具备内在思考的主动对话智能体
摘要
对话 AI 的一个长期目标是允许它们在对话中自主采取主动措施，即具有主动性。这在多方对话中尤其具有挑战性。以往的 NLP 研究主要关注从上下文（如前导对话）中预测下一个说话者。本文演示了此类方法的局限性，并重新思考 AI 在多方、人类-AI 对话中具有主动性的含义。我们提议，正如人类一样，而不仅仅是对 turn-taking cues 作出反应，主动 AI 在对话期间会 formulation its own inner thoughts，并 seeks the right moment to contribute.通过 formative study with 24 参与者 and inspiration from linguistics and cognitive psychology, we introduce the Inner Thoughts framework.我们的框架为 AI 提供了一个与 overt communication process 并行的 continuous, covert train of thoughts, which enables it to proactively engage by modeling its intrinsic motivation to express these thoughts.我们将该框架实现为两个真实系统：一个 AI 玩具网站应用和一个聊天机器人。通过 technical evaluation and user studies with human participants, our framework significantly surpasses existing baselines on aspects like anthropomorphism, coherence, intelligence, and turn-taking appropriateness.
引用
@article{arxiv.2501.00383,
title = {Proactive Conversational Agents with Inner Thoughts},
author = {Xingyu Bruce Liu and Shitao Fang and Weiyan Shi and Chien-Sheng Wu and Takeo Igarashi and Xiang Anthony Chen},
journal= {arXiv preprint arXiv:2501.00383},
year = {2025}
}