Human-computer interactions predict mental health
Veith Weilnhammer, MD
Max Planck UCL Centre for Computational Psychiatry and Ageing Research
It started with strange noises through the wall, quiet and barely recognizable, as if someone were eavesdropping.
Over time, I could make out the voice of my neighbor. I had to investigate.
At first, I only heard his voice in my apartment, but later he followed me to other places.
Then it dawned on me - he's an agent!
Mrs. X
It started with strange noises through the wall, quiet and barely recognizable, as if someone were eavesdropping.
Over time, I could make out the voice of my neighbor. I had to investigate.
At first, I only heard his voice in my apartment, but later he followed me to other places.
Then it dawned on me - he's an agent!
Dr. Y
Clinical representations
Clinical representations
Binary Diagnosis
Clinical representation
Two problems
Precision: Limited mechanistic understanding & accuracy
Scalability: Duration of undiagnosed & untreated illness
Goal 1: A mechanistic understanding of subjective experience
Goal 2: New ways of measuring mental health
New ways of measuring mental health
Global average screen time: 6 hours and 34 minutes per day.
Human-computer interactions for almost half of waking hours.
Passively collected data that can be gathered in real-time, with high precision, and at no significant cost.
Can human-computer interactions predict mental health?
Datasets and Methods
Results
Two problems
Precision: Limited mechanistic understanding & accuracy
Scalability: Duration of undiagnosed & untreated illness
Current projects
-
Public mental health
- Massively scalable and dynamic mental health monitoring
-
AI Safety
- Tools against survey pollution
- Portable tokens of humanness
- Contextualizing chatbots for MH safety
-
Evaluation & Open Source
- Ethical & data protection challenges
- Transparency and interpretability of AI-supported models
Outcome: A new modality for fair, reliable, and scalable mental health assessment
Thanks for your attention!