Neural Approaches to Conversational AI

Paper · arXiv 1809.08267 · Published September 21, 2018
Conversation Architecture and Structure

The present paper surveys neural approaches to conversational AI that have been developed in the last few years. We group conversational systems into three categories: (1) question answering agents, (2) task-oriented dialogue agents, and (3) chatbots. For each category, we present a review of state-of-the-art neural approaches, draw the connection between them and traditional approaches, and discuss the progress that has been made and challenges still being faced, using specific systems and models as case studies.1

Introduction. Developing an intelligent dialogue system1 that not only emulates human conversation, but also answers questions on topics ranging from latest news about a movie star to Einstein’s theory of relativity, and fulfills complex tasks such as travel planning, has been one of the longest running goals in AI. The goal has remained elusive until recently. We are now observing promising results both in academia sindustry, as large amounts of conversational data become available for training, and the breakthroughs in deep learning (DL) and reinforcement learning (RL) are applied to conversational AI. Conversational AI is fundamental to natural user interfaces. It is a rapidly growing field, attracting many researchers in the Natural Language Processing (NLP), Information Retrieval (IR) and Machine Learning (ML) communities. For example, SIGIR 2018 has created a new track of Artificial Intelligence, Semantics, and Dialog to bridge research in AI and IR, especially targeting Question Answering (QA), deep semantics and dialogue with intelligent agents.

Discussion / Conclusion. Conversational AI is a rapidly growing field. This paper surveys neural approaches that were recently developed. Some of them have already been widely used in commercial systems. We have discussed some of the main challenges in conversational AI, common to Question Answering agents, task-oriented dialogue bots and chatbots.

Synthesis notes from this paper's topics 8

This paper has no direct semantic matches yet — here are notes from the topic(s) it's filed under.

Lines of inquiry this paper opens 24

Research framings built by reading the notes related to this paper — the questions it feeds into.

What conditions enable agent collusion in verification systems? Can conversational agents overcome structural limitations and initiate proactive dialogue? How should LLM systems manage conversation history and dialogue memory? What coordination failures emerge when multiple language models work together? Does encoded knowledge in language models causally influence their outputs? Why do language models accommodate false information despite possessing correct knowledge? Does training for better reasoning reduce the ability to abstain? How can we understand and mitigate AI persuasion risks? Can prompting alone inject genuinely new knowledge into trained models? When should AI systems actively intervene versus remain silent? Can readers reliably distinguish AI-generated text from human writing? How do model confidence signals affect human judgment and error detection? What determines whether language models adopt prompted personas successfully? How can AI assistance improve human judgment without creating overreliance? What makes understanding genuine rather than merely producing correct outputs? Do models conceal internal reasoning that diverges from verbalized outputs? Why does model self-revision often amplify confidence while degrading accuracy?