Apparatus
Notes on Sources
This book does not require its readers to read the works it is built on. It makes them available to the readers who want to. What follows is a working bibliography, annotated lightly, ordered roughly by where each work first does its work in the book. A reader who wants to follow a single thread is given the place to start.
The empirical anchor of the book is the fieldwork of Deborah M. Gordon, who has studied the same population of harvester ant colonies near Rodeo, in the bootheel of New Mexico, since 1985. Two of her books are written for the general reader and carry most of what the colony chapters rest on: Ants at Work: How an Insect Society Is Organized (Free Press, 1999), and Ant Encounters: Interaction Networks and Colony Behavior (Princeton University Press, 2010). The second is the more technical of the two and is the source for the central mechanism this book returns to again and again — that a forager leaves the nest in response to the rate at which other foragers are returning, and that this rate is sensed through brief antennal contact rather than read from any central signal. Readers who want the primary record can follow Gordon's papers in Behavioral Ecology, Animal Behaviour, and Proceedings of the Royal Society B; the long-term study of colony lifespan and the developmental differences between young and mature colonies is reported across that body of work over four decades.
For the comprehensive reference behind nearly every factual claim about ants that is not Gordon's own, the source is Bert Hölldobler and Edward O. Wilson, The Ants (Harvard University Press, 1990). It is a large book and not a casual read, but it is the standard work, and it is where the figures on caste, colony size, species count, and chemical communication are grounded. Hölldobler and Wilson's later The Superorganism (W. W. Norton, 2009) is the more accessible companion and is the better starting point for a reader coming to the subject for the first time. Wilson's earlier work on the chemistry of pheromone communication — the decoding of the alarm, trail, and recruitment substances over the second half of the twentieth century — underlies the chapter on the pheromone trail.
The word at the center of the book's architecture comes from Pierre-Paul Grassé, who in 1959 described how termites coordinate the rebuilding of a nest without any of them holding a plan, each one's work prompted by the state of the work already done. The paper is La reconstruction du nid et les coordinations interindividuelles chez Bellicositermes natalensis et Cubitermes sp. (Insectes Sociaux, 1959). Grassé called the mechanism stigmergy. The book uses the idea throughout and the word sparingly; this is where it begins.
The observation that a price is a way of holding information that no single participant possesses belongs to Friedrich A. Hayek, The Use of Knowledge in Society (American Economic Review, 1945). It is short — a dozen pages — and it is the source for the chapter on the market. A reader who has never read it will find that it reads less like economics than like a description of distributed cognition, which is how this book treats it.
The principle that connections strengthen through joint use comes from Donald O. Hebb, The Organization of Behavior (Wiley, 1949), the origin of what is now compressed into the phrase neurons that fire together wire together. The mathematical form of that principle, and the proof that it extracts the most important structure from its input, is Erkki Oja, Simplified neuron model as a principal component analyzer (Journal of Mathematical Biology, 1982). These two works stand behind the chapter on the brain and behind the claim, made in the companion whitepaper, that the operation strengthening a synapse and the operation strengthening an ant trail are the same operation.
The bounds on what any channel can carry — and so on what any substrate can hold and move — come from Claude E. Shannon, A Mathematical Theory of Communication (Bell System Technical Journal, 1948). It is the founding document of information theory and it is the reason the book can speak about signals and channels with a straight face.
The formal results that turn the book's central observation into a defensible claim are gathered in the companion whitepaper, The Mathematics of Emergence, and rest on five established theorems. The convergence of trail-laying search under bounded reinforcement and non-zero exploration is Marco Dorigo's ant colony optimization framework, first in his doctoral thesis, Optimization, Learning and Natural Algorithms (Politecnico di Milano, 1992), and made rigorous in Thomas Stützle and Marco Dorigo, A short convergence proof for a class of ant colony optimization algorithms (IEEE Transactions on Evolutionary Computation, 2002). The convergence of evolutionary dynamics in a fitness landscape is John Maynard Smith, Evolution and the Theory of Games (Cambridge University Press, 1982). The convergence of a learner updating on evidence is the martingale convergence theorem of Joseph L. Doob, Stochastic Processes (Wiley, 1953). The result that a sufficiently capable network can approximate any continuous function is George Cybenko, Approximation by superpositions of a sigmoidal function (Mathematics of Control, Signals and Systems, 1989). The fifth is the one already named above, in the note on the brain: Erkki Oja's proof that Hebbian strengthening converges on the principal structure of its input (1982), which is why the operation that strengthens a synapse and the operation that strengthens a trail can be treated as one operation. The whitepaper is where these are combined; this book only points at the combination.
For the framework that gave biology its first clue that pattern can arise from nothing but local interaction and diffusion, the source is Alan M. Turing, The Chemical Basis of Morphogenesis (Philosophical Transactions of the Royal Society B, 1952). It is the earliest of the works on this list to describe, in mathematics, how structure can come from a substrate and a rule, with no plan anywhere. The imitation game that the Gordon Test is built as a counterpart to comes from the same author's earlier Computing Machinery and Intelligence (Mind, 1950).
The current frontier — the move from modelling stigmergy as discrete trails to treating it as a continuous field that can be steered — is surveyed in the Royal Society's Stigmergy: from mathematical modelling to control (Philosophical Transactions of the Royal Society A, 2024), which is the most recent work the book draws on and the best single pointer to where the research is now going.
The chapters on the city, the scientific community, the body, and the language draw on broad and well-established bodies of work rather than on single texts, and the book has tried to make no claim in them that a standard reference in urban studies, the sociology of science, immunology, or historical linguistics would dispute. Where the evidence in those fields is settled, the book has stated it plainly. Where it is contested or thin, the book has softened its language accordingly, and the reader should read those softenings as deliberate.
The chapter on the ledger refers to one document by name, because its title is part of the historical record and naming it dates nothing: the 2008 paper that described a method for maintaining a single agreed record across mutually distrustful computers. The book names the paper and not the many systems built after it, for the same reason it names no other product anywhere — the architecture is the subject, and the architecture outlasts its implementations.
That is the whole of the apparatus. The reader who wants to check the book against its sources can do so. The reader who wants only to keep looking at what the book has pointed at can close it here, and start noticing.