THE INTERNET AS A COMPLEX ADAPTIVE SYSTEM
By the late 1970s a community of economists had begun grappling with the role of ‘complexity’ in the field of economics. These scholars distinguish complex systems from merely ‘complicated’ systems, explaining that it is not possible to understand the whole simply by better grasping each of the parts.
As Miller and Page (2007, p. 27) note: ‘As the parts begin to connect with one another and interact more... we move from the realm of complication to complexity, and reduction no longer gives us insight into construction’.The Austrian critique of neoclassical equilibrium, initiated by thinkers like Hayek, paved the way for economists to consider theories in which diverse economic agents interact in this complex and networked fashion (Koppl, 2009). In recent years, ‘neo-Schumpeterian’ approaches have emphasized the evolutionary nature of these dynamics (Langlois and Everett, 1994).
A group of scholars affiliated with the Santa Fe Institute have done much of the seminal work in this area. This interdisciplinary group of thinkers draws heavily from biological and evolutionary models in order to inform their approach to economics. The canonical output of their thinking is catalogued in a three-part series of edited volumes entitled The Economy as an Evolving Complex System (Anderson et al., 1988; Arthur et al., 1997; Blume and Durlauf, 2006).
Eric Beinhocker has summarized these developments as what he calls simply ‘complexity economics’ (2006). In place of traditional economics’ conventional wisdom of equilibrium and the market’s ‘invisible hand’, Beinhocker emphasizes a notion of ‘fitness functions’. Various emergent structures may be more or less fit for the environment and the task at hand. The best chance of finding good fitness functions lies in leaving the emergent system open to subsequent experimentation, adaptation, and emergence.
Holland and Miller describe economies that function as ‘complex adaptive systems’ consisting of a ‘network of interacting agents’ that exhibit ‘a dynamic, aggregate behavior that emerges from the individual activities of the agents’. Each of these agents ‘behaves so as to increase this value over time’ - that is, their behavior evolves (1991, p. 365).
In the following four subsections, we describe these essential characteristics of complex adaptive systems and give examples of how the Internet exhibits each.
3.3.1 Agents
Agents are economic actors, and they are individual nodes in a network. Whether acting as consumers or investors, CEOs or government officials, all of us play this interactive role in the economy. The term agent is preferred to either consumer or user, both of which tend to reduce humans to a one-way transactional relationship.
Whereas the usual economic model of human behavior posits ‘incredibly smart people in unbelievably simple situations’ (Leijonhufvud, 1996), the agent-based view admits that we have diverse motivations and limited knowledge. We are not ‘homogeneous billiard balls or gas molecules’ (Taylor, 2004, p. 273). We have a variety of motivations, including those that are non-pecuniary and non-proprietary (Benkler, 2008, pp. 460-73). We utilize not just reason but imagination, intuition, and creativity. We are altruistic, cooperative, and sharing creatures. We can use intelligent action to ‘tip’ the world in certain directions. Perhaps most importantly, we learn - that is, we have ‘evolved the adaptation of adaptability’ (Shermer, 2008, p. 190).
Agents throughout the Internet ecosystem exhibit these characteristics. The billions of humans that use the network daily each have their own motivations and adaptive ways of achieving their goals. This includes a tremendous amount of creating and sharing of culture in a participatory fashion that was far more difficult in one-way broadcast media. Likewise, adaptation on the Internet is cheap relative to markets that are constrained by physical commodities.
Start-ups can try, fail, and adapt in a matter of weeks. The general purpose protocols at the heart of Internet software and hardware (which we discuss at more length in subsection 3.4) allow more entities to try their hand at being a productive agent - at any layer of the network.3.3.2 Networks
Economic agents do not exist in a vacuum. The full productive potential of agents comes from their interactions with each other, which facilitate sharing of information and effort. Any particular agent may have a link to several other agents, who in turn link to others through lines of communication, common tasks, market agreements, or any number of other relationships. In the language of network science, the agents are ‘nodes’ and the links are ‘edges’. The connections between agents can be unpredictable and ephemeral, helping to create complexity in the overall system. Network science explores how networks form and attempts to explain why certain dynamics in networks arise that do not appear in more static, linear systems (Johnson, 2001; Barabasi, 2002; Watts, 2003).
In many systems, individual actors end up having indirect positive effects on others. Economists call these effects ‘positive externalities’ and often describe the benefits that accrue to others as ‘spillovers’ (Frischmann and Lemley, 2006). For example, I may invent a new method for scanning bar codes that yields me great profit, but you might adopt or adapt this technology to your own benefit (provided that the law allows). Furthermore, to the extent that different agents share this standard - say, a manufacturer using bar codes for inventory management and a retailer using the same codes to automate checkout - the system benefits exceed the sum of the parts, becoming ‘network effects’. In complex networks, all of these benefits flow more freely than in disconnected islands. Each new node creates added value for the existing nodes.
These forces are strongly at work on the Internet, where new innovations are immediately available to the entire network of agents.
An improvement in the core Internet Protocols benefits all users, and an app built on top of layers of others’ work can grow exponentially overnight. This phenomenon is strikingly evident in the case of social networking tools like Facebook or with platforms like iOS or Android, but many of the same forces are at work throughout the Internet economy. The Internet’s engineers understand that more complex systems like the Internet display more non-linearities, which occur (and are ‘amplified’) at large scales (Bush and Meyer, 2002). Moreover, more complex systems often exhibit increased interdependence between components, due to ‘coupling’ between or within protocol layers.Arthur (2000, p. 1) notes that ‘the network is the dominant pattern of the new digital economy’, subject to increasing returns. This extends beyond the bounds of the Internet itself to all technologies touched by networked economic dynamics. In the language of neo-Schumpeterian evolutionism:
A network is useful, increases efficiency and systemic performance, but is altruistic in the sense of ‘systemic interests’... Networks are not institutions. The behavior of network units may be influenced by institutional parameters, but the emergent knowledge base of the network is basically open with regard to the future. (Dopfer, 1994, p. 154; original emphasis)
This helps to shift our focus from the ‘firm’ to the network-enabled configurations of and interactions between agents. Ronald Coase explained that firms are created in order to reduce transaction costs - the costs of finding and negotiating interactions with partners (1937). By bringing many entities together under a single umbrella, an organization can limit the transaction costs required. In complex networks, these units need not be literal ‘firms’, and the multitude of links can reduce transaction costs in a more dynamic fashion.
3.3.3 Evolution
To Schumpeter, ‘the essential point to grasp is that in dealing with capitalism we are dealing with an evolutionary process’ ([1942] 1976, pp.
82-3). ‘[Economics] is the study of how humans choose. That choice is inescapably a biological process’ (Glimcher, 2004, p. 336). Evolution is the algorithm for change in economic systems (Vermeij, 2004), and the iterative process of experimentation by agents contributes to optimal growth. Hayek noted that markets involve the ‘evolutionary formation’ of ‘highly complex selfmaintaining orders’ (1988, p. 9). Evolutionary economists use ‘replicator dynamics’ to represent how heterogeneous agents innovate, interact, and evolve in the Darwinian spirit of ‘survival of the fittest’ (Cantner, 2009).On Beinhocker’s analysis, the first step of evolution is ‘differentiation’, in which intelligent agents identify various possible approaches. Next, through experimentation, these agents sort through the variations in order to select the most fit solutions - to decide what works and what does not. Finally, the agents share and iterate on the most successful approaches, throwing out the others and amplifying the effects (2006, pp. 213-16). In other words, natural selection both ‘weeds out’ what fails and ‘weeds in’ (‘nurtures’) what works.
Complexity economists describe evolution as operating on ‘technologies’. One way of viewing this is that evolution operates on two broad types of technologies, which Richard Nelson refers to as ‘physical technologies’ and ‘social technologies’ (Nelson, 2003). Physical technologies are means or recipes for producing objects or ideas; they consist of specifications, instructions, shareable practices, and other ways of transforming materials to serve a goal. These technologies have a modular, building-block character consisting of components and architecture. Social technologies, on the other hand, are methods and designs for organizing people in service of a goal, and instilling order in the social realm.
Technological breakthroughs can come from unexpected directions. Perhaps most important is the process of adaptive tinkering. Chance and accident produce such efficient systems that we often forget that their logic is often the result of non-linear trial and error.
As Taleb has observed, ‘The reason free markets work is because they allow people to be lucky, thanks to aggressive trial and error, not by giving rewards or “incentives” for skill. The strategy is, then, to tinker as much as possible’ (2007, p. xxi).The Internet has, at all layers, evolved according to such trial-and-error. The Internet Engineering Task Force (IETF) is an example of a flexible ‘social technology’ premised on ‘rough consensus and running code’ with input from anyone who cares to participate. The IETF generates ‘physical technologies’ in the form of specifications - some of which are adopted widely and survive as the ‘fittest’, and others that fail to gain adoption. Likewise, the Internet has served as a platform for tinkerers, innovators, and entrepreneurs who constantly experiment with new business models. Amidst endogenous waves of boom and bust, transformative technologies and businesses emerge.
3.3.4 Emergence
Economic agents interact and evolve via networks, ultimately yielding macro systems - that is, emergence. The value generated by this emergent structure is more than the sum of its parts, and the fact that it is complex means that its form is hard to pin down or to predict. The scholarship on this phenomenon spans the physical sciences, social sciences, economics, and interdisciplinary studies (Holland, 1998; Johnson, 2001; Morowitz, 2002).
Emergent systems have no single ideal structure. They exist in an ever-changing environment and consist of complex interactions that continuously reshape their internal relationships. The many independent actions of agents unify, but they do not necessarily work toward one particular structure or equilibrium. For example, emergent systems can be robust to change, and they can be far better at evolving toward efficiency than topdown systems. On the other hand, emergent structures can fall apart when their basic conditions are altered in such a way that they work against the health of the system as a whole.
Agents’ actions in turn affect the other agents, setting off both positive and negative feedback loops. Fortunately, we have developed some understanding of what types of conditions lead away from such negative feedback loops, and towards more productive emergence. Generally speaking, a greater ability of agents to connect and explore new modes of production will facilitate the chance connections that a top-down designer might not foresee. Better global information sharing and feedback between agents facilitates better local decisions. The system as a whole can leap forward when new innovations come out of this process and are replicated throughout the network. Inductive tinkering by a single agent can lead to breakthroughs with widespread payoff.
Emergent systems are often described as being organism-like. One canonical example is the ant colony. Each ant follows rules for when and how to forage, leaving pheromone trails to food. In one strikingly literal parallel to the Internet, a team of biologists and computer scientists found that harvester ants employ congestion control techniques that mirror those of the Internet’s Transmission Control Protocol (TCP) (Prabhakar et al., 2012). Beyond this compelling anecdote of a real-world ‘anternet’, the Internet exhibits widespread emergent structure. For instance, the Border Gateway Protocol (BGP) that controls network routing at the core of the Internet relies on information passed on from neighbors, adapting dynamically to changes in network structure (with the help of human network operators when things go awry). The economic relationships between these interconnecting entities are the subject of constant individual negotiation and adaptation. Nevertheless, the Internet continues to function largely as a single unified network.
The Internet is not, however, merely the routing of traffic and the negotiation of interconnections. It is a modular, layered system, in which the core standards in the ‘middle’ of the Internet allow others to build on top of the network to create their own standards and applications. Applications may themselves facilitate follow-on innovation - technologically via something like an application programming interface (API) or simply as a matter of agents finding new ways of using a tool. For instance, Twitter relies on the emergent structure of the Internet - globally interconnected wired and wireless networks, protocols like HTTP, and the ‘agent base’ of millions of Internet users. Others have built on top of Twitter’s API to create new applications, or have used it to establish new cultural practices or form political movements. The flexibility at the heart of the Internet is reflected to varying degrees in all of the technologies built on top of its core protocols. The more that these technologies permit evolution and experimentation, the more complex the emergent Internet becomes. At each layer, and within each layer, complex adaptive systems may develop and influence each other. In this sense, the Internet can be seen as a complex adaptive system of complex adaptive systems - a constellation of emergent structures.
3.4