Proactive AI-Initiated Video Calling for General-Purpose Human–AI Interaction: A Unified Architecture for Autonomous Assistance Initiation, Context-Continuous Communication, and Agentic Task Execution
Year: 2026
Type: preprint
Abstract: Contemporary artificial intelligence systems predominantly rely on user-initiated interaction: users must recognize a need, open an AI application, formulate a request, and initiate a conversation before receiving assistance. Although recent advances in large language models, multimodal models, and agentic systems have substantially expanded AI capabilities, the interaction paradigm remains largely reactive. This paper proposes Proactive AI-Initiated Video Calling for General-Purpose Human–AI Interaction (PAIVC-GPA), a conceptual architecture in which a general-purpose AI system autonomously identifies potential assistance opportunities, determines whether proactive communication is appropriate, and initiates a video-call invitation to the user. Such invitations may be delivered through the AI system's own application or through other supported applications and communication platforms. Following explicit user acceptance, the video-call session becomes a unified real-time interface through which the user can access the AI system's general-purpose capabilities and communicate naturally with the AI through speech. Within the established interaction, users may also switch between video, text-based conversation, and voice-only conversation according to their preferences and task requirements. The proposed framework is distinguished by an integrated interaction sequence comprising autonomous assistance-opportunity detection, proactive communication decision-making, AI-initiated video-call invitation, cross-platform communication delivery, explicit user-controlled session establishment, contextual continuity between pre-call reasoning and the accepted session, flexible interaction-modality switching, and general-purpose agentic task execution. Unlike systems that merely embed an assistant into an existing video meeting, the proposed architecture places the AI itself in the position of initiating the communication. Unlike conventional proactive notification systems, it uses an accepted video-call session as a general-purpose AI interaction environment rather than restricting the interaction to a predefined notification or service. This paper formalizes the architecture, defines its operational states and decision mechanisms, and proposes an evaluation methodology for investigating its effectiveness. Potential applications include research, programming, education, document processing, information retrieval, data analysis, visual guidance, and authorized multi-step task execution. The paper also examines interruption costs, consent, privacy, contextual data governance, cross-platform communication, modality switching, and the technical requirements of persistent agentic systems. The author claims priority for the specific conceptual formulation of an AI system that autonomously initiates a video-call invitation to a user and transforms the accepted session into a context-continuous interface for full-spectrum general-purpose AI assistance. This claim is explicitly distinguished from earlier work on proactive digital assistants, opportunistic user-to-user calls, and digital assistants operating within established video communication sessions. The proposed framework is a conceptual and technical research contribution; no experimental superiority or completed implementation is claimed.
Author: Zhang, Jincheng
Language: English
Topic: Speech and dialogue systems
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225595539