工智能的路径变得愈发清晰。然而,生成式人工智能技术让人类喜忧参半的特性,决定了人类在面对未来可能出现的通用人工智能技术时必须保持谨慎态度。为确保其设计和应用符合社会期望及人类利益,需要建立一种伦理基础:AI对齐。目前的关键问题是,AI对齐需要整合学界对AI认识的理论分歧,攻克机器学习的风险性和人自身的脆弱性等现实挑战。当下人工智能业界和学界已初步形成AI对齐的目标和方法,但这些方法尚无法满足人工智能技术从生成式迈向通用式的现实需求。基于此,亟需探索AI对齐的新路线:政府应加快推进AI对齐的法律和技术伦理规范出台,人工智能企业要强化多样性和包容性的数据训练,多领域专家要加强对话与合作,用户要积极反馈AI系统使用中出现的问题。鉴于人工智能技术从生成式迈向通用智能技术的现实需求,AI对齐对于实现安全有益的通用人工智能意义重大。通过AI对齐,能够避免人工智能出现价值观偏差,防止其对人类产生不当影响,切实保护人类利益。From ChatGPT to DeepSeek,the intelligence of generative AI technology continues to expand,and its immense potential to generate benefits for human society is becoming increasingly evident.Concurrently,the substantial commercial value generated by generative AI technology has prompted leading AI companies to accelerate the iterative development of this technology,thereby advancing generative AI towards the trajectory of general artificial intelligence.The potential benefits and risks associated with generative AI technology have prompted a cautious approach to the development of general-purpose AI technologies.To ensure the design and application of these technologies align with societal expectations and the interests of humans,an ethical foundation,known as AI Alignment,is essential.The current key issues involve integrating the theoretical fragmentation of academic understanding of AI,overcoming the risks associated with machine learning,and addressing the real challenges of human vulnerability.The scientific and academic communities have preliminarily formed objectives and methods for AI alignment,yet these methods are still insufficient to meet the practical needs of AI technology transitioning from generative to general capabilities.Based on this,there is an urgent need to explore new approaches to AI alignment:governments should formulate legal and ethical standards for AI alignment,AI enterprises should enhance diversity and inclusiveness in data training,multidisciplinary experts should strengthen dialogue and collaboration,and users should actively participate and provide feedback in the design and decision-making processes of AI systems.Given the realistic call for AI technology to transition from generative to general intelligence,AI alignment is of significant importance for achieving safe and beneficial general AI.This primarily includes avoiding value biases,preventing undue influence,and protecting human interests.

关 键 词：AI对齐 ChatGPT DeepSeek 大型语言模型 通用人工智能 人类安全 

分 类 号：B0-0[哲学宗教—马克思主义哲学]