A while back, I wrote about emergent cooperation phenomena amidst groups on the internet. I’m revisiting this idea, this time around layering in some additional context based on a recent exposure to the idea of causal emergence.
I was captured by this idea because it seemed intuitive, and I think that intuition is something that isn’t followed enough these days, especially in science.
The idea of causal emergence as a whole is a balance check to reductionism, which says that the deeper we look, the more answers we’ll find. If we look at the last hundred years, that hasn’t exactly been the case. The deeper we go, it seems to be that we are finding only more unanswered questions, and that our science is getting less bang for its buck.
What is causal emergence?
Causal emergence happens when the causal mechanisms of a system are more informative at a higher scale as opposed to at the scale of observation where things are happening. When stuff happens, noise gets made. When dealing with noise, an observer may have a more challenging time obtaining new information about a system. The fact that more information is able to be obtained through observation at a higher scale, suggests that the causal elements of the stuff that is happening at the lower scale may not originate at that scale, but rather at the scale which contains more effective information.
Effective information is a measure of uncertainty between causal relationships. This measure of uncertainty helps to come to conclusions about the effectiveness of given causes that perturb a system to produce certain effects.
So for example, if a microscale’s measure of effective information is lower than it’s macroscale, then the macroscale offers more information that helps come to conclusions about the system as a whole.
But what would this look like in practice? Returning to this idea of emergent cooperation phenomena, let’s look at how we might be able to measure the causes for a group of strangers to cooperate without a clear top-down leadership structure.
From the perspective of causal emergence, we can look and measure this phenomena at its microscale and its macroscale. Looking for causes at the former scale of observation would likely start by having us drawing a social graph of everyone involved that might look something like this:
We are assuming that this graph consists of causal relationships between individuals working towards an objective. Each node in the graph represents an individual, and each edge between nodes means that information of some kind gets exchanged between those individuals. As mentioned above, we can measure these causal relationships by their effective information, in order to come to conclusions about the ability of certain causes to promote certain outcomes.
In this image specifically—which I (stole/borrowed?)—there are 1764 nodes/people with which we’ll say we can derive {X} bits of effective information from. Is that enough to come to any definitive conclusions? Well, not exactly.
Notice how upon first glance, the image may seem to be following a deterministic gradient from the first few nodes on the right unto the cascades on the left or vice versa. If you look closer, the flow of information quickly begins to pickup uncertainty—or noise. This is due to the varying directional flow of information indicated by the arrows in the graph and the multiple points which contain 1:many or many:1 flows of information.
It may be impossible to ever have deterministic relationships at a micro-scale. In a sense evolution deals with a constant source of noise, noise similar to that defined in information theory wherein sending a signal always has some degree of error. Therefore, in order to make sure that causes lead to reliable or deterministic effects, biology necessarily needs to operate at the level of macro-scales.1
From micro to macro
Now, let’s take for example the same network after having been elevated into a macroscale. High level, this has been done by running an algorithm over the graph to identify clusters of nodes/people that could be compressed into a ‘macronode’. This macronode represents a summarized index of the effective information of the cluster.
In our new graph, we’ve managed to reduce our number of nodes to 596 and increase the effective information of the system by {X}%!
And so how does this help us- the lowly onlooker who simply wants to know what’s causing the phenomena of cooperation amongst our hypothetical group of internet folk. Well for starters, the ability to look at the social graph and identify key macronodes that are more effectively transmitting information than others. Some macronodes may themselves be a top level cause that is influencing the network as a whole. Furthermore, the identification of these clusters allows for a maniacal observer to jump into the social graph where they might be able to either:
Receive new information with greater velocity
Transmit information with greater velocity
Therefore, we can define causal communities as being when a cluster of nodes, or some subgraph, forms a viable macronode. Fundamentally, causal communities represent noise at the microscale…2
Albeit for my explanation here I’m being a bit more literal in applying the term ‘communities’. Mapped to the original idea of emergent cooperation, a causal community as explained here could be analogized to something like a Discord, Telegram, Slack or other group communication channel.
…this continued point is the most interesting:
The closer a subgraph is to complete noise, the greater the gain in Effective Information by replacing it with a macronode.
Lets say we have a communication channel with 5000+ members sending messages that contain information 24/7 on a given topic. There may be hundreds of messages exchanged between members daily, but with a very low noise:signal ratio. For the broader subject area, there may be a dozen+ other communication channels all day every day doing the same thing across areas such as stock investing, blockchain/crypto, etc.
What the idea of causal emergence shows is that rather than trying to follow the flow of effective information across each individual node, we can group the most noisy players across these channels into their own subgraphs. Then we need only to track it across one macronode with an indexed statistic of the total noise represented by that grouping, thereby improving an observer’s ability to come to conclusions about the network’s overall causation.
Beyond Emergent Cooperation
I had the pleasure of coming across this idea from Erik Hoel, who in addition to publishing a delightful essay newsletter which you can subscribe to here is also a neuroscientist and neurophilosopher.
His primer and synthesization of ideas surrounding the concept captured my attention. Moreover I was surprised and pleased to find in one of his references another paper applying this concept specifically to to biology, coauthored with someone who’s work I’ve expressed admiration for here before in past posts.
Many types of models in biology, from protein networks to physiological ones, can now benefit from a quantitative analysis of their causal structure, revealing the “drivers” of specific system-wide states and thus suggesting new strategies for rational interventions. Moving beyond traditional definitions of information to analyzes of causation in networks across scales can help drive new experimental work and applications in regenerative medicine, developmental biology, evolutionary cell biology, neuroscience, and synthetic bioengineering.
However it’s clearly noted, that the range of application does not stop at biology…
…We find that informative higher scales are common in simulated and real networks across biological, social, informational, and technological domains.
We need models like causal emergence to help narrow the search space of solutions that elude our understanding. This narrowing is not only of potential solutions to problems that we know we want to solve, but also to expose ourselves to problems that we may have never dreamed we’d have the power to solve.
Dream big.
-Benjamin Anderson
Send me signal on Urbit: ~padlyn-sogrum
Erik Hoel & Michael Levin (2020) Emergence of informative higher scales in biological systems: a computational toolkit for optimal prediction and control, Communicative & Integrative Biology, 13:1, 108-118, DOI: 10.1080/19420889.2020.1802914
Klein, Brennan & Hoel, Erik. (2020). The Emergence of Informative Higher Scales in Complex Networks. Complexity. 2020. 1-12. 10.1155/2020/8932526.