Related but distinct: Information cascades

Information transmission is analogous to infectious disease transmission: information that sustains $R_0>1$ spreads "virally" to significant fractions of the world; information with $R_0<1$ has limited spread.

Any information that remains in circulation stays at $R_0=1$^[1], analogous to an endemic infectious disease^[2]. This is an example of self-organised criticality, where the system as a whole tunes itself to the critical point without requiring external intervention.

For individual ideas: Small changes in the information (presentation or substance), or the information environment (social media or the social world), can change the $R_0$ of a piece of information across this critical threshold of 1, and thus make the difference between "everybody knowing" and "nobody knowing" that information.

Thus, small changes to an idea can change whether "everybody knows" it, and it is part of general background knowledge which can be built upon, or whether "nobody knows" it. But this is rare: most ideas have $R_0\ll 1$ once outside their niche audiences.

For the overall information space: Adding up all pieces of information that thus became viral / ceased to be viral, and the small change in social media (algorithms or UI) or social world can tip the information space into different attractor spaces.

That "different attractor space" can mean changing the norms of conversations, as well as moving the Overton window as a whole. The overall information space is metastable.^[3]

Other examples of self-organised criticality

Eric Raymond has previously written an example, on just-in-time and efficiency pushing supply chains towards criticality.

Here I highly recommend the video format. Self-organised criticality is a system dynamics property, not a system statics property. Veritasium has made a great video, which includes simulations of sandpiles, ferromagnetism, and forest fires. (Of which, ferromagnetism can exhibit criticality, but not self-organised criticality.)

1. When averaged over the long term. ↩︎

2. How do endemic infectious disease stay at an effective reproduction number of $R_0=1$ (averaged over the long term)?

   An endemic infectious disease likely started as a (perhaps local) epidemic with $R_0>1$, infecting a significant fraction of the (local) population, giving them some amount of immunity (and reducing the effective population of infection-naive individuals), reducing $R_0$ to near 1.

   Thereafter, human behaviour (e.g. acting more cautious when the community sees infections rising, and less cautious when infections seem to fade) cause the disease to fluctuate between $R_0<1$ and $R_0>1$, averaging over the long term to $R_0=1$.

   I technically should use the effective reproduction number $R_t$ instead of the basic reproduction number $R_0$. The difference is effective immunity: $R_t=R_0\times (1-\text{effective fraction of the population that is immune})$. ↩︎

3. To be precise:

   Any piece of information in circulation, like any particular endemic infectious disease, is in the metastable state of "in circulation" or "endemicity". The stable state is the piece of information ceasing to circulate, or the disease being eradicated.

   A piece of information "going viral" is an information cascade, a special case of a threshold cascade.

   The set of all circulating information, like the set of all endemic diseases, is self-organised critical.

   The broader system (including policy inputs like social media algorithms or UI, and non-explicitly-controllable "policies" like norms and Overton window) is metastable. ↩︎ ↩︎