New preprint on the limits of detecting higher-order interactions in microbial communities.
www.biorxiv.org/content/10.6...
We find that the dominance of additive and pairwise interactions on community function may not reflect biological simplicity, but fundamental limits of statistical detection.
SFI's Laurent Hébert-Dufresne (@lhd.bsky.social) is the 2026 recipient of the Young Scientist Award for Socio- and Econophysics by the German Physical Society (DPG). Honoring “outstanding original contributions that use physical methods to develop a better understanding of socio-economic problems.”
Many are appropriately outraged by Altman’s comments here implying that raising a human child is akin to “training” an AI model.
This is part of a broader pattern where AI industry leaders use language that collapses the boundary between human and machine.
🧵/
I hope people (speaking to myself) will talk more and more about higher-order interactions and less and less about higher-order networks.
p.s., we could not have proven the result above with a standard, node-based mean-field approx, since, in fact, it completely forgets about the interaction structure. For the same reason, it cannot discern repeated interactions with a same agent from a single interaction with many.
e.g., how interactions of different orders are distributed across the network (inter-order correlations) matters for contagions. This is true no matter the representation you choose. I just find easier to think about and communicate it in terms of HGs than BNs.
journals.aps.org/prl/abstract...
Choosing one representation among several equivalent is about convenience, in this case to help intuition. Hypergraphs offer a "direct" representation of interactions; bipartite nets, less intuitively, map interactions to new nodes, but preserve a pairwise description and are more flexible.
Glad this piece is out! I never understood why some literature went obsessed with hypergraphs per se when a lot had already been done (and yet ignored) for bipartite nets.
What's interesting about higher-order interactions is...🥁...interactions – not how they are represented.
Here are your 10 -essential- AI prompts for academics ... make your life easy with help from @profserious.bsky.social profserious.substack.com/p/10-ai-prom...
Really happy to see this out in PRX Life! There you can find an eco-evolutionary framework integrating the evolution of viral infectiousness and antigenic features. While the former determines contagion events among hosts, the latter tells us how quickly viruses can escape population immunity 1/4👇
Excited about this paper and the interactive story to accompany it. Congrats @lhd.bsky.social @juniperlov.bsky.social @giulioburgio.bsky.social @sfiscience.bsky.social @unioflimerick.bsky.social and nice story telling @jstonge.bsky.social!
Very cool interactive story, @jstonge.bsky.social!
"[...] real social cascades aren't simply branching processes with fixed rules." A self-reinforcing mechanism is what we propose in a recent piece led by the one and only @lhd.bsky.social.
It might start as a joke, belief, or rumor, easy to dismiss. But then it twists, builds momentum, and spreads like wildfire. Why do some ideas die out while others go viral?
A new study by researchers from the University of Vermont and the Santa Fe Institute offers answers: santafe.edu/news
Our team had an amazing week at @ic2s2.bsky.social in Norrköping Sweden and we will post pictures of our posters and talks soon - the big news is that we're so excited to host #IC2S2 in Burlington in 2026! youtu.be/p412S4GnPkc
There's amazing work on group effects in higher-order networks, but not a lot of connections to social ontology, collective action, and group selection.
Led by @jstonge.bsky.social with expert guidance of @rharp.bsky.social we reviewed and formalized these connections.
arxiv.org/abs/2507.02758
Defining and classifying models of groups: The social ontology of higher-order networks arxiv.org/abs/2507.02758
Not sure we'll ever understand adaptive systems enough. But what we're sure of is that one basic reason is that you can't even start to describe them properly w/o preserving local dynamical correlations.
A fun and frustrating long way to go.
w/ the amazing @lhd.bsky.social & @gstonge.bsky.social.
During a pandemic such as COVID19, we hope (but fail) to accurately estimate the incidence of the disease. In this paper, we propose a new approach to machine-learn models of the real incidence from readily available information (tests and detected cases) dx.doi.org/10.1371/jour...
Cartoon of a top-down central state combined with a local network of decisions units.
Cartoon of decisions as a statisfiability problem (decision network) where each decision is solved by a higher-order network (governance network) of agents with their own opinions on what decisions should be made.
Some decisions are best made quickly and locally. Governance can work better as a higher-order network, not a pyramid around a central state. How should we design these networks?
We looked at this with law and complexity scholars and found "effective governance" networks.
arxiv.org/abs/2412.03421
We are thrilled to share our new pre-print, “Self-Reinforcing Cascades: A Spreading Model for Beliefs or Products of Varying Intensity or Quality,” now available on arXiv! arxiv.org/pdf/2411.00714