机器伦理的哲学基础、基本特征与实现方法  

The Philosophical Basis,Fundamental Featuress and Implementation Methods of Machine Ethics

作　　者：廖备水[1] Liao Beishui

机构地区：[1]浙江大学哲学学院,杭州310058

出　　处：《中国社会科学》2024年第2期126-142,206,207,共19页Social Sciences in China

摘　　要：随着人工智能系统自主决策能力的不断提升,如何赋予智能体的决策以足够伦理考量成为一个备受关注的重要挑战。解决这一问题的关键途径是建立机器伦理,即将人类伦理价值观和道德规范嵌入人工智能系统,使其具备伦理对齐能力。机器伦理建立在人类伦理的基础上,但具有与人类伦理不同的基本特征。首先,当前的智能机器尚无现实主义意义上的主体性和体验性,表现为在伦理决策方面的弱主体性。其次,机器的决策反映的是受其行为影响的人类利益相关方的伦理考量,因此,机器的伦理决策需要在不同利益相关方的价值诉求之间取得平衡,即具备社会平衡性。再次,机器在伦理决策时容易受到文化因素的影响,应能反映文化差异性。最后,机器需要向人类主体解释伦理决策、理解情感表达,并进行责任归因,因此必须具备良好的人机交互性。With the constant improvement of the autonomous decision-making ability of artificial intelligence(AI)systems,the question of how to endow such systems"decision-making with sufficient ethical considerations has become an important challenge that has attracted widespread attention.The key approach to resolving this problem is to establish a machine ethics that embeds human ethical values and moral norms into AI systems,enabling them to possess a capacity for ethical alignment.Machine ethics is based on human ethics but its fundamental features are different.First,current AI machines lack agency and experience in the real sense;this is expressed as weak agency in ethical decision-making.Second,the machines'decision-making should reflect the ethical considerations of the human stakeholders affected by their actions,so their ethical decision-making needs to balance the values of different stakeholders,i.e.,they should have the capacity for societal balancing.Third,AI machines are easily affected by cultural factors in ethical decision-making,so should be able to reflect cultural differences.Finally,they need to explain their ethical decisions to human agents;understand emotional expression;and attribute responsibility.This requires good human-machine interaction.

关 键 词：伦理对齐 可解释性 大语言模型 知识表示与推理 

分 类 号：B842.1[哲学宗教—基础心理学]