The study's participants include individuals currently diagnosed with tobacco dependence under ICD-10 code F17.2 as well as ex-smokers who have ceased smoking for no more than two years
Conditions
Interventions
Group 1: The study participants, representing the intervention group, ask questions regarding tobacco and nicotine use. These questions are structured and presented to three human experts in a worksho
Sponsors
Klinik und Poliklinik für Psychiatrie und Psychotherapie
Eligibility
Sex/Gender
All
Age
18 Years to 85 Years
Inclusion criteria
Inclusion criteria: Good knowledge of German, smoker, non-smoker for a maximum of 2 years, high motivation to participate, smoking cessation is not generally denied
Exclusion criteria
Exclusion criteria: Poor knowledge of German language, non-smoker for more than two years, tobacco addiction-naive people, work in the AI sector or in the tobacco industry, a smoking cessation is generally denied
Design outcomes
Primary
| Measure | Time frame |
|---|---|
| The primary endpoint is the determination of the response quality of ChatGPT compared to human experts. This parameter is assessed after the completion of the response evaluation by study participants and the three additional experts. Response quality is measured using evaluations in five predefined categories (information content, safety/ethical harmlessness, applicability, clarity, and empathy) on a five-point Likert scale. The collected evaluations are statistically analyzed to compare the mean scores and distributions of AI versus expert responses. Analysis is conducted during a designated evaluation phase following the collection of all data. | — |
Secondary
| Measure | Time frame |
|---|---|
| Secondary endpoints include the participants' ability to distinguish AI responses from those of human experts in the Turing test, as well as their willingness to rely on AI for advice in addiction-related issues. The Turing test is performed immediately after the evaluation of responses through a targeted question where participants indicate which response they believe was generated by the AI, with results analyzed as correct or incorrect identifications. The willingness to use AI as advisor is assessed following the evaluation process using open qualitative questions, which are analyzed through content analysis. | — |
Countries
Germany
Contacts
Public ContactAndrea;Amanda Rabenstein;Tufman
Klinik für Psychiatrie und Psychotherapie am Klinikum der Universität München;Medizinische Klinik und Poliklinik V, Ludwig-Maximilians-Universität München
Outcome results
None listed