ISSR Humanlike AI Systems and Trust Attribution

As AI assistants adopt humanlike names, conversational tone, and confidence framing, users form impressions of AI competence from interface design alone — independently of actual system capability. This can produce overtrust, underuse, or miscalibrated reliance. Most existing research measures trust through self-report scales, which cannot capture when or how quickly users actually defer to or override an AI recommendation. This project builds a modular, open-source web-based experimentation platform that makes behavioral trust measures tractable for researchers. Built in Next.js, the system supports randomized condition assignment, configurable manipulation of humanlike interface cues (agent name, tone, confidence framing), and a recommendation-acceptance task that produces clear reliance-versus-override outcomes under controlled AI accuracy. Fine-grained event logs — decisions, timestamps, response latency — are exported as clean JSON and CSV datasets, and a reproducible Python notebook reports reliance rate, override rate, and latency by condition. The platform is designed as reusable research infrastructure: cue logic is configuration-driven, task logic is decoupled from condition management, and the event schema is fully documented so future researchers can extend the system with new cues, tasks, and study conditions. Deliverables include a functional experimental prototype, cue manipulation framework, behavioral logging backend, sample dataset, analysis notebook, and complete documentation in a public GitHub repository.