In Japan for the next two weeks, Tokyo and Kyoto, if anyone in my network is available to meet up!
01.01.2026 21:31
👍 2
🔁 0
💬 0
📌 0
In Japan for the next two weeks, Tokyo and Kyoto, if anyone in my network is available to meet up!
While this work is currently theoretical and highly preliminary, we are conducting work on formalizing these measures to apply to real-world collectives and forms of emergence.
@ketikagarg.bsky.social @grf.bsky.social
Open access link below:
www.sciencedirect.com/science/arti...
🚨Excited to share our new theoretical framework to classify and spark mechanistic inquiries into various forms of collective intelligence!
Soon to be published in @cognitionjournal.bsky.social
Thanks to @divintelligence.bsky.social
Thread by Cody Moser @culturologies.co 👇🏽
Thrilled to see this paper out, two years after starting our collaboration at @divintelligence.bsky.social
Such decompositions can be applied invariant to a given system's goals or end-states: using a recently published "piano mover" task comparing groups of ants and humans at the same task, we show that forms of information processing are a more useful means for decomposing collectives than their goals.
Tying these different intelligences into known concepts of emergence, we argue integrated information decomposition can help classify behavior of collectives into different emergent regimes; in aggregate systems, upward causation dominates; while in structural systems, downward causation dominates.
Viewing these systems across these axes, we classify different collective phenomena into separate forms of collective intelligences: high IP, low GP aggregate intelligences; low IP, low GP statistical intelligences; low IP, high GP structural intelligences; and high-high synergistic intelligences.
Using Wimsatt's rules of aggregativity, we propose consilience between these phenomena is possible by examining their information processing capabilities across two dimensions: individual processing of group-level information (IP) and group-level sensitivity to individual arrangements (GP).
A core issue with this emerging discipline is a lack of underlying principles for theoretical unification. While multiple phenomena are classified as collective intelligence (flocks, swarms, human organizations, and culture), it's unclear what theoretical binding such disparate systems have.
My co-authors and I are happy to present our framework "Collective Intelligence as Collective Information Processing (CIP)."
Here we propose decomposing different information processing mechanisms to unify disparate phenomena traditionally classified as "collective intelligence."
Gave what I thought would be objectively my most boring lecture for my students last week and they absolutely loved it. I had no idea students would be so passionate about contract theory!
Fast forward to what I thought would be my most interesting lecture this week ... and I saw heads nodding off.
You can check out the pre-print here:
Moser, C., & Smaldino, P.E. (2025). Limit Cycles in Opinion Dynamic Networks with Competing Stubborn Agents.
osf.io/preprints/so...
Several implications emerge regarding online information architectures: the space of analytic solutions is constrained by networks with these cycles. Second, in an increasingly connected world, more leverage points exist for nefarious actors to disrupt our online ecosystems.
Importantly, the length and complexity of these cycles is dependent on the path length of the network: in centralized networks, centrality is devalued such that multiple optimal positions emerge and agents spend more time chasing optimal positions in these cycles.
Known as the counter-optimal stubborn agent placement problem, previous research has looked at the use of greedy algorithms for finding optimal positions in this NP-hard task.
We find that as agents attempt to disrupt each other certain network configurations lead to limit cycles in agent position.
We examine agent positioning strategies with repeated iterations: once a node has established its optimal influence over a network, we ask how counter-stubborn agents with opposite opinions can "disrupt" the influence of their counterpart.
New pre-print with @psmaldino.bsky.social on an agent-based model of propaganda in online spaces.
Using the voter model from physics, we simulate networks with stubborn agents (zealots) who do not change their opinions, asking about their optimal positioning for influencing the network.
Cognitive scientists and computer science PhDs talking about David Marr:
Not to resurrect overtrodden "AI art" discussions, but I think I'm closer to viewing the image on the left as AI-generated art than the image on the right.
The fact the journal editors and authors missed it feels like a tremendous statement in itself.
The man is a machine
Are the laws of nature subjective or Platonic revealings of our world? What does it mean to know what one cannot articulate? What is the connection between scientific theory and riding a bike?
In this episode we tackle the work of Michael Polanyi in Personal Knowledge:
m.youtube.com/watch?v=d1gs...
Theories of organizational structure emphasize different dimensions of managerial design, including span of control (Simon 1947), decision-making speed (Aswamenakul et al. 2025), task delegation (Moser & Smaldino 2022), and managerial monitoring (Zefferman 2023).
Cody and his committee: Tyler Marghetis, Alex Petersen, me, Cody, Justin Yeakel.
Congrats to Dr. Cody Moser (@culturologies.co)! Cody successfully defended his dissertation today, and is off to Morocco to start a faculty position at UM6P’s School of Collective Intelligence!
Cody's dissertation cover
NHB paper linked in the tweet
Mega congrats to Dr Cody Moser @culturologies.co who passed his thesis defense at UC Merced
Cody was one of my first students at @themusiclab.org, co-leading work on the sounds of infant-directed vocalizations in our species doi.org/10.1038/s415...
He is starting a lab in Morocco soon! Woo!
I appreciate these comments, by the way, this is good for tightening up the text in the paper to specify what we are doing. Thank you!
New preprint! 🚨How do groups process information? What does it mean for part of a network to "explore" and part to "exploit"? How do we move beyond phenomenological correlation in our analyses of networks to a causal theory of collective intelligence?
Here, we try to answer these questions:
Yes, I agree.
I agree we can't judge accuracy, hence we don't use it as a benchmark. We're looking for convergence in the Collins dataset and merely used inter-rater reliability to see if the two came to similar classifications.
Sure, they probably used the same texts, but I don't see this as a problem.
