Summary: In both social groups and neural circuits, relying on familiar connections feels safe — but it can also trap new information in a self-reinforcing loop. A recent Northwestern University study introduces a theoretical framework showing how Hebbian learning, the principle often summarized as “neurons that fire together, wire together,” can unexpectedly hinder the spread of activity across a network.
While positive reinforcement strengthens existing ties and makes repeated patterns more likely, it can confine signals, ideas, or behaviors to narrow loops. In contrast, weakening or “negative” reinforcement allows activity to escape entrenched routes and explore new pathways, promoting broader propagation.
Key Facts
- Hebbian reinforcement forms loops: The study embeds the rule that repeated interactions strengthen links. As connections grow stronger, they become more insular, acting as barriers to external information and trapping activity in feedback loops.
- “Ant mill” analogy: Lead researcher István Kovács likens strong positive feedback to ant mills, where ants follow pheromone trails in a circle until they perish. In social and neural networks, the same mechanism can create a “death spiral” that keeps ideas or signals circulating among the same nodes.
- Spread requires weakening old ties: For an idea, infection, or signal to spread efficiently, systems must deviate from established routes. Negative reinforcement — weakening familiar links — forces exploration of new nodes and supports wider propagation.
- Broad applicability: The model applies to many systems where activity spreads, including social media echo chambers, disease transmission, and neural signal propagation in the brain.
Source: Northwestern University
Sticking with familiar people feels comfortable, but it can limit reach.
A new study from Northwestern University demonstrates that repeatedly interacting with the same contacts can confine information to tight echo chambers. Conversely, when interactions shift away from familiar connections and toward new ones, activity spreads farther across the network.
To study how activity propagates, the researchers developed a network model that incorporates simple learning rules. Traditional network models typically assume fixed relationships, but this framework lets connections evolve in response to experience. As links strengthen or weaken with each interaction, the whole network reshapes over time.
The results are relevant beyond social networks: they apply to the spread of infections, the propagation of neural signals across brain regions, and the diffusion of behaviors in animal groups. The research suggests a basic trade-off: reinforce existing ties and activity stays local, or reduce reinforcement to encourage exploration and broader spread.
The study was published online on April 27 in Communications Physics.
“Learning and adaptation are central to biological and social systems, yet their effects are often missing from simple models,” said István Kovács, lead author and assistant professor of physics and astronomy at Northwestern. “We found that positive incentives, by strengthening existing connections, can surprisingly prevent activity from spreading. Weakening connections, on the other hand, encourages the system to avoid old routes, enabling more effective spreading.”
Kovács, a complex systems researcher and member of the Northwestern Institute on Complex Systems and the NSF-Simons National Institute for Theory and Mathematics in Biology, led the study with co-first author Will Engedal, a recent graduate of his research group.
“Fire together, wire together”
The team explicitly modeled Hebbian learning, a principle introduced by Donald Hebb in 1949 to explain how repeated co-activation strengthens connections and helps form memory. In plain terms, when two nodes activate together, their link becomes stronger, making future joint activation more likely.
In the new model, nodes represent entities such as people, neurons, or animals, and links change based on outcomes. The researchers tested two learning modes: positive reinforcement, where successful interactions strengthen links, and negative reinforcement, where interactions lead to link weakening.
Depending on whether the source node, the target node, or both adjusted their links after interactions, the network produced different emergent behaviors.
Trapped in a “death spiral”
When the source node applied positive reinforcement, activity tended to loop back along the same routes and become trapped in closed circuits. This feedback prevented the signal from reaching new parts of the network. By contrast, when connections weakened after interactions, activity was forced to seek fresh paths and spread more widely.
“This mirrors the ant mill phenomenon,” Kovács explained. “Blind fire ants can get trapped following their own pheromone trail in a closed loop. The trail strengthens as they continue, so they persist until they die. Positive feedback in networks can create a similar self-sustaining loop.”
Because the model centers on how past interactions shape future ones, Kovács expects the findings to apply across many spreading processes. The team plans next to investigate whether these learning-driven effects appear in real-world networks and how they interact with more complex behaviors.
Funding: The study, titled “Activity propagation with Hebbian learning,” was conducted in collaboration with HUN-REN Wigner RCP in Hungary and supported by 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 the Baker Program of Undergraduate Research at Northwestern University.
Key Questions Answered:
A: Not necessarily socially. But from a network perspective, constant agreeability fosters echo chambers: talking only to like-minded people keeps ideas circulating within the same group. To spread an idea, you need the friction introduced by new, less-familiar connections.
A: It helps explain habit and thought loops. Repeated co-activation strengthens neural pathways, creating efficient, well-worn routes. While useful for memory and skill, those pathways make it harder for new or conflicting information to be integrated because the brain favors the established path.
A: Yes. When content keeps recirculating among the same users, it loses momentum and fails to penetrate new clusters. Effective viral spread requires breaking out of initial comfortable networks.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- The journal paper was reviewed in full.
- Additional context was added by editorial staff.
About this neuroscience research news
Author: Amanda Morris
Source: Northwestern University
Contact: Amanda Morris – Northwestern University
Image: The image is credited to 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
Biological and social systems — from infection spread and inter-regional brain activity to population dynamics — show learning across many scales. These processes require models that incorporate local learning rules into contact processes.
The authors introduce learning as either positive (Hebbian) or negative (anti-Hebbian) reinforcement of the activation rate between two sites after each successful activation.
They show that Hebbian learning produces a wide range of emergent behaviors: local incentives can produce opposite global outcomes. Generally, positive reinforcement tends to eliminate the globally active phase, while negative reinforcement can convert an inactive system into one that is globally active.
Analytical and numerical results indicate that in two dimensions and higher, negative reinforcement both promotes spreading and creates effectively immune regions, producing two distinct critical points. Positive reinforcement can lead to Griffiths-like effects with non-universal power-law scaling, reflecting the “ant-mill” phenomenon described in the study.