The article explores the challenges faced by Conversational Agents (CA) and suggests improvements through cognitively inspired components. The authors note that while large language models like GPT3 have enhanced the quality of CA, technical issues and the human perception of CAs as social actors create problems. Technical problems include the limited scope of retrieval agents and the nonsensical responses of generative agents. The social problems arise from the failure of CAs to meet social expectations, which can lead to poor interaction or perceived threat from the user. The paper proposes that incorporating computational facsimiles of semantic and episodic memory, emotion, working memory, and the ability to learn can help to address these issues.
Publication date: 10 Nov 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2311.05450
