iiinova — where your digital self keeps evolving

Reposted by Valerii Chirkov

Félicie Dhellemmes @felicie-dhellemmes.bsky.social · 25/09/2026

Proud to present our latest work "Tracking human foragers and their prey reveals adaptive predator-prey dynamics", now hosted on BioRxiv! 🧵 doi.org/10.64898/202... @ralfkurvers.bsky.social @scioi.bsky.social @arc-mpib.bsky.social @alexschakowski.bsky.social @dominikdeffner.bsky.social

26427

Valerii Chirkov @chirkovv.bsky.social · 24/09/2026

Huge thanks to co-authors @ralfkurvers.bsky.social, @promanczuk.bsky.social, & @dominikdeffner.bsky.social (@scioi.bsky.social & @mpib-berlin.bsky.social) 📄 Main paper: www.nature.com/articles/s41... 📄 Related computational study: doi.org/10.1098/rsos...

011

Valerii Chirkov @chirkovv.bsky.social · 24/09/2026

Key Finding 3: Our agent-based simulations reproduced these behavioral dynamics. They illustrate that collective intelligence doesn't just happen by default—it only emerges under specific environmental conditions.

A Model illustration: At each time step, agents rotate based on either private information (with probability 1 − psocial) or social information (with probability psocial). Probability of acting based on social information (psocial) depends on social information quality (qt). If agents followed social information, they rotated towards the center of the locations of other agents in their field of view. If they followed private information, the rotation magnitude was proportional to the change in detector value (δt) and the rotation direction depended on the outcome of the previous rotation. Bottom right: Agents' selectivity to social information quality; colored lines represent degrees of selectivity. Highly selective agents (yellow) only followed social information if agents in their field of view were performing better than they were at the time; non-selective agents (blue) followed others independently of quality. B Average tracking time vs. tracking efficiency of groups relative to solitary agents (black dot). Colors show selectivity levels; opaque points correspond to the two experimental speeds (40% and 80%). Data points in B represent the average of n = 2000 independent simulation runs per parameter combination.

150

Valerii Chirkov @chirkovv.bsky.social · 24/09/2026

Key Finding 2: Visibility networks & modeling showed that payoff selectivity is the crucial ingredient. People selectively tune their behavior toward the position and movement direction of their most successful peers. Just seeing others isn't enough!

Top: Illustration of a resource discovery event, where a discoverer (brown) enters the resource and the others (blue) shift their visual attention (indicated by arrows) to the discoverer and approach the resource center 20 s later. Middle left: Probability of remaining on the resource for the initial discoverer, conditional on time since the discovery event. Middle right: Probability that the others (i.e., non-discoverers) are on the resource, conditional on time since discovery. Bottom left: Average in-degree of the discoverer over time. Bottom right: Average out-degree of others over time. Black line indicates participants in the Alone condition. Dark lines show posterior means of a Bayesian Gaussian process regression; shades represent 90% HPDIs. Statistics across all panels are derived from n = 621 participants (Group conditions [A–C]: Fast NP n = 72, VP n = 86, FP n = 74; Slow NP n = 80, VP n = 79, FP n = 86. Alone condition [Panel C]: Fast n = 73, Slow n = 71).

120

Valerii Chirkov @chirkovv.bsky.social · 24/09/2026

Key Finding 1: Groups only outperformed solo individuals when full payoff info was available—meaning they could see who was succeeding. Without it, groups actually performed worse, especially with fast-changing environment.

Mean tracking time, tracking efficiency, and total score (colored isoclines) per treatment. The dashed box indicates the slow resource environment, and the dotted box indicates the fast resource environment. A: Alone Condition; NP: No Payoff-Sharing; VP: Voluntary Payoff-Sharing; FP: Full Payoff-Sharing. Dots and error bars show posterior means and 90% highest posterior density intervals (HPDIs). Statistics are derived from n = 621 participants (Fast resource: A n = 73, NP n = 72, VP n = 86, FP n = 74; Slow resource: A n = 71, NP n = 80, VP n = 79, FP n = 86).

110

Valerii Chirkov @chirkovv.bsky.social · 24/09/2026

🎮 The Setup: We challenged 621 people (alone & in groups of 5) with a realistic search-and-tracking task in an immersive 3D environment. By changing resource speeds & social cues, we tracked how they adapted their visual attention & social learning.

Payoff visibility conditions. Alone: Participants are alone in the environment; No Payoff-Sharing: Participants are in groups of five and can observe others' positions, but not their current payoffs; Voluntary Payoff-Sharing: Participants are in groups of five and can continuously decide whether to share their current payoff (at a cost of 1 point/s); Full Payoff-Sharing: Participants are in groups of five and their payoff information is always shared (at no cost).

120