Thread by @MattPirkowski

Do you ever wonder why it feels like the Internet has made us collectively dumber, despite the fact that it increases individual access to information?

The answer lies in the concept of a Complexity Catastrophe, and how our present Internet usage patterns create one.

A thread:

Before describing exactly what a Complexity Catastrophe (CC) is, why CCs emerge, and how they're relevant to our present Internet landscape, we must first unpack a few theoretical concepts from evolutionary biology.

First up, the Adaptive Landscape model of evolution.

The Adaptive Landscape allows us to discuss how a given combination of (typically genetic) traits perform in a given environment (fitness), relative to other combinations of said traits.

As shown below, you can think of the possible combinations as a "mountainous landscape"...

Each dot on the above landscape therefore represents an actual organism with a specific genetic combination.

Movement across the landscape thus maps to a change in the underlying genetic combination, and results in a different level of "fitness", according to the landscape.

As is noted in the first image, local selection pressures will typically move a population to the top of a nearby peak. This is what's called an Adaptive Walk (AW).

So, then, what does it mean for a population to walk upon such a landscape?

We call that "evolution".

Here we should pause to clarify what the landscape itself actually represents, because its meaning lies far deeper than genetics.

Fundamentally, an AL represents the *stable information advantage* gained via the realization of a possible combination, given a specific context.

Thus one can generate an AL for "product space", in which the genetic combinations are replaced with feature combinations, and "fitness" maps to "product market fit".

This concept guides the product development strategy at Netflix (at least it did prior to 2015, when I left).

As you can see, the concept of an AL is quite flexible, and can be generalized to any situation in which there exists a relation between some combination of generative features and the degree to which said combination of features varies with respect to a desired metric.

But you might ask: "don't these landscapes also evolve?"

Which would be an excellent question, to which the answer is, most emphatically: yes!

The evolution of the AL represents the ever-changing relation between the fitness of specific combinations and their environment.

This is a crucial point to understand: the fact that an adaptive landscape has meaningful features (i.e. is not either flat or random) means that there *exists information that selection can use to discover higher fitness combinations*.

Can we study the nature of these features?

Yes, we can. One of the most important, though somewhat overlooked theoretical studies of these features was performed by Stuart Kauffman, and is described in his book The Origins of Order.

static1.squarespace.com/static/5657eb5…

The scope of his explorations is vast, and extends far beyond what I can relate here.

But it contains the best description of an Error Catastrophe I've come across, as well as outlines what Kauffman coins a Complexity Catastrophe, which as far as I'm aware was a novel discovery.

First, the Error Catastrophe (EC). To conceptualize an EC, imagine pouring a "slippery" substance upon an AL, such that it becomes more difficult to climb via selection, and also more difficult to maintain any advantageous position.

This feature emerges from replication error.

This implies that if an organism loses the capacity to replicate itself with high fidelity, it cannot retain whatever beneficial mutations it might have obtained via an Adaptive Walk.

It becomes a leaky container with respect to adaptively valuable information, so to speak.

Consider this a lower bound for evolutionary processes. Remaining above the EC threshold is adaptive table stakes.

But what about an upper bound? Does that concept even make sense?

This is where the Complexity Catastrophe enters our story...

To study ALs, Kauffman created a framework called the NK-tunable Adaptive Landscape, which gave rise to the theoretical prediction of CCs upon ALs.

In this framework N represented the number of mutational loci upon a gene sequence. e.g. for a sequence "ACTG", n=4.

Next, "K" represents the number of loci *affected by mutations to an average loci*.

In other words: if one A becomes G, how many other letters need to update their state?

Here's a compelling visual metaphor for the idea, apropos of our times.

As the video above demonstrates, an increase in K results in chain reactions that make the landscape quite chaotic / random.

Yet the evolutionary reality is such that each mousetrap automatically "reloads", so that such a chaotic process can continue indefinitely.

And this chaos cuts to the heart of the Complexity Catastrophe: increasing the capacity of the average component within a system to causally impose upon components beyond a certain threshold will invariably "melt" the entire AL into a sea of constant change.

Unlike the EC, where an organism can gain information from the AL, but the AL becomes slippery because the organism's "adaptive information container" is leaky, the CC *precludes the possibility of an organism gaining any adaptively stable information whatsoever*.

Think of it this way: if you stand on a smooth slope blindfolded, your immediate environment contains information about which way is generally "up".

But if sheer walls and steep falls surround you randomly, your position provides no actionable information.

Hence NK "tunable".

With these basics out of the way, we can finally turn our attention to our initial question: "why does it seem like we're getting dumber as a society despite this amazing tool for information sharing?".

The answer should now be somewhat obvious: we've landed ourselves in a CC.

This is to say that the specific networks that characterize our present Adaptive Walk have optimized for attentional engagement, and this has led them to a strategy that locally benefits from the chaos of the ping pong ball effect of open (high-K) networks of connectivity.

Twitter and Facebook (among others) turned the ping pong ball chaos into a kind of economic fission reaction capable of free energy extraction in the form of shareholder returns, all the while *completely melting* the AL upon which all prior institutions and processes evolved.

For those versed the theories of ALs, AWs, ECs, and CCs, the looming consequences were obvious: as you melt the historical AL, the sensemaking and agentic capacities of all prior structures will dissolve, given that there can no longer exist the required informational footholds.

And so we find ourselves somewhat trapped.

We stand atop a local optima in which our networks optimize for an outcome (attentional engagement) that fundamentally precludes large-scale sensemaking and selective evolution at the emergent scale of society...

Thus it's not so much that we have become dumber, so much as it is that we've created a context in which it's essentially impossible for any large scale system to gain and retain the very information upon which adaptive evolution depends.

We've suffocated our adaptive capacity.

What to do, then?

Well, the answer is obvious, despite the path to its realization being quite difficult:

We must create far more robust mechanisms of semi-permeability that insulate us from one another, contrary to the utopian naivete of the Religion of Openness.

Openness is not a virtue in and of itself. It is a feature of ALs that allows for the flow of information, both adaptive and maladaptive.

But if you set the dial on openness to its maximum, you will simply preclude evolution itself.

Thus, the need for membrane-aware protocols.

It is only through introducing an awareness of these evolutionary realities at the lowest levels of our informational stack that we will revivify the emergence of meaningful ALs.

And it is only through ensuring they're structured bottom-up that we may prevent their manipulation.

That is to say, so long as our economic incentives trap lock us into a trajectory that optimizes for the creation of chaotic open interactions as a fissionable substrate for profit, we will remain incapable of democratic sensemaking in the 21st century.

Of course, we're seeing an intuitive reaction to these realities in which individuals begin to pull away from these liquifying networks, into their own subsets of smaller bulkheads.

Yet such an extreme reaction carries with it the risk of the Error Catastrophe.

Specifically: secluding ourselves within entirely insulated subnets re-introduces the difficulty of replicating adaptive discoveries across the entire network, or even within a given community.

It precludes knowledge transfer, and lowers redundancy.

This is why such change must occur at the lowest levels of the stack, and why projects such as Holochain and others working toward fundamentally novel network topologies interest me, and why talented minds should focus on these problems, not short term profit maximization.

Because profits, as symbols of relative value creation, truly cease to matter if we liquify the fundamental structures upon which our cultural evolution depends.

In other words, captains still sink alongside their ships.

And unfortunately, we live in a world that has allowed said captains to defer this reality, as well as their responsibilities in light of it, and this deferral has led to the liquefaction of governance systems upon which our survival depends, as recent world events demonstrate...

But it's possible for those who see the depth of the problems we face, and who possess the capacity to contribute to building something better, to shift their purpose in life toward this end, even if they might not live to see its realization.

These are the seeds we must plant.