Eyes don’t lie: Indifferent to AI-generated depression screenings

This study explored how users respond to a depression self-diagnosis app with two outcomes: “doctor-generated” and “AI-generated.” The doctor-generated result was based on participants’ actual PHQ-9 depression scores, while the AI-generated result was randomly selected and sometimes matched the true diagnosis. Participants (n = 47) with depressive symptoms used the app while pupil dilation was recorded using a Tobii Pro Spectrum eye tracker. Pupil dilation was analyzed before and during the decision phase. Participants who chose the doctor-generated result showed no difference in pupil dilation compared to those who chose AI. Our Extreme Gradient Boosting (XGB) model predicted trust with an AUROC of 0.871. These findings show that users engage equally with both AI and doctor-labeled diagnoses, but pupil diameter is indicative of their trust., , # Eyes don’t lie: Indifferent to AI-generated depression screenings

Dataset DOI: [10.5061/dryad.80gb5mm2d](10.5061/dryad.80gb5mm2d)

## Description of the data and file structure

All files named P28.csv through P80.csv contain pupil diameter (mm) data for individual participants.

* Time\_Seconds is the time stamp during the experiment (ms)

* Pupil\_Filtered\_Smoothed is the preprocessed pupil diameter (mm)

The file Trust.csv is the label dataset, where 0 indicates no trust in AI and 1 indicates trust in AI.

### Files and variables

#### File: P28.csv

**Description:** 

##### Variables

* Time_Seconds: time stamp
* Pupil_Filtered_Smoothed: pupil diameter

#### File: P29.csv

**Description:** 

##### Variables

* Time_Seconds: time stamp
* Pupil_Filtered_Smoothed: pupil diameter

#### File: P31.csv

**Description:** 

##### Variables

* Time_Seconds: time stamp
* Pupil_Filtered_Smoothed: pupil diameter

#### File: P30.csv

**Description:** 

##### Variables

* Time_Seconds: time ...,