Shriya Singh
Master thesis, Human-AI interaction2025 to 2026

Clone Testing

When does a conversation begin to feel artificial?

Clone Testing project cover

A two-phase investigation

Trainers first role-played personas that informed editable conversational profiles. Evaluators later spoke with a human trainer and an associated AI clone while recording how their suspicion changed throughout each conversation.

The study focused on how judgments formed and shifted, including moments when participants reversed an earlier assumption.

Participant using the Clone Testing setup
A participant during the study
Talk with a mysterious partner study sign
The study invitation and physical setting
Two phase study design diagram
Trainer and evaluator phases
Suspicion trajectories from the study
Suspicion trajectories across conversations

Try the study prototype

The deployed prototype preserves the interactive system used for the two-phase study. It opens in a separate site.

ACCESS PASSWORDadmin-password
Open the live prototype

What the study surfaced

Evaluators correctly classified 19 of 20 conversations in this small study. Repeatedly polished responses and weak contextual adaptation appeared as recurring clues. These results describe this study and should not be treated as a general AI detection rate.

Read the public thesis