Beyond the Conversational Paradigm: Models, Design Principles, and Tools for Advancing the Spectrum of Human-LLM Interaction

Overview

This research examines the design space of human-LLM interaction interfaces as a continuum between conversational user interfaces and graphical user interfaces with direct-manipulation capabilities. The work challenges the assumption that LLM interaction must be strictly conversational, investigating how systems can integrate visual and interactive elements to create hybrid interface configurations. The research employs Human-Centered Design methodology to understand user preferences, interaction patterns, and contextual factors that influence interface design decisions across this spectrum.

Methods and approach

The doctoral research employs a research-through-design methodology grounded in Human-Centered Design principles. The approach involves the development of representative interface prototypes distributed across the conversational-to-graphical spectrum, followed by comparative empirical studies to evaluate user preferences and behaviors. These studies systematically investigate factors that influence users' interface configuration preferences, examining how different contextual and individual variables affect interaction choices. The research iteratively refines understanding through prototype evaluation and user feedback cycles.

Key Findings

The research identifies that human-LLM interaction exists on a continuous spectrum rather than being constrained to purely conversational paradigms. Empirical findings from comparative studies reveal that user preferences for specific interface configurations vary based on task characteristics, user expertise, and contextual factors. The work identifies distinct regions along the spectrum that present different opportunities and constraints, with varying implications for user experience, task completion efficiency, and interface effectiveness.

Implications

The findings support the development of adaptive LLM interfaces capable of dynamically adjusting the balance between conversational and graphical interaction elements in response to evolving user needs and contextual requirements. This adaptive approach contrasts with static interface designs and enables interfaces that adjust their presentation across the CUI-GUI spectrum. The research establishes a foundation for moving beyond one-size-fits-all interface design paradigms in LLM systems.

The work contributes a comprehensive characterization of the interaction spectrum alongside empirical evidence of design trade-offs and risk-benefit analyses for different interface regions. Design principles and practical tools emerging from this research provide guidance for practitioners developing next-generation LLM interfaces. These principles inform decisions regarding when conversational interaction is optimal, when graphical direct manipulation enhances usability, and how hybrid configurations can be leveraged to support diverse user populations and task requirements.