Focus on Arts and Ecology

By Sara Goudarzi | October 13, 2025 

From conspiracy theories to new fashion trends, some online ideas explode in hours—capturing attention and shaping discourse—while others hardly gain attention. What makes some ideas go viral while others fizzle out? A recent study in Physical Review Letters sheds new light on this question. Researchers from the University of Vermont and the Santa Fe Institute devised a model that treats a piece of information not as static and unchanged as it spreads, but one that gains or loses strength in the course of propagation, much like a forest fire. Unlike traditional models which treat a joke, idea, or even a virus as fixed while it is circulating, this new model shows that the ability of an idea to shift in intensity during transmission is what can lead to the viral patterns seen online.

This insight could help experts better understand—and possibly curb—misinformation, by identifying how and when information gains or loses strength as it moves through digital networks. I spoke with study co-author Sid Redner, a professor at the Santa Fe Institute, to find out more about the new finding.

This discussion has been edited and condensed for length and clarity.

Sara Goudarzi: What typically happens when something goes viral, and what it is that you found with your model that was new?

Sid Redner: I can’t speak to why in general things go viral: That’s kind of a random process. But the basic idea in our model is that when something like a rumor or a fad or something else contagious becomes popular, that somehow the quality of the fad itself changes because more people are doing it. So, to give maybe a more concrete example. If you imagine a forest fire: When it first starts, it’s not very hot around it. You can approach to within a few feet of it, and you won’t feel much heat. But as it gets larger, the heat generated by the fire heats up the air around it, and so it spreads more easily. So that’s the kind of feature that we tried to encapsulate in this model—that the actual physical thing that is being transmitted changes character as it becomes more popular or more widespread.

Goudarzi: And you found that it’s the change that helps it spread further, correct?

Redner: Yeah. Say, in an epidemic, you get the flu. The nature of the flu is, roughly speaking—except for the few who die from it—more or less the same for everybody. You feel miserable for a few days or a week. But sometimes the flu virus mutates so that it has more serious effects, or it mutates to a form where more people would get it—or it does both at the same time. That would be a feature that is encapsulated in our model.

Goudarzi: So, is it the mutation that makes something more viral?

Redner: One could say maybe it’s a mutation. In the case of forest fires, it’s not; it’s the actual heat itself which is generating more heat. But yeah, you could think of it as a mutation that’s engendered by the fact that it’s become more popular.

Goudarzi: So, to use an example from social media: If you make a post, and the post gets repeated exactly as it were, there is less of a chance for it to spread than say, if somebody adds something to it, editorializes a bit, and then reposts it. Is that correct?

Redner: That’s a very good analogy. Yes, it’s like the post itself might have a certain quality; for example, you tell a joke; it’s sort of funny, but if somebody embellishes it so that it becomes funnier, then that enhancement makes the joke spread more easily.

Goudarzi: How does this model differ from previous models of virality?

Redner: To my knowledge of the literature, [in previous models] the actual physical attribute that is spreading is not changing as it spreads. It’s just that everybody has a certain rate of susceptibility of either joining the fad or getting sick [in the case of disease], but the actual thing which is spreading is not changing character.

Goudarzi: So previous models were treating it more like a branching tree?

Redner: When you say branching tree, this is sort of a simplified model for the spatial spread that allows one to analytically attack the problem. There are not any closed loops in the tree so that when you tell a rumor to a friend, it cannot come back to you through multiple cycles of personal connections.

Goudarzi: How were you inspired by forest fires to devise this model?

Redner: This is really a convoluted story. Three years ago, I wrote a paper with a collaborator that was called “Birds on a Wire,” (with apologies to Leonard Cohen). I do a lot of bike riding in New Mexico, and I’m always fascinated by birds sitting on wires. And I thought, how far apart are the birds? I came up with a very simple-minded model, which is birds are skittish but they’re also sociable.

If a bird lands too close to another bird on a one-dimensional wire, the [sitting] bird will fly away. But if it lands close, but not too close to an already sitting bird, then they just stay there. And then, if you imagine that there are birds always coming onto the wire at a fixed rate, and then some birds are flying away because somebody elbows them, one can ask, what is the distribution of birds on the wire? What is their separation? What are the avalanches that happen when birds fly away? So, we solved that model. It was kind of cute, but I was not happy with that model because it didn’t encapsulate the feature that sometimes you see that one bird lands and 20 fly away. I was trying to think of some mechanism which was just that: If one skittish bird starts flapping its wings that maybe excites the next bird over, and then two birds are flapping their wings, and it makes sufficient commotion that a bird far away is disturbed, and a whole cluster of birds start flapping their wings, and the whole cluster flies away.

It turns out that this type of mechanism also occurs in schools of fish: When one fish darts away from a school, typically nobody else goes. But if multiple fish go, all of a sudden, many of them go. So, I was thinking about a model of that kind. The bird model as stated above was too complicated. But we physicists always try to simplify models as much as possible to make them solvable while still respecting the phenomenology. And I thought, suppose that the disturbance just propagates in one direction on the wire—birds only downstream. As soon as I thought of that, I realized this is nothing more than a forest fire in strong wind—where the forest fire propagates downstream. If the fire comes to a gap between trees, and if the fire is intense enough, it can jump the gap. When the fire jumps the gap, it loses some intensity. But then after the fire crosses the gap and starts burning more trees, the fire increases intensity. So, there’s this cascade happening that will only stop when the intensity of the fire at a gap is not strong enough to jump the gap. That was the starting point. And then [study co-author] Laurent [Hébert-Dufresne], who is very astute and clever said, “this could be a model for the spread of rumors.” And he took off with that.

Goudarzi: Do you think that this idea of change in your model is responsible for the proliferation of misinformation?

Redner: I don’t have an answer for that right now, because our model is purely abstract. We haven’t applied it to real data or tried to make a model for something like disinformation or misinformation. That’s certainly a possibility and something we might look at down the road, but at the moment, it was a purely theoretical model that we could study analytically.

Goudarzi: But can it be used for better understand the information landscape?

Redner: Absolutely. To give you an example, there was the speeches by Hegseth and Trump to all the generals; they were all sitting there completely quiet; no applause, little acknowledgement. Even Trump said, “well, I’ve never been in a crowd this quiet.” And, I just thought, they’re military guys. They’re being respectful and quiet, because that’s what they’re taught to do. But then, a lot of the talk show hosts were making fun of the whole event, and then I realized, wait a minute, maybe they really disrespected him, and this is a way of showing disrespect. So, somehow by people commenting on it, it changed my perception of the event for me. I think that that kind of feature could also be used in the notion of spread of good or bad information.

Goudarzi: If you apply this idea to real data and find out that it does play a role, how could we use this understanding to fight misinformation and disinformation?

Redner: Now you’re asking me a hard question that I don’t have a good answer for. I’m a theoretical physicist. I don’t usually delve into policy issues or the world of data too much, except to verify the basic ideas of a theoretical model. So, I don’t want to touch that one at the moment, because I just haven’t thought about it seriously enough.

Goudarzi: What’s the significance of this finding?

Redner: I think the significance is that there are lots of models for spreading of disease, rumors, fads, whatever, but they all had this basic feature that the susceptibility of how the fad would spread was on the side of the user. How susceptible am I to getting sick or buying a new car or something like that? But what’s not encapsulated in all these models is the fact that the actual object which is spreading can also change in character by the very spreading phenomenon itself. So that’s the new feature of this model.

Author:

Sara Goudarzi is associate editor, disruptive technologies at the Bulletin. Her work has appeared in Scientific American, The New York Times, and National Geographic News, among others. She is the author of the debut novel The Almond in the Apricot and several children’s titles including Leila’s Day at the Pool. Find her at saragoudarzi.com and on Twitter at @Saragoud.

(Sources: Bulletin of the Atomic Scientists)