Interactive Experiment Platform
Carrier is a browser-based platform for designing, deploying, and monitoring controlled studies of mixed-agent, multimodal group interaction. It enables researchers to run repeatable experiments involving human participants, AI assistants, and agents across text, audio, and video channels — without building a bespoke system for each study.
Background
Why Carrier?
Online interaction is being reshaped by two converging shifts. Conversational spaces increasingly mix humans with AI actors — overt assistants, covert bots, and humans who use AI to draft or optimise messages. At the same time, everyday communication is moving beyond text toward audio and video as synthetic media becomes cheaper to produce. These shifts complicate agency and authenticity judgments and widen the surface for persuasion and misinformation, especially in modalities that carry stronger credibility cues than text.
Empirical understanding has not kept pace, in part because many studies still rely on bespoke, text-centric systems that are difficult to reuse, hard to instrument consistently, and poorly suited to systematic manipulation of group composition, roles, and modality. Carrier addresses this gap by providing configurable interaction protocols, integrated surveys, and standardised logging of interaction traces — including AI prompts and outputs — in a single reusable platform.
Design Framework
Three Axes of Interaction
Carrier treats group interaction structure as an explicit experimental object. Each chamber (interaction episode) is specified along three orthogonal axes that can be manipulated independently or combined factorially.
Who Participates
Define the interaction ecology with three participant types.
- Human participants
- AI assistants (LLM-driven)
- Scripted bots (rule-based)
What They Do
Separate participant type from functional role within the group.
- Communicator — primary interactant
- Mediator — group-facing broadcasts
- Processor — draft-time assistant
How They Interact
Configure the communication channel per chamber.
- Text chat
- Audio messaging
- Video conferencing
Capabilities
What Carrier Enables
Carrier reduces the engineering burden that typically leads to one-off implementations, enabling clearer replication and cross-study comparison.
Configurable Protocols
Define group assignment rules, session timing, matching policies, and dropout handling.
Flexible AI Roles
Switch AI between visible peer, mediator, or behind-the-scenes drafter across conditions.
Integrated Surveys
Attach surveys globally and per chamber, with responses driving dynamic matching criteria.
Standardised Logging
Capture interaction events, role states, and AI prompts/outputs in a consistent schema.
Live Monitoring
Real-time dashboard for participant status, matching queues, and experimenter interventions.
Session Continuity
Automatic reconnection support so participants can resume without losing progress.
Applications
Research Directions
Carrier supports a range of experimental paradigms across the spectrum of mixed-agent interaction research.
Human–AI interaction studies examining trust, persuasion, and communicative norms when people engage with AI agents
AI-mediated communication experiments where LLM facilitators moderate group discussions or reduce polarisation
Human–AI collaboration tasks exploring co-constructive dialogue, empathic response, and creative co-creation
Multi-agent simulations modelling social dynamics, cooperation, and emergent coordination in configurable group topologies
Multimodal credibility studies comparing how people interpret and respond to identical content across text, audio, and video