Multicellular Computing: Emergence at Internet Scale

The more elements in a system and/or the richer the possible interactions between them, the more likely that surprising emergent behavior will arise.

How many interacting elements are required for emergent behavior to arise? The answer depends upon the complexity of the elements and their interactions.

At one extreme, vast numbers of simple elements with simple interactions can form complex behavior. Gravitational attraction between atoms created galaxies, stars and black holes (see other examples of emergence). At the other extreme unexpected behavior can emerge from only two interacting elements if their interactions are rich and they are sufficiently adaptable (consider a marriage).

In general, the less complex the interactions the more individuals are required to produce surprises. But unless the interactions are very simple, it doesn't take all that many. Small flocks of birds or schools of fish can provide surprising collective behavior. And, as every programmer has discovered, even small software systems all too often demonstrate surprising emergent behavior (bugs).

Multicellular systems, whether in biological or digital realms, have many elements that are themselves complex dynamic systems that interact with each other in quite complex ways.

Neurons are much smaller than grains of sand yet their interactions are far richer. Neural "circuits" of only a few hundred to a few thousand neurons can produce amazing behavior. The entire central nervous system of  C. elegans, a roundworm commonly used in biological studies of development, has 302 neurons and about 5000 synapses, yet it demonstrates social behaviors of a primitive sort and something akin to associative learning. The nervous system of Aplysia (a mollusk) contains 18-20 thousand neurons. These animals aren't noted for their intellectual gifts, yet their behavioral repertoire is far greater and more "purposeful" than a hurricane or a sand dune: they seek food, avoid danger, mate, and learn. Aplysia, demonstrates learning in both classical conditioning experiments and operant reward conditioning.

The earliest computers had a few thousand gates and programs with a few thousand machine instructions. While their planned emergent behavior was primitive by today's standards, they still could calculate artillery ballistic tables, prime numbers and the like. Unplanned emergent behavior (bugs) popped up even then. In modern computers with upwards of a billions gates and hundreds of millions of lines of code the behavior, both planned and unplanned, is beyond our comprehension.

Emergence at Internet Scale

What, then, of the Internet with more than a billion elements, each quite adaptable and each interacting in complex ways with multiple other elements? The simplest answer is, no one knows, nor can know, nor ever will know in detail. Emergence in the Internet is exploding at such a pace and with such subtle and high-speed interactions that it is already literally incomprehensible. A given element (say one PC or smart-phone in the net) simultaneously participates in many emergent multicellular phenomena that are operating at different time scales. For example, a smartphone, wireless Netbook, or iPad is interacting at millisecond scales with cell towers or WiFi hubs while perhaps doing some web eCommerce interaction at second or minute scales, or interacting with FaceBook blogs or Twitter at minute or hourly time scales, or requesting a Google search of Internet-scale information that itself is changing second by second as Google's crawlers and ever-adapting page ranks do their work.

Individual computers participating in the Internet are only the first metalevel of Internet interaction. The trails of browsing left by these individual computers become the fodder for huge advertising networks, e.g. Google Adsense among many others, that compete with each other to organize consumer marketing campaigns in other sets of computers, or to guide political messaging campaigns. Highly organized sets of computers create and/or exploit financial markets that affect the world banking system and therefore its economy.  Massive intelligence operations track and analyze all emails, blogs, and phone conversations world-wide in an effort to identify and/or foil unfriendly efforts by international bad-actors.  Malware botnets form and are exploited by cyber-criminals in the cracks of the systems. Geotracking and social networks create their own sots of stigmergy structures that are the territory of the next generation of emergent behavior.

We can no more understand the emergent systems in the Internet than a neuron can "understand" the brain of which it is a part.

Contact: sburbeck at
Last revised 8/6/2013