Books : reviews

Manuel De Landa.
War in the Age of Intelligent Machines.
Zone Books. 1991

Manuel De Landa.
A Thousand Years of Nonlinear History.
Zone Books. 1997

rating : 2 : great stuff
review : 19 September 2010

Amazon kept recommending this book to me, for some reason I couldn't fathom. After all, I don't think I've bought a single history book from them. The cover is garish enough that I noticed they way it kept popping up (and later I found the internal typography to be execrable: san serif like a pre-school Janet and John book, and ragged right margins obscuring the paragraph structure; but I digress). After a while I looked at the recommendation in a bit more depth than merely reading the title. It looked as if it might be interesting, so in June I bought it, as is my wont, and added it to my ever-growing unread collection. And Amazon still kept recommending it! So I looked a bit closer, and eventually decided to read it. 50 pages in, I went back to Amazon, and bought all the rest of De Landa's books. (And later, I bought a book by another author referenced in the notes. Reading good books makes my unread pile grow.)

Amazon's recommendation was right for me because this isn't "history" in the traditional sense. De Landa discusses the last 1000 years of (mainly European) history as intensifications of flows of energy, material, genes, and language through a combination of non-equilibrium complex dynamical systems. He provides a tour de force, that synthesises and builds on the work of a huge range of authors from a diverse range of backgrounds: geology, biology, sociology, economics, linguistics, philosophy, and more. (One author he builds on in particular is the French philosopher Deleuze. After reading the numerous quotations given, I am not tempted to buy his works, but rather am thankful that De Landa has provided such an excellent translation and synthesis as presented here.)

Below I try to provide an overview of the book. Warning -- this is long, since there is so much covered in the book -- and it is incomplete -- for the same reason. The book is a complex and detailed meshwork of an account, and to be fully appreciated, I highly recommend reading it in full. I have attempted a fairly linear account, but this is doomed to failure. De Landa emphasises that we find it difficult to think in non-linear ways:

p67. We may wonder why, given the ubiquity of self-consistent aggregates, it seems so hard to think about the structures that populate the world in any but hierarchical terms. One possible answer is that stratified structures involve the simplest form of causal relations, simple arrows going from cause to effect. … Western thought has been dominated by notions of linear (nonreciprocal) causality for twenty-five hundred years. It was not until World War II that the work of Norman Wiener … gave rise to the study of negative feedback and with it the beginning of nonlinear thinking.

The whole work is embedded in the notions of complex non-linear dynamics, at multiple levels.

p25. We live in a world populated by structures---a complex mixture of geological, biological, social, and linguistic constructions that are nothing but accumulations of materials shaped and hardened by history.

It is about themes and patterns: there is barely the mention of any specific historical characters or events (no specific wars, for example, but a lot about the effect that preparation for war had on nations). It starts a thousand years ago, at the "first industrial revolution", of agriculture and water power, with a break at the Industrial Revolution, of steam power.

p34. eleventh century industrial revolution, fueled by solar (agricultural) and gravitational (water) energy.

He does this in terms of dynamical systems concepts such as "attractors" and "bifurcations" (he never goes into specific details of the underlying dynamical systems formalism, but it is clear he means this concretely, not metaphorically).

p263. Attractors are the simplest type of abstract machine, operating at the level of nonlinear, destratified flows.

The attractors define "stable" states in the system, which can move between them.

p15. much as a given chemical compound … may exist in several distinct states … and may switch from stable state to stable state at critical points in the intensity of temperature (called phase transitions), so a human society may be seen as a "material" capable of undergoing these changes of state as it reaches critical mass in terms of density of settlement, amount of energy consumed, or even intensity of interaction.

(Note here it's "matter-energy", not "matter-energy-information", but information and its flows do play a role later.)

p21. reality is a single matter-energy undergoing phase transitions of various kinds, with each new layer of accumulated "stuff" simply enriching the reservoir of nonlinear dynamics and nonlinear combinatorics available for the generation of novel structures and processes.

Because of this openness---flows through the far-from equilibrium-system---and this growth---of material, of kinds of material, and hence of the dynamical system's phase space and dimensionality---conventional non-linear dynamical systems theory is inadequate. It deals with fixed phase spaces, not ones constantly being "enriched" by novel materials and processes. But since De Landa only uses high level concepts of attractors and bifurcations, this does not pose a problem (at this level of discourse).

De Landa moves beyond these dynamical systems ideas of attractors and bifurcations to ideas of "structure-generating processes" that change the very dynamical system's landscape. He identifies three further "abstract machines" that are important here: those that create hierarchies, those that create meshworks, and evolution. The idea of an abstract machine that can have different realisations is an important concept. All too often writers take structures in one domain and translate them metaphorically (or via only a naive analogy), merely "copying the form" into another domain. This just doesn't work (beyond possibly some trivial results), and can all too easily result in "cargo cult" reasoning.

p28. We must be careful when drawing these analogies, however. In particular, we must avoid the error of comparing cities to organisms, especially when the metaphor is meant to imply (as it has in the past) that both exist in a state of internal equilibrium, or homeostasis. Rather, urban centers and living creatures must be seen as different dynamical systems operating far from equilibrium, that is, traversed by more or less intense flows of matter-energy that provoke their unique metamorphoses.

p281, note 11. Mirowsky shows how the concept of the "invisible hand" was formalized in the nineteenth century by simply copying the form of equilibrium thermodynamics. ... Elsewhere he warns that recent attempts to apply Ilya Prigogine's theories to economics are making the same mistake---for example, assuming the existence of attractors without specifying just what it is that is being dissipated (i.e., only energetically dissipative or "lossy" systems have attractors).

But getting a good abstraction, then reinstantiating appropriately in a different domain, give much more powerful results. It tells you which parts are contingent, and which are essential to the process. For example, it is important to have some realisation of the abstract concept of dissipation, or "friction", in the system to get the attractor dynamics. (Frank Herbert at least recognises the necessity of "friction" in society, with his invention of BuSab: the Bureau of Sabotage.)

p41. Contemporary studies in nonlinear urban dynamics teach us that, in many cases, friction (delays, bottlenecks, conflict, uneven distribution of resources) plays a crucial role in generating self-organization. Hence, eliminating it from our models (by postulating an optimizing rationality, for instance) automatically eliminates the possibility of capturing any real dynamical effect.

De Landa insists his abstract machines are genuine abstractions that can be applied non-metaphorically across domains.

p58. There is, no doubt, some element of metaphor in my use of the terms [hierarchy and meshwork], but there are, I believe, common physical processes behind the formation of meshworks and hierarchies which make each different usage of the terms quite literal. These common processes cannot be fully captured through linguistic representations alone; we need to employ something along the lines of engineering diagrams to specify them.

No actual diagrams are drawn, but several are described. The first abstract structure generating process discussed is that of sedimentation, of making strata, or hierarchies. Initially this is examined in terms of sedimentary rock formation, and then abstracted, and later applied to other systems.

p61. Since pebbles do not come in standard sizes and shapes, some kind of sorting mechanism must be involved here, some specific device to take a multiplicity of pebbles of heterogeneous qualities and distribute them into more or less uniform layers. …
     ... a second operation is necessary to transform these loose collections of pebbles into a larger-scale entity: sedimentary rock. This operation consists in cementing the sorted components together into a new entity with emergent properties of its own, that is, properties such as overall strength and permeability which cannot be ascribed to the sum of the individual pebbles. …. it consolidates the pebbles' temporary spatial relations into a more or less permanent "architectonic" structure.
     Thus ... a "double articulation" transforms structures on one scale into structures on another scale. … these two operations constitute an engineering diagram and so we can expect to find isomorphic processes (that is, this same "abstract machine of stratification") not only in the world of geology but in the organic and human worlds as well.

This abstraction (the structure generating process, and the resulting structure) is applied across domains.

p62. sedimentary rocks, species, and social classes ...are all historical constructions, the product of definite structure-generating processes that take as their starting point a heterogeneous collection of raw materials (pebbles, genes, roles), homogenize them through a sorting operation, and then consolidate the resulting uniform groupings into a more permanent state. The hierarchies ... are a special case of a more general class of structures, stratified systems, to which not only human bureaucracies and biological species belong, but also sedimentary rocks. (And all this without metaphor.)

A metaphor would be something like "sedimentary rocks have layers, social castes have layers, therefore castes are like rocks", but here we have the abstraction "sedimentary rocks are formed by these processes, we can abstract these processes into "sorting" and "consolidation", we can see instances of sorting and consolidation processes in other systems", but those need not be instantiated in the same way as the sorting and consolidation in rocks: social systems don't use gravity to sort people into classes!

The consolidation step is a crucial part of the process, as it slows down the flow, produces a new kind of entity, by stopping the components from just drifting apart again. The dynamics is irreversible at a fundamental level.

p268. the cementing together of the sediment … acting as a ratchet mechanism

In fact, there's an even more abstract machine, of which this "sorting and cementing" is itself an instantiation:

p290, note 82. Deleuze and Guattari call these two operations "content" and "expression" and warn us against confusing them with the old philosophical distinction between substance and form. Content and expression each involves substance and form: sedimentation is not just about accumulating pebbles (substance) but also about sorting them into uniform layers (form); while consolidation not only effects new architectonic couplings between pebbles (form) but also yields a new entity, a sedimentary rock (substance). It is this form of the diagram (one operation involving substances and forms, another operation involving forms and substances) that is the most abstract and, hence, the most useful. The particular instantiation that I will be using in this book (sorting + consolidation) may be seen as a particular form of this more general diagram.

"Substance and form" corresponds to "stuff and relationships", which might be drawn as a network of "nodes and links". In complex systems, we are used to thinking of particular complex networks, such as "small world networks". But of course a hierarchy is also a kind of network: a tree structure. The kinds of more horizontally structured networks we think of as "complex systems" De Landa calls meshworks, and they form his second abstract machine.

p64. there are three elements in this diagram. First, a set of heterogeneous elements is brought together via … an interconnection of diverse but overlapping elements. … Second, a special class of operators, or intercalary elements, is needed to effect these interconnections. (... this is the role played by catalysts...) Finally, the interlocked heterogeneities must be capable of endogenously generating stable patterns of behavior (for example, patterns at regular temporal or spatial intervals).

Moving from the domain of sedimentary rocks to that of markets, one intercalary element identified is money, as it serves to interconnect various transactions.

p66. In the absence of price manipulation, money … functions as an intercalary element ... Other intercalary elements are also needed to make markets work. … constraints … to reduce transaction costs and allow the interlocking of complementary demands to take place.

Catalysis plays a key role in meshwork formation. What is meant by catalysis? It can't be just the chemical definition: that is tied to the particular realisation. Again, we need an abstraction.

p291, note 92. Deleuze and Guattari … constantly refer to catalysis in their theories of meshworklike structures … They tend to view catalysis in terms of one specific (albeit very important) type of catalyst: the allosteric enzymes …
     What is needed here is to make the notion of a "catalyst" more abstract so that the specific functions of a chemical catalyst (to perform acts of recognition via a lock-and-key mechanism, to accelerate or decelerate chemical reactions) are not what matters, but the more general notion of aiding growth "from within" or "from in between." One step in this direction has been taken by Arthur Iberall … by defining catalytic activity as the ability to force a dynamical system from one attractor to another.

So money can be a catalyst.

p35. Money … is like energy, only it runs in the opposite direction: energy flows from agricultural villages to the towns they feed, while money flows from town to countryside, to pay for the food. … in many cases peasants would come to a market town to sell their goods, not to buy other goods, but to get cash to pay their rent to the owners of their land. … monetary flows regulate (inhibit or intensify) energy flows ...
     Money is best defined as a catalyst or stimulant of trade (and its absence, an inhibitor). Barter, the exchange of goods for goods, is relatively inefficient in that people must wait for their complementary needs to meet. … once … money comes into existence, complementary demands can be meshed together at a distance, greatly increasing the intensity of market exchanges.

Here money is explicitly compared to (negative) energy flow, rather than say being an embodiment of information. But like energy flows, money flow can be intensified: it be stored and accumulated (and restricted, to provide "friction"). If we think of money as information, then this implies that information can be a catalyst. Another example is given, this time directly informational (despite "know-how" being embodied in people):

p79. Skills and know-how provided what one might call "catalytic information," that is, information capable of bringing together and amplifying flows of energy and materials.

Abstraction mitigates the danger of "cargo cultism". If certain features or properties in one instantiation are not part of the abstraction, there no reason why those features or properties hold in a different instantiation.

p65. much as hierarchies (organic or social) are special cases of a more abstract class, strata, so autocatalytic loops are special cases of self-consistent aggregates. And much as strata are defined as an articulation of homogeneous elements, which neither excludes nor requires the specific features of hierarchies (such as having a chain of command), so self-consistent aggregates are defined by their articulation of heterogeneous elements, which neither excludes nor requires the specific features of autocatalytic loops (such as growth by drift or internal autonomy).

Systems aren't simple instantiations of one or the other of these abstractions, either hierarchies or meshworks, but are truly complex:

p32. self-organized meshworks of diverse elements, versus hierarchies of uniform elements. … meshworks and hierarchies not only coexist and intermingle, they constantly give rise to one another. … hybrid form[s]: a hierarchy of meshworks … a meshwork of hierarchies

So systems have multiple levels and scales of interlocking hierarchies and meshworks. The dynamics within and between these levels is important. The emphasis is that history is understood not in terms of individual people and events, but in terms of dynamical flows, trends, and processes.

p43. to understand the role of decision making in the creation of social order, we need to concentrate … on the dynamics … among many interacting decision makers. The hierarchies and meshworks that develop from these interactions … in turn become elements of other homogeneous and heterogeneous structures .... At each level, different nonlinear dynamics take place, with their own multiple equilibriums and bifurcations between alternative stable states. Hence, individual decision making, while important, is simply one element in the mix, interacting and influencing dynamics on only one of a number of scales.

However, non-linear dynamics permits small causes to have large consequences. If a higher level is near a bifurcation point, a small lower level action can affect it.

p286, note 48. There are several exceptions to this "rule" (individual decision making affects only one level of scale). One of them involves special situations where the level immediately higher ... is near a bifurcation point in its own dynamics. Here the decisions and actions of individuals may be amplified and have effects beyond their scale.

But even so, we still need to understand how that higher level got to, and was maintained near, that bifurcation point.

p147. A command given by someone of high rank in a hierarchy, for example, can set off disproportionately large flows of energy, as in the case of a declaration of war. However, the military order itself is powerless unless backed up by a chain of command that has been kept in working order through constant drill and discipline …, all of which involves enormous expenditures of bodily energy.

The third abstract machine is evolution. This is more readily acceptable as being an abstract machine, because of the well-know non-Darwinian instantiations, such as memes and evolutionary algorithms.

pp138-9. the two abstract machines … (one generating hierarchies, the other meshworks) are adequate to account for living structures ... there is another abstract machine involved in the production of biological entities which has no counterpart in the geological world, therefore distinguishing species from sedimentary rocks. ...
      Darwin's basic insight was that animal and plant species are the cumulative result of a process of descent with modification. Later on, however, scientists came to realize that any variable replicator (not just genetic replicators) coupled to any sorting device (not just ecological selection pressures) would generate a capacity for evolution. ... Birdsongs are the most thoroughly studied example of these
[other kinds of] replicators ("memes," as Dawkins calls them) ...
      … the coupling of variable replicators with a selection pressure results in a kind of "searching device" (or "probe head") that explores a space of possible forms … This probe head is the abstract machine we were looking for ... although the new machine is characteristic of life-forms, the same basic diagram applies to memes and genetic algorithms. It would be incorrect to say that evolutionary concepts are used metaphorically when applied to computer programs and birdsongs, but literally when talking about genes. …. Hence, it does not constitute an "essence" of life, in the sense of being that which makes life what it is.
      The flow of genes through replication is indeed only a part of what life is. The other part is constituted by the flow of biomass.

So evolution isn't the essence of life, because non-living things evolve. Life requires both evolution and metabolism. (One might speculate that engineered life doesn't need evolution. Unless the evolutionary "probe head" is the only feasible way of finding/constructing life-like regions of the state space?)

The evolutionary probe head isn't blindly wandering around an unstructured space: the space has structure because of the underlying dynamics, and the process can exploit this.

p140. As with any physical system, the intense flow of energy moving through an ecosystem pushes it far from equilibrium and endows it with the ability to generate its own dynamic stable states (attractors). ... the space that the probe head blindly explores is not completely unstructured but already populated by various types of stable states (static, cyclical, chaotic, autopoietic). This prestructuring of the search space by intensifications of the energy flow may indeed facilitate the job of the abstract machine (blind as it is). For example, since one possible endogenously generated stable state is a periodic attractor, which would automatically draw gene activity and gene products into a cycle, the searching device may have stumbled upon the means to generate a primitive metabolism very early on. Further evolutionary complexification may have been achieved as the probe head moved from attractor to attractor, like so many stepping-stones.

This kind of meshing everything together, the interactions and comlpexifications of all the different components discussed, is typical of this book.

pp146-7. we have been exploring exclusively the interactions between culture and genetics, but nonetheless we must never lose sight of the fact that the flow of replicators … constitutes only half the story. The flow of matter and energy through a system (which often means the flow of biomass, either living or fossil) is of equal importance, particularly during intensifications. The role of genetic and cultural replicators (or, more accurately, of the phenotypic effects of those replicators) is to act as catalysts that facilitate or inhibit the self-organizing processes made possible by intense matter-energy flows. It is these flows that determine the nature of the thermodynamic stable states available to a system; the catalysts act merely as control mechanisms, choosing one stable state over another. Another feature of catalytic action is that low expenditures of energy can bring about high-energy transformations. …
      Cultural replicators may be viewed as having phenotypic effects similar to catalysis. ... we cannot be content with a description of society expressed exclusively in terms of replicators and their catalytic effects, but must always include the material and energetic processes that define the possible stable states available to a given social dynamic.

It this multi-level, multi-scale complexity, and the dynamism of the structure generating processes, that raises this work above a naive application of complex systems ideas. Indeed, relative to this model, most "complex" systems appear disturbingly simple: single kinds of entities, single kinds of network structure, no environment, no analogue of energy-matter flows. Here we have ever-growing complexity, as emergent properties become crystallised and reified, and themselves enter into the dynamics of the system.

The emphasis here isn't on some "final" static structure, because there isn't one; it's on the various structure generating processes, and how they are constrained.

pp80-81. the process of industrial takeoff may be viewed as a bifurcation, from a state in which self-stimulating dynamics were not complex enough to overcome diminishing returns, to a state in which the series of nodes forming the circuit became a self-sustaining entity. The addition of new nodes to the meshwork as it complexified did not occur according to a plan but simply following internal constraints; that is, each new node had to "mesh well" with the existing ones (i.e., catalyze and be catalyzed by existing nodes).

These processes aren't free. It takes input to drive them, and if there is insufficient driving force, the changes will falter, and the structures will not develop.

p78. autocatalytic loops need to achieve a threshold of complexity before they acquire the resilience and versatility needed to overcome diminishing returns.

This helps to explain why the Industrial Revolution occurred when and where it did, and just as importantly to explain why it didn't occur elsewhere, or at another time (eg earlier): "you can't railroad until it's time to railroad" (this point is also alluded to earlier, when discussing the "first" Industrial Revolution, in 11 century). The process needs more than one "intensification" to fuel a revolution: single intensifications tend to peter out after they exhaust their single resource, while true revolutions have a (meshwork of) multiple resources. And homogenisation does not help fuel revolutions.

Examples mentioned in this second Industrial Revolution include the railroad and the telegraph: essential infrastructure. Russell J Abbott, in his paper "Emergence Explained", has argued that one should build infrastructure, rather than closed and isolated functionality, since the infrastructure can "serve as a possible basis for systems yet to come", and can support diverse functionality that you never even thought of originally. Here we see that infrastructure can provide the intercalary elements needed to help bind the meshwork together.

p50. The emergence of powerful nation-states, and the concomitant decrease in autonomy of the cities they absorbed … could have brought the different forms of self-stimulating dynamics … to a halt. That this did not happen was due to yet one more form of autocatalysis unique to the West: continued arms races. … the nations of Europe, unlike China or Islam, were never able to form a single, homogeneous empire, and have remained until today a meshwork of hierarchies. It was within this meshwork that advances in offensive weaponry stimulated innovations in defense technology, leading to an ever-growing armament spiral

Despite this feeling of inevitability, however, it would have been interesting to have a discussion on how much role contingency plays. Is it all down to European topography (mountains separating regions), its relative smallness (relative to the other continents, that is) resulting in relatively long coastline, and everywhere being relatively close to the sea? There is the discussion (above) about military orders having large effects (when systems are near bifurcations point), but nothing about how things might have been different. As I've already mentioned, there is very little here about specific characters or specific events. But certain decisions, such as whether to locate a capital city inland or on the coast, clearly had important consequences. What if the tape were played again? How much would things have had to be different to end up with a qualitatively different history? Which is more important, butterflies or canalisation?

De Landa is careful to emphasise that these structure generating processes do not result in a notion of "progress" as such. But the dynamics does result in enrichment, greater complexity, more crystallised emergent "stuff" to use in the next iteration/timestep. This might not be "progress", but it certainly has a direction; it is somehow "more developed".

Homogenisation, although it does not help to fuel revolutions, is a recurrent theme in this ongoing developmental process.

p84. the process of routinization of production in arsenals, mines, and civilian factories underwent a great intensification on both sides of the Atlantic, and this implied a large increase in the command element in the economic mix.

One consequence of routinisation is the reduction in the importance of information (as in formal skills and informal know-how). The information has been transformed from being held by individuals to being located in the "corporate memory" of documented processes and procedures. Quoting Hohenberg and Lees:

p86. "The nature of information as an input to production is that it ceases to be important once a given process becomes routine. At that point other costs---for machines, basic labor, and space---take over…"

One advantage of homogenisation can provide is economies of scale. But homogenisation isn't the only route to this. Heterogeneous "economies of agglomeration" also exist. These still require a certain scale, of course, but of the environment (many small others), not of the self.

p85. there are other possible types of positive feedback for cities and towns, other connections between efficiency and size-not the size of a homogenized enterprise and its homogeneous mass-produced products, but the size of a highly heterogeneous urban center which provides small firms with a variety of mutually stimulating links. These are not economies of scale, but economies of agglomeration

Nevertheless, this homogenisation, or routinisation, of processes within the system is itself an important process: De Landa talks about it with respect to manufacturing processes, and even the training and "processing" of people. This description seems to have some relationship to the "progressive mechanization" in biological systems noted by Bertalanffy in his General System Theory. De Landa makes this link, but does not think the resulting "systems approach" is the last word in this respect.

p97. Maturana notes that one characteristic of autocatalytic loops is that their internal states determine most of their behavior, with external stimuli playing the role of triggers. He compares this to push-button machines whose behavior is not caused by the pushing of a button, only triggered by it. Automated factories are very complex push-button machines of this type and, as such, planned autocatalytic loops. Indeed, as late as the 1960s, a routinized, rationalized production process that generated economies of scale was thought by many to be the perfect example of a whole that is more than the sum of its parts. That so-called systems approach celebrated routinization as the crowning achievement of modern science. Today we know that planned loops of triggers and flows are only one of a number of systems that exhibit emergent properties, and that spontaneously generated loops may be more adaptive and resilient than rigidly planned ones.

This resilience comes about because of heterogeneity, not homogeneity.

p97. while economies of scale and economies of agglomeration, as forms of positive feedback, both promote growth, only the latter endows firms with the resilience and adaptability needed to cope with adverse economic conditions.

p107. heterogeneity endows these [ecosystem] meshworks not so much with stability (the capacity to maintain a state with relatively minor internal fluctuations) as with resilience (the capacity to absorb major external and internal fluctuations by switching between several alternative stable states).

p34. McNeill's hypothesis is that explosive, self-stimulating ("auto-catalytic") urban dynamics cannot emerge when hierarchical components overwhelm meshwork components.

In a world view without a notion of "progress", how can we bring about desirable change. In particular, how can we reduce heterogeneity, and produce heterogeneous meshworks to get this important resilience? De Landa warns that it it not simple -- it is not a matter of mixing things up, or destroying hierarchies.

p68. the increase in diversity that mutually stimulating loops bring about will be short-lived unless the heterogeneous elements … come to form a meshwork. As Maruyana writes, "There are two ways that heterogeneity may proceed: through localization and through interweaving. In localization the heterogeneity between localities increases, while each locality may remain or become homogenous. In interweaving, heterogeneity in each locality increases, while the difference between localities decreases." In other words, the danger with positive feedback is that the mere production of heterogeneity may result in isolationism (a high diversity of small cliques, each internally homogeneous). Hence the need for intercalary elements to aid in articulating this diversity without homogenization

In fact, radical destratification can make things worse.

p266. hypothesis: that the creation of novel hierarchical structures through restratification is performed by the most destratified element of the previous phase.

De Landa advocates change, but not grand sweeping changes, but smaller, tentative experiments.

pp272-3. To simply increase heterogeneity without articulating this diversity into a meshwork not only results in further conflict and friction, it rapidly creates a set of smaller, internally homogeneous nations. (Hence, the balkanization of the world would increase heterogeneity only in appearance.)
... … our world is governed not only by nonlinear dynamics, which makes detailed prediction and control impossible, but also by nonlinear combinatorics, which implies that the number of possible mixtures of meshwork and hierarchy, of command and market, of centralization and decentralization, are immense and that we simply cannot predict what the emergent properties of these myriad combinations will be. Thus the call for a more experimental attitude toward reality and for an increased awareness of the potential for self-organization inherent in even the humblest forms of matter-energy.

One such small experiment that a company can perform is to start a research laboratory.

p90. One way a corporate hierarchy may internalize knowledge is by funding a research laboratory. ... The [General Electric] lab, and the many that were later created in its image, may be viewed as an internalized meshwork of skills

What else could we do to effect change in this complex hierarchical meshwork that is today's world? De Landa here has a deep and rich model of our world, and it would be worthwhile using the concepts of interlocking meshworks and hierarchies, and the various abstract machines and structure generating processes, to develop a framework to think about truly complex systems.

Some more quotations that I found thought-provoking, but don't fit into the flow of the above review.

Colonies initially started off importing the more high-tech products from Europe. But some started developing their own alternatives.

p82. [in the USA] what mattered was the reservoir of interlocking skills and procedures generated by import-substitution dynamics.

That is, the colonists developed skills locally to manufactures things previously imported. These skills could then be exploited further, to do other things. This might possibly be an analogue of gene doubling, a genetic process that preserves the original functionality, whilst allowing innovation too, as the copy can change. Here, the colonies first had to first "unfreeze" the functionality, by replicating it locally, in order to be able to "double" it, to use it for other purposes too. Maybe there is a lesson here about outsourcing (the opposite process), inevitably leading to loss of a valuable source of internal innovation?

Further thoughts that might be related to ideas on infrastructure (in terms of pervasiveness, at least):

p91. The miniaturization of motors allowed the gradual replacement of a centralized engine by a multitude of decentralized ones (even individual tools could now be motorized). Motors began disappearing from view, weaving themselves into the very fabric of reality.

The same miniaturisation process is true for computers: mainframe to PC to ambient. ("Clouds" are different: they are more about raw compute power than making the whole environment "smarter".) Is this progression true for any miniaturisable tech?

More on the importance of flows, and thinking of intensifications of those flows:

pp104-5. living creatures and their inorganic counter-parts share a crucial dependence on intense flows of energy and materials. … Our organic bodies are … nothing but temporary coagulations in these flows ….
     Flesh, or "biomass," circulates continuously … giving the ecosystem its stability and resilience. … The foundation of any food web is its plants, which "bite" into the stream of solar radiation …. Plants are the only nonparasitic creatures in an ecosystem, its primary producers, while the animals who eat flesh (plant or animal) are mere consumers. The complex microflora and microfauna that process the ecosystem's waste are as important as plants, since these organisms remineralize and reinject dead plant and animal bodies back into the web. Compared to plants and microorganisms, "higher" animals are just fancy decorations in an ecosystem, consuming and transforming biomass with decreasing efficiency as their size increases.

Downward causation gets a look in (if not by that name):

p112. the crucial function of genes is to force individual molecules within a cell to obey the cell itself, and similarly for individual cells in a tissue, individual tissues in an organ, and individual organs in an organism. At each rank of the hierarchy, the genes' purpose is to constrain the lower level to behave in ways determined by the immediately upper level.

The third part of the book is about language. There is some intriguing speculation on a common underlying process for the evolution of natural language in terms of "combinatorial constraints", and of the more formal mathematics and less formal music:

pp222-3. by making the combinatorial constraints more rigid [than for language] we can generate strings of inscriptions like those belonging to systems of logic or mathematics, while by making them more flexible we can produce musical strings. … This is not to deny that explicit rules exist in mathematical or musical systems, much as they do in standardized languages. The question is whether mathematics or music could have originally developed as a decentralized system of constraints that only later was formalized as a centralized body of rules.

One important process in language generation is pidginisation. (De Landa argues that English itself underwent such a process after the Norman conquest.)

p236. simplification alone cannot account for the birth of (more or less) stable entities …. Hymes adds the requirements that the new, simplified pidgin be used by several groups (each with its own mother tongue distinct from the pidgin) ... the language being pidginized … must be absent both as a source of stigmatization and as a reference model. That is, the crystallization of a pidgin involves a barrier (geographical or social) that distances the emerging entity from the conservative tendencies of the prestigious target language. Only under these conditions can a pidgin achieve autonomy from the dominant norm, and it is this autonomy that defines it as a separate entity.

Note that this description, like most of the rest of the book, does not concentrate on specific contingent events and people, but rather on processes, trends and patterns.

The summary in the final part of the book brings together much of what has been covered, in a more condensed form. It possibly only makes sense if the preceding material has been digested, however.

pp258-60. It is almost as if every part of the mineral world could be defined simply by specifying its chemical composition and its speed of flow: very slow for rocks, faster for lava.
      Similarly, our individual bodies and minds are mere coagulations or decelerations in the flows of biomass, genes, memes, and norms. ... Over the millennia, it is the flow of bio mass through food webs, as well as the flow of genes through generations, that matters, not the bodies and species that emerge from these flows. ... And a similar point applies to our institutions, which may also be considered transitory hardenings in the flows of money, routines, and prestige, and, if they have acquired a permanent building to house them, in the mineral flows from which the construction materials derive.
      This book has concerned itself with a historical survey of these flows of "stuff," as well as with the hardenings themselves, since once they emerge they react back on the flows to constrain them in a variety of ways. .... On one hand, the flows of materials whose history we described involved more than just matter-energy. They also included information, understood … in dynamic terms, as patterns capable of self-replication and catalysis. That is, we have considered … the "phenotypic" effects of these replicators, their ability to trigger intensifications or diminutions in the flows of matter-energy and their ability to switch from one stable state to another the structures that emerge out of these flows. On the other hand, among these structures we distinguished coagulations that have undergone a process of homogenization, which we calledhierarchies …, from those wherein heterogeneous components were articulated as such, which we referred to as meshworks ...
      … the flows of lava, biomass, genes, memes, norms, money (and other "stuff") are the source of just about every stable structure that we cherish and value (or, on the contrary, that oppresses or enslaves us).

p263. A dynamical system whose behavior is governed by ... endogenously generated stable states is further characterized by a certain number of key parameters. … at critical values of these parameters, bifurcations occur which abruptly change one set of attractors into another.

p264. abstract machines may also be viewed as equipped with "knobs," controlling parameters whose intensity defines the dynamical state of the structure-generating process. .... Other key parameters are those controlling the strength and thoroughness of the sorting process and the degree of consolidation or reproductive isolation of the double-articulation machine; or the degree of connectivity that determines when a meshwork becomes self-sustaining; or the rates of mutation and recombination that define the speed of the probe head, as well as the strength of the flow of biomass and of the coupling between coevolving species---parameters that define the kind of space that the probe head explores.

p270-1. whether these or other diagrams are used to model the structure-generating processes involved in the genesis of social forms, what matters is explaining this genesis in an entirely bottom-up way. That is, not simply to assume that society forms a system, but to account for this systematicity as an emergent property of some dynamical process. … we must take into account that the larger-scale structures that emerge from the actions of individual decision makers, such as formal organizations or informal networks, have a life of their own. They are wholes that are more than the sum of their parts, but wholes that add themselves to an existing population of individual structures, operating at different scales

And we finish off with a nicely acidic comment on not being too smug about thinking we've solved the problem.

p273. No doubt, the entrenchment in the academic and scientific worlds of certain discursive practices informed by linear thinking and linear representation is indeed part of our problem. But to try to reduce a complex situation to a question of representations is, in turn, a homogenizing force very much alive today among social critics.

Manuel De Landa.
Intensive Science and Virtual Philosophy.
Continuum. 2002

Manuel De Landa.
A New Philosophy of Society: assemblage theory and social complexity.
Continuum. 2006

Manuel De Landa.
Philosophy and Simulation: the emergence of synthetic reason.
Continuum. 2011