OLN | Cloned Testing: An Investigation of Trust and Suspicion in Identity Recognition During Human-Machine Dialogue
Project information
submitted by
Marcel Gohsen
Co-Authors
Shriya Singh
Mentors
Marcel Gohsen, Benno Stein
Faculty:
Media
Degree programme:
Human-Computer Interaction (Master of Science (M.Sc.))
Type of project presentation
Final project
Semester
Summer semester 2026
- Bauhausstraße 11
(Webis-Labor (R012))
10. Juli 2026, ab 17:00 Uhr
Project description
In recent years, large language models (LLMs) have developed the ability to engage in social interactions that are virtually indistinguishable from human interactions. For example, OpenAI's flagship models have passed the so-called Turing test, in which human test subjects were unable to distinguish generated text from text written by humans. The social capabilities of LLMs not only offer advantages, but can also be exploited to make phishing attacks on social media even more convincing. The effectiveness of this attack depends on whether the victim can recognize the received messages as machine-generated.
This thesis examines how people behave in situations where it is unclear whether they are engaging in a chat with a human or with a clone of that person. In particular, a study will investigate how and why trust in the dialogue partner is built up or eroded, and what strategies are used to identify the dialogue partner as a machine.