Summary: Staying within familiar social circles or neural pathways can feel safe, but a new theoretical study shows this habit can create a “death spiral” that prevents new information from spreading. By incorporating simple learning rules into network models, researchers demonstrate that Hebbian learning — the principle often summarized as “neurons that fire together, wire together” — can reinforce existing routes so strongly that activity becomes trapped. In contrast, weakening connections allows activity to break free and explore new parts of the network.
Positive reinforcement strengthens established links and makes them more likely to be reused, but that very reinforcement can trap ideas, signals, or infections in tight loops. Negative reinforcement, which reduces the strength of previously used connections, opens alternative routes and improves the system’s ability to spread activity more broadly.
Key Facts
- The Hebbian Loop: Incorporating the rule that repeated interactions strengthen links, the study finds that stronger connections can act as barriers to outside information, keeping activity confined to a reinforcing feedback loop.
- The Ant Mill Effect: Lead researcher István Kovács compares unchecked positive feedback to “ant mills,” where ants follow their own pheromone trail in a circle until they perish. Social and neural networks can form analogous loops that prevent outward spread.
- Efficiency Through Weakness: For an idea, contagion, or signal to spread efficiently, the system must avoid overusing old paths. Weakening connections (negative reinforcement) drives activity to discover and engage new nodes.
- Universal Dynamics: The model applies to many spreading processes: social media echo chambers, viral epidemics, signal propagation in the brain, and behavioral spread among animals.
Source: Northwestern University
Sticking with the same people or pathways may feel comfortable, but it can limit reach.
A new study from Northwestern University shows that repeatedly interacting with the same contacts — or repeatedly activating the same neural routes — can confine new ideas or signals to closed loops. When interactions shift away from the familiar and toward new partners, activity has a much better chance to travel further across the network.
To investigate how activity spreads across networks that learn from experience, the researchers developed a theoretical framework that adds simple learning rules to traditional spreading models. Whereas classic network models assume fixed links, this approach lets connections strengthen or weaken depending on past interactions, allowing the network’s structure to evolve as activity propagates.
These findings extend beyond social conversation: they apply to any system where activity propagates and links adapt, including the spread of infections, regional brain signaling, and collective behaviors in animal groups. The core insight is simple: repeated interactions that reinforce the same links can confine activity, while mechanisms that reduce reliance on those links can facilitate broader exploration and spreading.
The study was published online April 27 in Communications Physics.
“Learning and adaptation are intrinsic to biological and social systems, but the effects of learning are still largely unexplored in even simple models,” said István Kovács, the study’s lead author. “We found that positive incentives can strengthen existing connections, which unexpectedly prevents activity from spreading. When connections weaken, the system avoids old paths and promotes more efficient spreading.”
Kovács, an assistant professor of physics and astronomy at Northwestern’s Weinberg College of Arts and Sciences and a member of the Northwestern Institute on Complex Systems, led the work. Will Engedal, a recent graduate from Kovács’ research group, is co-first author.
‘Fire together, wire together’
The team formalized Hebbian learning — the idea that repeated co-activation strengthens connections — within a spreading-process model. Originally proposed by Donald Hebb in 1949 to explain how the brain forms associations and memories, Hebbian learning is often summarized as “neurons that fire together, wire together.” When two nodes repeatedly activate together, their link grows stronger and becomes more likely to carry future activations.
In the model, nodes (representing people, neurons, animals, or other entities) exchange activity along links whose strengths change after interactions. The researchers tested two learning modes: positive (Hebbian) reinforcement, which increases link strength after successful transmission, and negative (anti-Hebbian) reinforcement, which decreases it. They also explored how learning at the source node, the target node, or both affects system-wide outcomes.
Stuck in a ‘death spiral’
When positive reinforcement predominates, activity tends to circle back along the same strong routes, forming confined loops rather than reaching new regions. Kovács likens this to the ant mill phenomenon: blind ants that follow a pheromone trail can become trapped in a circular path that gains strength as more ants follow it, eventually causing exhaustion and death. In networks, similar positive-feedback loops can lock activity into echo chambers.
Conversely, when connections weaken after use, activity is pushed to explore alternative routes, increasing the likelihood of wide-scale propagation. Because the model captures a fundamental mechanism — how past interactions bias future ones — the authors expect these results to be robust across many spreading scenarios. The next steps include testing whether these effects appear in empirical networks and how they interact with more complex behavioral rules.
Funding: The study, “Activity propagation with Hebbian learning,” involved collaboration with HUN-REN Wigner RCP in Hungary and received support from Hungary’s National Research, Development and Innovation Office (award K146736), the U.S. National Science Foundation (award PHY-2310706), the Hungarian Academy of Sciences, and Northwestern University’s Baker Program of Undergraduate Research.
Key Questions Answered:
A: Not necessarily in everyday interactions. The study highlights that, from a network perspective, constant agreeability can create echo chambers. To spread ideas beyond a core group, some engagement with unfamiliar contacts — and the friction that comes with them — helps an idea jump to new communities.
A: The results help explain habit formation and persistent thought patterns. Repeatedly co-activated neural pathways become very efficient, which supports memory and skill but also makes it harder for new or conflicting signals to take hold because activity preferentially follows the entrenched route.
A: Yes. Content that circulates mainly within the same saturated cluster loses momentum for reaching new groups. Successful viral spread often requires breaking out of the initial, comfortable network into fresh clusters.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- The journal paper was reviewed in full by the editorial team.
- Additional context was provided by staff to clarify implications for social and neural systems.
About this neuroscience research news
Author: Amanda Morris, Northwestern University
Source: Northwestern University
Contact: Amanda Morris – Northwestern University
Image credit: Neuroscience News
Original Research: Open access. “Activity propagation with Hebbian learning” by Will T. Engedal, Róbert Juhász & István A. Kovács. Communications Physics. DOI: 10.1038/s42005-026-02638-z
Abstract
Activity propagation with Hebbian learning
Learning occurs across many biological and social systems — from the spread of infections to the propagation of brain activity and the movement of populations. These processes call for models that incorporate local learning rules rather than assuming static links.
The study introduces learning as either positive (Hebbian) or negative (anti-Hebbian) reinforcement of activation rates between pairs of sites following successful activations. The analysis shows that Hebbian learning produces a range of emergent behaviors: positive reinforcement often eliminates the globally active phase by confining activity, while negative reinforcement can convert an otherwise inactive system into one that supports widespread activity.
Analytical and numerical results indicate that in two dimensions and above, negative reinforcement both promotes spreading and creates effectively immune regions, giving rise to two distinct critical points. By contrast, positive reinforcement can generate Griffiths-like effects with non-universal power-law scaling, reflecting the ant-mill phenomenon described in the study.