Chapter 9 2.6 Advantages and disadvantages of clustering

There are two extreme paths to produce "more".One approach is to build systems with the idea of ​​sequential operations, like an assembly line in a factory.The principle of this type of sequential system is similar to the internal logic of a clock - through a series of complex movements to reflect the passage of time.Most mechanical systems follow this logic. There is another extreme path.We find that many systems are stitched together of parts that work in parallel, much like a brain's network of neurons or an ant colony.The actions of such systems arise from a messy and interrelated set of events.They are no longer driven and manifested in discrete ways like clocks, but more like thousands of clockworks working together to drive a parallel system.Since there is no chain of instructions, a specific action of any one clockwork will be transmitted to the entire system, and the partial performance of the system is more easily concealed by the overall performance of the system.What emerges from the crowd is no longer a series of pivotal individual actions, but a multitude of synchronized actions.The group patterns exhibited by these synchronized movements are much more important.This is the cluster model.

Both extremes of organization exist only in theory, since all systems in real life are a mixture of these two extremes.Some large systems prefer a sequential model (such as a factory), while others prefer a network model (such as a telephone system). We found that most of the most interesting things in the universe are near the network mode end.The same is true of intertwined lives, intricate economies, bustling societies, and unpredictable minds.As dynamic wholes, they share certain qualities: a certain vitality, for example. These parallel systems go by all kinds of names: bee colonies, computer networks, brain neuronal networks, animal food chains, and agent swarms.The categories to which the above systems belong also have their own names: networks, complex adaptive systems, swarm systems, living systems, or swarm systems.I use all of these terms in this book.

Each system is organized as a collection of many (thousands) of autonomous members. "Autonomous" means that each member reacts individually according to internal rules and local environmental conditions.This is very different from following orders from the center, or reacting in unison to the overall environment. These autonomous members are highly connected to each other, but not to a central hub.They form a peer-to-peer network.Since there is no control center, people say that the management and center of such systems are decentralized and distributed in the system, in the same form as the management of hives.

The following are four salient features of a distributed system, from which the idiosyncrasies of a living system derive: no mandatory central control Subunits have the trait of autonomy Subunits are highly connected to each other Point-to-point influences form a non-linear causal relationship through the network The importance and influence of the above features in distributed systems has not been systematically examined. One of the themes of this book is that distributed artificial living systems—such as parallel computing, silicon neural network chips, and vast online networks such as the Internet—expose both the fascination of organic systems and their certain defects.Here is my overview of the pros and cons of distributed systems:

Benefits of the group system: Adaptable - One can build a clockwork-like system that responds to preset stimuli.But to be able to respond to stimuli that haven't been seen before, or to be able to adjust to changes over a wide range, requires a swarm—a hive mind.Only a whole consisting of many components can continue to survive or adapt to new stimulus signals when some of its components fail. Evolvable—Only the swarm system can transfer the adaptability acquired by local components over time from one component to another (from body to gene, from individual to group).Non-population systems cannot achieve (similar to biological) evolution.

Resilience - Since swarm systems are built on top of many parallel relationships, there is redundancy.Individual behavior is irrelevant.A glitch is like a fleeting little spray in a river.Even a large fault is equivalent to a small fault in the higher layers and thus can be suppressed. Infinity - For traditional simple linear systems, positive feedback loops are an extreme phenomenon - such as the chaotic feedback of sound reinforcement microphones.In a group system, however, positive feedback can lead to an increase in order.By gradually expanding new structures beyond the scope of their initial state, groups can build their own scaffolding to build more complex structures.Spontaneous order helps to create more order—life begets more life, wealth begets more wealth, information breeds more information, all of which transcend primitive limits and are forever endless.

Novelty - There are three reasons why swarm systems can be novel: (1) They are sensitive to "initial conditions" - the subtext of this academic phrase is that the consequences are not proportional to the causes - thus, swarm systems can Turn small mounds into surprisingly large mountains. (2) The combination of interconnected individuals in the system grows exponentially, which contains countless novel possibilities. (3) They deemphasize the individual and thus allow for individual differences and flaws.In a group system with hereditary possibility, individual variation and defect can lead to permanence, a process we also call evolution.

Obvious flaws of the swarm system: Non-optimal - Swarm systems are inefficient because of redundancy and no central control.Its resource allocation is highly chaotic and duplication of effort abounds.Frogs produce thousands of eggs at a time, only for a few offspring to become frogs. What a waste!If swarm systems have contingency controls—such as the price system in a free market economy—the inefficiency can be suppressed to some extent, but it is by no means possible to eliminate it completely like a linear system. Uncontrollable - there is no absolute authority.Leading a flock system is like a shepherd herding sheep: you need to exert force on key parts, reverse the natural tendency of the system, and turn it to a new target (use the nature of sheep to be afraid of wolves, and use dogs that love to chase sheep to gather them).The economy cannot be controlled from the outside, it can only be adjusted bit by bit from the inside.One cannot stop the dream from happening, only reveal it when it appears.Wherever the word "emergence" appears, human control disappears.

Unpredictability - The complexity of a swarm system affects the development of the system in unforeseen ways. "The history of biology is full of surprises," says researcher Chris Langton.He is currently developing mathematical models of swarms. The term "emergence" has its dark side.The novelty that emerges in a video game is a lot of fun; the novelty that emerges in an air traffic control system can lead to a national emergency. Agnostic - causality as we currently know it is like a clockwork system.We can understand sequential clockwork systems, whereas nonlinear network systems are downright mysteries.The latter are drowned in their self-made sleepy logic. A leads to B, and B leads to A.A swarm system is an ocean of intersecting logic: A indirectly affects everything else, and everything else indirectly affects A.I call this horizontal causation.The true cause (or rather, the true cause combined of several elements) will spread laterally through the network, and eventually, what triggered a particular event will never be known.Let it be.We don't need to know exactly how tomato cells work to be able to grow, eat, and even improve tomatoes.We don't need to know exactly how a massive swarm computing system works to be able to build it, use it, and make it better.Still, we are responsible for a system whether we know it or not, so knowing it certainly helps.

Not instantaneously—ignite a fire and generate heat; flip a switch and a linear system operates.They are ready to serve you.If the system stalls, restarting will do the trick.Simple swarm systems can be awakened with simple methods; but complex swarm systems with rich layers take some time to start.The more complex the system, the longer it will take to warm up.Every level must be settled; horizontal causes must be fully diffused; millions of autonomous members must become familiar with their environment.This, I think, will be the hardest lesson for humanity to learn: organic complexity will require organic time.

Choosing between the pros and cons of group logic is like choosing between the costs and benefits of biological systems—if we had to.But because we grew up with biological systems and had no choice, we always accept their costs without consideration. In order for the tool to be powerful, we can allow it to be a little bit flawed in some ways.Likewise, we have to tolerate nasty worms or blackouts without reason or warning in order to ensure that the entire network of 17 million computer nodes doesn't go down.Multiple routing is wasteful and inefficient, but we can use it to ensure the flexibility of the Internet.On the other hand, I would wager that as we build autonomous robots, we will have to constrain their ability to adapt in order to prevent them from escaping from our full control on their own. As our inventions move from linear, predictable, causal mechanical devices to criss-crossing, unpredictable, and ambiguous living systems, we also need to change our expectations of machines.Here's a simple rule of thumb that might help: For jobs where absolute control is a must, the reliable old clock system remains. Where ultimate adaptability is required, out-of-control groupware is all you need. Every step we bring machines toward swarms is a step toward life.And every step our contraption takes away from the clock means it loses some of that icy but quick and optimal efficiency of machinery.Most missions will seek a balance between control and adaptability, so the equipment most conducive to the job will be a hybrid of a cybernetic system partly controlled and partly swarm.The more mathematical properties of general group processing that we can discover, the better our understanding of biomimetic complexity and biological complexity will be. Groups highlight the complex side of real things.They are unconventional.The mathematics of swarm computing continues Darwin's revolutionary study of plants and animals undergoing random variation to produce random populations.Group logic attempts to understand imbalances, measure instability, and measure unpredictability.It was an attempt, in the words of James Gleick, to map out "amorphous morphology"—that is, to give shape to forms that seem inherently formless.Science has solved all the simple tasks—clear and concise signals.Now all it faces is noise; it must face the chaos of life.