Thank you so much for sharing!!

Read the full paper using this free access link here:

www.sciencedirect.com/science/arti...

Some to-dos for the field:
- identify amenable types of real-world data
- integrate them into data-collection pipelines alongside lab-based tasks
- assess links to mental health
- assess lab-based versus real-world convergent validity of parameters
- expand models to accommodate real-world data

…and challenges of this approach:
⚠️ Noisy data
⚠️ Analytical complexity
⚠️ Data collection burden
⚠️ Ethical considerations

It remains to be seen whether real-world data can better computational psychiatry.

Finally, we explore opportunities:
👍 Testing the generalizability of in-lab computational psychiatry findings to every-day life
👍 Incorporating linguistic behavior into cognitive models using LLMs (see figure)

We also provide a (non-exhaustive) taxonomy of cognitive processes paired with modeling frameworks and types of real-world data that could be used to probe them:

We review recent studies that probe cognition using real-world behavior, primarily in a reinforcement learning framework.

Some of these have uncovered novel links to mental health (e.g. linking depression to blunted reactivity to positive prediction errors, or higher sensitivity to social rewards)

By real-world data we mean any data that reflect a person’s every-day behavior. We distinguish 3 types:
- experience sampling data (active, self-report)
- passive sensing data (e.g. geolocation, physiology, social proximity)
- digital-behavior data (e.g. social media, texting, phone/app navigation)

Computational psychiatry often relies on behavior in cognitive tasks, which are simpler, less engaging and often less social than real-world environments.

By contrast, real-world data have intrinsic ecological validity and allow the continuous assessment of cognition and its variation over time.

Thanks so much Angela!

Thanks for sharing Ondrej!! 🙂