Finally, to prove our model insights could explain the observed dynamics, we developed an ABM reproducing key aspects of the behavioral data.

Simulations also provided generalizable insights beyond our experiment, revealing why collective intelligence can only emerge in certain environments.

We then used a comp model predicting movement decisions to test how private and social features guided participants' behavior.

Results confirmed that payoff information let participants selectively tune behavior to the position and direction of successful peers, unlocking collective intelligence!

Using high-resolution time-series data of participants' visual information and movement trajectories, we found that payoff information boosted performance by allowing collectives to flexibly reorganize visibility networks over time and adaptively guide information flow between group members.

Results show that sociality was a double-edged sword:

Groups could outperform solitary individuals through both superior tracking and search performance, but only when full payoff information was available.

In the absence of payoff information, they sometimes even performed worse!

We challenged individuals with a spatially-explicit search-and-tracking task in an immersive 3D environment.

By manipulating task complexity and the availability of social cues, we study how individuals adapt their visual attention and social learning strategies to different dynamic contexts.

Previous research (including my own) has predominantly studied these processes in simplified paradigms with unrealistic environments and predefined features!

Thus, it is largely unknown how collective intelligence can emerge in mobile human groups coping with dynamically changing environments.

Causal inference folks, can you please explain to me why conditioning on Z is bad here (case-control bias in journals.sagepub.com/doi/full/10....) I've simulated it, it's bad, I just don't get the intuition 🤔
@pwgtennant.bsky.social @dingdingpeng.the100.ci @rmcelreath.bsky.social @p-hunermund.com

In case you need another reason to apply:

There are eagle-owls nesting in the towers of Marburg's 13th-century Gothic church and you can observe (and hear) them almost every day at dawn 😱

🏰 PhD position on computational modelling | collective dynamics | cultural evolution 🏰

Get in touch with any questions you might have and join me in beautiful Marburg!

stellenangebote.uni-marburg.de/jobposting/c... (Deutsch)

stellenangebote.uni-marburg.de/jobposting/3... (English)

Very happy (and a bit nervous) to share that I've officially accepted a TT-professorship for Computational Modeling of Behaviour at the University of Marburg 🏰

Focus will be on (comp models of) collective decision-making and cultural evolution using both experimental and real-world data!

Part of an exciting special feature celebrating 50 years of quantitative cultural evolution!

Check out the rest here: bit.ly/CulturalEvolution

Finally, mapping group-level spatio-temporal dynamics revealed a collective explore-exploit trade-off across different timescales.

Group-level incentives reduced participants’ overall responsiveness to social information and promoted higher selectivity over time.

We then developed a computational Social Hidden Markov Decision model (inspired by animal movement models) to infer decisions between latent states and describe how different factors influence participants’ decisions to use social information.

Behavioral analyses revealed individual-level benefits (but group-level losses!) of high social information use and spatial proximity. Crucially, group incentives alleviated the negative consequences of excessive scrounging and improved performance

Using an immersive-reality approach, we study how groups of four participants search for resources ("coins”) in a 3D virtual environment with different resource distributions and incentive structures.

Now out in Nature Comms 🎊

Collective incentives reduce over-exploitation of social information in unconstrained human groups [link: rdcu.be/dCC3Z]

🧵below

How can we compare constructs ("prosociality", "extraversion", etc.) across groups?

How can we represent and test measurement invariance as part of modeling process?

We propose a Causal Framework for the Comparability of Latent Variables!

Link: osf.io/preprints/ps...