Ramezani Aida, Xu Yang
Department of Computer Science, University of Toronto.
Cognitive Science Program, University of Toronto.
Top Cogn Sci. 2025 Jan;17(1):120-138. doi: 10.1111/tops.12774. Epub 2024 Nov 25.
Automated moral inference is an emerging topic of critical importance in artificial intelligence. The contemporary approach typically relies on language models to infer moral relevance or moral properties of a concept. This approach demands complex parameterization and costly computation, and it tends to disconnect with existing psychological accounts of moralization. We present a simple cognitive model for moral inference, Moral Association Graph (MAG), inspired by psychological work on moralization. Our model builds on word association network for inferring moral relevance and draws on rich psychological data. We demonstrate that MAG performs competitively to state-of-the-art language models when evaluated against a comprehensive set of data for automated inference of moral norms and moral judgment of concepts, and in-context moral inference. We also show that our model yields interpretable outputs and is applicable to informing short-term moral change.
自动化道德推理是人工智能中一个新兴的极其重要的话题。当代方法通常依赖语言模型来推断一个概念的道德相关性或道德属性。这种方法需要复杂的参数设置和高昂的计算成本，并且往往与现有的道德形成心理学解释脱节。我们提出了一种用于道德推理的简单认知模型——道德关联图（MAG），它受到道德形成心理学研究的启发。我们的模型基于用于推断道德相关性的词关联网络构建，并借鉴了丰富的心理学数据。我们证明，在针对用于自动推断道德规范、概念的道德判断以及情境道德推理的综合数据集进行评估时，MAG的表现与当前最先进的语言模型不相上下。我们还表明，我们的模型能够产生可解释的输出，并且适用于为短期道德变化提供信息。
