I picked this up hoping for it to shed some light on complex adaptive systems, in particular: how adaptation builds complexity. What I read instead was a book that was unfocused, unclear and rushed. Instead of taking the proper time to explain the nuances of the theory, John H. Holland instead assumes everyone already knows what he's talking about and then proceeds to explain things in a rushed and unfocused manner. I do not recommend this book to anyone, as it did not shed any more light on the ideas behind complexity or in particular complex adaptive systems.
A great book, but difficult to review. This book raises hundreds of interesting questions but answers very few. If you liked Anti-fragile (Taleb), the Misbehavior of Markets (Mandlebrot), or the Selfish Gene (Dawkins) you'll probably like this as well.
I also think reading Thinking in Systems by Donella Meadows before reading this was helpful.
If you read books to settle scores, however, you won't like this at all.
The book presents an attractive view of adaptation in complex systems. The first three chapters are great in laying the foundation of Holland's idea, however, the proposed methodologies/simulations in the rest of the book seem outdated in 2022.
If I can summarize the book in one sentence:
Adaptation is the ability of a system to change its behavior when facing a perturbation.
My favorite notes from this book are:
"Adaptation, in biological usage, is the process whereby an organism fits itself in its environment"
"If we remove one kind of agent from the system, creating a "hole," the system typically responds with a cascade of adaptations resulting in a new agent that "fills the hole." "
"Diversity also arises when the spread of an agent opens new niche opportunities for new interactions that can be exploited by modifications of other agents"
"...in complex adaptive systems, a pattern of interactions disturbed by the extinction of component agents often reasserts itself, though the new agents may differ in detail from the old"
"We gain experience through repeated use of the building blocks, even though they may never twice appear in exactly the same combination. "
"We gain a significant advantage when we can reduce the building blocks at one level to interactions and combinations of building blocks at a lower level"
Short, quick read. The opening is worth the read as it provides a system perspective of the world where pattern rather than substance defines reality or at least is the only conduit through which reality can be experienced. Specifically interactions in space and time render meaning to objects and we do not experience the objects themselves per se but rather interpolate they exist from myriad of interactions with the same location in space over time using diverse sensory experiences.
The book then spills into a discussion of genetic programming which is markedly less creative modeling completely after DNA. This part of the books is quite dated and to be honest after a while I just skimmed over.
That said, the opening alone is worth the read and merits the 5 star rating.
Good back-to-the-basics read on properties and mechanisms that describe complex adaptive systems and adaptive agents. Many problems in this category are now tackled as part of the work being done on reinforcement learning and artificial intelligence. Amazing to see how in the past 5-10 years we moved from research into a series of impactful practical applications.
The reader is taken on a scientific journey leading toward a theory explaining Complex Adaptive Systems, by John Holland; one of the legendary thinkers and original members of the Santa Fe Institute.
Holland readily admits theory is a distant goal and he is trying to lay the foundation that will be needed to eventually develop a theory, beginning with a way to conceive of and communicate the phenomena being modeled.
It was sometimes taxing to keep reminding myself the mental constructs being described were themselves a way to describe complex adaptive systems.
The first third of the book covers “The Seven Basics” that Holland has observed in all complex adaptive systems. Here are the seven basics and some other observations using a model of “adaptive agents” sensing their environment and taking action based on If/Then conditional rules.
Properties • Aggregation: sets • Nonlinearity: outputs greater than sum of inputs • Flows: inputs processed into outputs • Diversity: constant evolution to fill niches
Mechanisms • Tags: labels necessary for aggregation and differentiation • Internal Models: anticipation and prediction • Building Blocks: segments recombining to create internal models
Using “credit assignment” processing rules compete and successful rules are strengthened. This feedback is one way a system adapts. Competing rules actually support each other even when they contradict. By having varying strengths, default hierarchies will naturally form and a model may evolve over time through competition.
Even with a relatively small number of “detectors”, such as 100, the possible combinations might as well be infinite (10^30). No system can efficiently evaluate every possible configuration at that scale; much less act on it, but this building- block-and-rules model allows a system to describe novel situations in terms of familiar components and take action even in a completely new set of circumstances.
Competing rules form internal models. By default, hierarchies emerge where more general rules give way to the more specific. Over time experience creates even more specifics.
At this point the “adaptive agents” that do the processing (e.g. cells in a body or businesses in a city) include: 1. A performance system with detectors, If/Then rules, and effectors. Or put another way: Information > Processing > Action. The type of detector determines the properties of the environment to which the agent is sensitive. 2. A credit assignment algorithm that provides a system with hypotheses which anticipate future consequences and strengthen rules accordingly. Competition is the basis for credit assignment and rewards begin by keeping reservoirs of basic needs away from empty. 3. A rule discovery algorithm. Rule discover (generating plausible hypotheses) centers on the use of tested building blocks (sub-assemblies). Tags create associations and form default hierarchies.
Holland is looking for “lever points”; places where a small action has large impact such as vaccines to an immune system or enzymes redirecting cell activity. He’s trying to find general principles that underpin all complex adaptive systems.
The final two thirds of the book focus on his “Echo” models and go into excruciatingly technical details. Since I wasn’t reading this book to follow Holland’s academic enterprise with my own research, I began skimming instead of reading intently for the remainder of the volume.
In the early chapters he seemed to be forcing the model to become something that could be tested using a computer. This was a little alarming, because it reminded me of the way “traditional economics” made all agents perfectly rational so the math would work out, where “behavioral economics” is a much better explainer of the chaos that is human behavior.
In the final chapter he acknowledges the limits on computer simulation but also says they are the only way to experiment with complex adaptive systems. “We cannot follow the traditional experimental path, varying selected variables under repeated runs, while holding most variables fixed, because controlled restarts are not possible with most complex adaptive systems, and because some complex adaptive systems operate over long time spans.”
Finally, Holland (writing in 1995) says we need new math to support a theory especially in two key areas. “One is an organized theory of a dynamics based on sets of equations that change in number (cardinality) over time. Another is a theory relating Building Blocks to hierarchical structure.”
This entire review has been hidden because of spoilers.
It's been hard for me, as a biologist, to find much useful in 'complexity' books. This one continues the disappointing trend, this time exploring agent-based systems and genetic algorithms. The author invented genetic algorithms, which are a very useful optimisation technique modelled upon biological mechanisms of genetic transmission, and can generate unexpected and surprisingly robust solutions. Despite being their sources of inspiration, they have little to offer back to genetics or evolutionary theory.
Similarly, agent-based systems are inspired by living systems of particular kinds and can be endowed with simple rules that express a wide variety of behaviours. From the start, though, the seven "properties" and "mechanisms" the author uses as essential to a (any?) cas -- complex adaptive systems, a lowercased and italicised acronym -- do not feel particularly insightful. I don't want to take the time to try tightening them up, but eg, seeing 'nonlinearity' as an essential property is yawn-inducing. A Lotka-Volterra predator-prey model comes in in equation form but we don't get a graph of the nonlinear cycling and what it means ecologically; instead the author wastes pages developing a billiard ball example to again explain the endogeny of nonlinearity, a model that helps not at all for the remainder of the text. Despite this attention to its existence, basic insights into the implications of nonlinearity remain undeveloped.
Engagement with each field is minimal. The reference used for economics, for example, was 45 years old at the time of the lectures. Citing The Selfish Gene is fine, but so? Overall, this book feels underdeveloped and inward. I'm hoping some of the author's other books are better, perhaps Adaptation in Natural and Artificial Systems (1975), which started genetic algorithms? As for complexity, I'm expecting that Gell-Mann's 1994 book is also not good, we will see. Hopefully complexity books get better as we move out of the 1990s.
J. Holland describes the key properties and mechanisms of complex adaptive systems (CAS). These mechanisms are the presence of building blocks, tagging, and internal models. Properties are aggregation, nonlinearity flows and diversity. He then describes a thought experiment to reproduce "morphogenesis": the emergence of specialisation and aggregation through evolution only.
What I got out of it Fascinating book on how the universe seems to produce order for free via coherence, spontaneous self-organization and complex adaptive systems.
Summary Holland walks us through how coherence emerges from unstructured agents in environments of continuous flux; coherence under change and complex adaptive systems (CAS)
Key Takeaways 1. Behavior depends much more upon interactions of agents than their actions 2. Catalog of all activities does not equal understanding the effect of changes in the ecosystem Read more at https://blas.com/hidden-order/
Curious 1995 introduction to "complex adaptive systems" (from the human body / immune system to cities and economies), the second half is a semi-interesting detailed overview of computer modeling a genetically-evolved agent system to understand these more directly than where the author feels mathematical modeling was struggling to deal with the non-linearity and hierarchical/relational aspects of these systems.
A good overview of the framework for complex adaptive systems and their base principles. This book is already a bit outdated as there are lots of thought experiments while not reasonably enough real computer simulations.
A work trying to put complexity theory or general systems theory into a complete package, the first Ulam Lecture at the SFI. A bit thin, missing context from the fields it takes inspiration from, and unsatisfying. Some interesting pieces in it, but I had greater expectations coming into the book.
Estamos perante um tratado extraordinário sobre a complexidade. O autor consegue, numa linguagem acessível, transmitir o state-of-the-art (em 1995) sobre este fenómeno que é um dos mistérios que a ciência ainda não conseguiu explicar nem caracterizar. O autor defende que, através de regras e processos simples, descritos em detalhe, a adaptação gera a complexidade. O único ponto menos forte da obra é o detalhe e as páginas dedicadas a certos pontos que não irão interessar ao comum dos leitores.
Aconselho a todos os que queiram aprender sobre sistemas complexos e adaptativos.
I'm going to give a 5 stars rating to this book because its first half and conclusion are worth it. Introduction to "Complex Adaptive Systems" (CAS), their properties, and mechanisms to to understand them. I was quite surprised to read a book that is almost a quarter of century old and still very relevant as an introduction to the field. It speaks of the vision of the author. The two tiers theory (theories of different scales) hinted at the end looks very promising as well.
Chapters 3 and 4, about the Echo model, were unconvincing to me because of the lack experimental data. I guess there was some progress since then?
An interesting primer on the subject of designing and building a model for complex adaptive systems. Holland walks through his work in a way that is accessible to the lay reader, and builds from foundation to completion an understanding of the subject. This work should be of interest to anyone who works in modelling or simulation, artificial intelligence, biology, or mathematics.
Though this book's heart is in the right place, it is better titled: The Echo Runbook. Its diagrams appear to come from a nearby whiteboard, not its text. Unless you're an aspiring geneticist, do not attempt this book before reading Complexity.