Accretive Growth Logics

I made up a term: Accretive Robotics. Robotics driven by accretive growth logics, as opposed to organic growth logics.

Two examples, both from cartoons (I overindex on cartoons clearly). First: Pickle Rick from Rick and Morty, where Rick starts out by turning himself into a pickle and then gradually adds more capabilities, such as by killing a cockroach and a rat and taking their body parts.

Second: The Akira-inspired South Park trapper-keeper monster, in which Cartman’s trapper-keeper (a kind of pencil case) grows by swallowing all sorts of devices and gadgets.

In both cases, a seed of partial organizing logic embodied by a primitive physical element (a pickle and a trapper-keeper respectively) grows inorganically, through improvised accretion, via a somewhat chaotic architectural scheme, into a much more capable embodiment: an accretive robot.

Despite the resemblance, an accretive robot is not the same thing as what in software architecture is known as a big ball of mud. Big balls of mud are the result of organic growth logics going wrong and stalling out due to insufficiently thoughtful organization. Accretive growth is marked by ongoing incorporation of bits and pieces into an improvised, emergent architecture that has a small, conceptually coherent kernel and a large, wild shell. It is the material-embodiment analogue to the AI/big data principle of “simple code and lots of data beats complex code and little data.” Mutatis mutandis: simple chassis and lots of scavenging beats complex chassis and little scavenging.

The main ongoing architectural task in accretive growth is expanding the range of things that can be “assimilated” into the Borg-like core, and shrinking the range of what must be rejected as incompatible.

Accretive growth logics violate our intuitions because all our growth intuition is based on organic, smooth growth. Even where there are discontinuous changes via “level ups,” they tend to fit into a harmonious ontogenic map of the growth future that is “continuous” at some meta level. The logic of the entire map is contained in a nearly complete ab initio architectural-information seed with genetic characteristics; ie the thing starts with meaningfully complete “DNA.”

Accretive growth on the other hand, continuously absorbs not just raw material and growth feedstocks from the environment, but organizing logics via a sort of Lamarckian or epigenetic capability. In the most extreme case, there is zero DNA to start. It’s all accreted during the growth process.

Organic growth only “eats” things at lower abstraction levels (usually several levels lower). Animals eat other animals and break them down into very low-level reusable parts (amino acids, fat molecules etc). They do not bolt on organs and such from their prey. But in accretive growth, you use the input at the highest level of abstraction you can make use of, often comparable to your own. To my knowledge, not counting such degenerate cases as hermit crabs, humans are the only organisms that can do some biological accretive growth through organ transplants.

But in all our design impulses, we are governed by an organic growth bias, and eschew accretive growth. I’ve been trying to develop the opposite bias in myself. In my own robotics tinkering, a vision of this sort of robotics, driven by junkyard scavenging, has emerged as my personal true north.

Speaking of hermit crabs, last year, on behalf of the Yak Rover project, I entered this design sketch for a hermit-crab robot into the Natural Robotics contest (which is open for entries again for the new year). The prize was that they’d actually try to build the design. Sadly, I didn’t win, but did get a special mention.

In corporations, accretive growth maps to growth via acquisitions rather than internally generated scaling of in-house innovations. There are huge but obscure companies that are really good at this, such as Illinois Tool Works (ITW).

At a lower level, accretive growth maps to being able to tolerate higher levels of “buy” in the build-buy tradeoff curve, by being really good at system integration. Again, there are huge but obscure companies that are really good at this (offhand I can’t think of the best example though, which itself is revealing).

In the Yak Rover project, my collaborators and I initially started out with a few pure-build and pure-buy rovers, but are increasingly starting to explore accretive design spaces, where rovers come together in ways that look more like Pickle Rick than Perseverance. I myself started out wanting to design and build an entire rover from scratch, with as few off-the-shelf parts as possible. I got reasonably far:

But I have now swung to the other extreme. I’d like a Pickle Rick rover: as scavenged and accretively grown as possible. I don’t yet know what the hell that means. Ideally, I’m imagining something like a Raspberry Pi board running an LMM that somehow magically picks up bits and pieces from the environment to build a body around itself. It’s an extreme thought experiment, and a fun one. Especially if you, like me, have one eye on how far AIs can get by themselves, starting with just LxMs running in regular computers. The trapper-keeper Akira monster is one heterodox vision of a Skynet-type AI. Ironically, despite the fact that it looks incoherent, it’s actually much more conceptually coherent than the idea of an “AGI” (which is based entirely on flimsy angels-on-pinheads organic growth logic). The trapper-keeper monster is “general” in a very specific sense — it has omnivorous assimilation capacity. It can eat and system-integrate almost anything. Almost nothing needs to be rejected as incompatible.

There is a more substantial difference too. Because accretive growth absorbs things at the highest level of tolerable abstraction, it absorbs a lot of information, including information about goals, values, beliefs, and so on. So thought experiments like “paperclip maximizer” don’t work with accretive visions. Goals themselves will evolve as chaotically as the body and mind of an accretively growing entity, and will matter much less. In fact, the more I think about complex, large scale systems, the more I realize “goals” are a very unimportant feature of their behavioral profile. Accretive growth logics prioritize the next round of growth, self-perpetuation, and survival, not long-term goals.

Think of a modified version of the Ship of Theseus thought experiment: if it evolves in accretive ways, it might actually transform from a ship into an airplane along the way, and ship-like persistent goals will not make sense. The embodiment is the medium is the message. Who you are shapes what you want to pursue, and what you eat next to take a next step reshapes who you are and what you want to pursue.

Going back to organizations, accretive growth logic really screws up Silicon-Valley style startup doctrines in particular, which have an uncritical faith in pure-paradigm organic growth. And as with all pure paradigm doctrines, they do contain a seed of the opposite. In the yin of the startup theology of organic growth, you find a seed of the yang of accretive growth logics: Paul Graham’s well-known advice to “do things that don’t scale.”

This line has always bothered me as being much too limited and unnecessarily framed as anomalous. I captured my misgivings with a series of casts on farcaster recently, which I’m expanding on below.

Cap: “now might be a good time to do startuppy things that don’t scale!”

Banner: “That’s my secret cap, I’m always doing things that don’t scale”

Paul Graham’s advice implies that there is something unnatural about doing things that don’t scale, and that success involves at some point switching to doing mostly “scalable things.” At which point, presumably you switch from “startup” to “grown-up” company.

The truth is, at every level n you’re doing level >n things in unscalable ways, and the proportion of unscalable things does not necessarily converge monotonically to zero. In fact, it almost never does. And every time you replace an unscalable behavior with a scalable one, you usually require a new unscalable behavior at another locus. Sometimes more than one. The entropic chaos of unscalability can keep increasing longer than you can remain architecturally sane, if you’re attached to organic growth logics.

I think Silicon Valley has this bias towards organic growth because it’s been riding Moore’s Law (which is inorganic and accretive growth in terms of its internal logic, but looks like an organic-growth boundary condition to its “customers”) for so long. Higher-order “organic” phenomena like network effects ultimately rest on Moore’s Law. I think that era is ending. Accretive growth logics are starting to dominate once more.

My reference data set is almost the exact opposite of Paul Graham’s. Almost all my consulting has been somewhere between Series-B-sized companies to big old corps, and in a few cases, government agencies. Mostly skewing larger, and all with a big chunk of growth vectors being accretive rather than organic. I’ve never seen an org that isn’t doing strategically critical things that don’t scale within its current operational structure. Startups aren’t special. They’re not even special in the sense of being the “starting point” initial condition. It’s just an arbitrary scaling level marked by formal incorporation and taking-on of formal investors. There are like 7 scaling levels before “startup”: paper napkin, weekend prototype, sabbatical hack, friendly collaboration, friends-and-family round, kickstarter campaign, pre-seed…

In fact you can define corporate wind-downs (opposite of startups) as hitting a level where there is no way to restructure what you’re doing in a scalable way for a more ambitious next chapter. And as with startups, formal wind-down of a corporate structure isn’t the “end” of the journey. Bits and pieces survive to be accretively incorporated into other entities, with various levels of breakdown and organ-transplant activity. Often with a lot of midwifery by the private equity sector (selling old companies for parts is part of the playbook there; the bigger the chunk you can salvage and sell, the higher the value).

A generalization of the idea is that at any given scale, an organization can be factored into:

• a) a scalable part that still has headroom for growth
• b) a scalable part at its limit (ossified)
• c) a part that doesn’t scale and will require the invention of a new scheme of organization to sustain past an urgent near-term horizon.

Normally we think things should have a proportion a > b > c, but this is not necessarily true. When accretive growth logics dominate, you can have c > b > a. A persistent SNAFU equilibrium that to organic-growth purists looks like a dysfunctional broken mess that has no right to persist and resist disruption, but does.

In fact, it’s not a good idea to think about accretive growth in terms of sequential or circular stage-growth schema like roadmaps and lifecycles. Accretive growth often has temporal messiness, with a lot of out-of-order and out-of-place growth elements that gradually shake themselves into the right spatio-temporal place-times. The histories of accretive growth phenomena are necessarily non-sequential.

A better way to think about accretively growing things is in terms of levels of disorganization. Pure organic growth, ideally managed, should exhibit strong and continuous organization throughout its (very sequentially staged) life. Pure inorganic growth should be like disorganized chaos all the way, with doubts about its entityhood along the way. Is it even a thing at all, or just a rolling knot of entanglement in other things?

Real things are always somewhere in between. There’s a legibly organized, sequentially staged part, and a disorganized, out-of-order, out-of-place part. That’s partly a mark of chaotic accretive growth, and partly a mark of structural “depression” of things being unmotivated due to being out of place/time in the current legibly organized scheme of things (notably, legibly organized growth kernels, if they exist, often try to label the illegibly accretive shell as deadweight and get rid of it, out of sheer authoritarian high-modernist anxiety rather than an actual case for elimination).

Scaling with accretive growth logics is much harder than scaling with organic growth logics. It takes conscious intelligence and more active steering to do. Very simple creatures can grow organically. It takes human intelligence to invent organ transplants that work. Dumb, unmanaged accretive growth isn’t a thing.

In the economic sphere, people are suspicious of accretive growth logic because historically it has been associated with inefficient command-economy patterns of organization. But the examples of post-war Germany, Japan, and Korea show that such patterns, once you get away from the Italian-corporatist or Soviet-style extremes, can grow quite big and powerful.

Organic growth logics are systematically over-rated, and over-credited with the success of real things. Accretive growth logics are systematically under-rated, and under-credited.

Subsets of accretive growth logics, rejoicing under politically loaded banners like “composability,” “modularity,” and “repairability,” enjoy some modest subcultural cachet, but do not seriously challenge the cultural dominance of organic growth logics. But I increasingly find that I like pushing the case for the fullest forms of accretive growth logics, which do not have the conceptual coherence and elegant degeneracy of limit-case concepts like composability, modularity, and repairability. If you champion Lego, but not Akira-trapper-keeper monsters and Pickle Rick robots, you’re not really grokking the idea of accretive growth logics. Instead, you’re stuck in a different kind of organic growth regime (of standard interfaces and open protocols for example).

Accretive growth entities of the most complex variety, such as large, messy liberal democracies like the US and India (for the time being), are acutely anxiety provoking to people with strong disgust responses and a yearning for aesthetic coherence. But a thing can have extreme architectural coherence, resilience, and functional capability, without having aesthetic coherence. In fact, arguably, beauty is actually a sign of all sorts of bad things: architectural degeneracy (unused dimensionality) fragility (monoculture vulnerabilities), and functional limitation (limited openness and interoperability within walled gardens). They also breed intolerance and resistance to pluralism.

I suppose you could call this post a critique of pure organic growth.

Accretive growth logics tend to wax in popularity as organic growth logics cycle through phases of over-reach and under-performance, as I believe is happening right now. But accretive growth logics are rarely championed directly, on their own merits. They tend to fill in the vacuums left behind by failing charismatic organic growth ideologies that overpromised and under-delivered. They tend to be negatively defined by the absence of organic growth logics. By the lack of Steve-Jobs style benevolent dictators for life (BDFLs) at the heart of absorbing charismatic theaters of visionary evolution.

A rare counterexample was the brief and somewhat faddish era of microservices and polyglot persistence in cloud architectures in the last decade (which seems to have run into trouble and waned somewhat in popularity lately). That was people taking accretive growth logics seriously.