January 20 2009
Five percent of malformed data packets on the internet cannot be explained - they cannot be traced to tampering or software and hardware glitches. Is this evidence that the internet is evolving into a globally connected single intelligence? The biggest puzzle is that an emergent intelligence would be so bizarre and counter-intuitive, we are unlikely to understand any immediate clues to its manifest existence and even if we did, we would have no clear method to communicate to it. But we do know about some of the mechanisms from which such a phenomenon might burst.
Genetic algorithms are known to produce various strange organic conflations in electronic circuits. In Adrian Thompson's experiments with FPGAs, successive generations of a circuit were trained to detect a square wave of a particular frequency. In these experiments, the evolved circuit seemed to jump beyond the constraints of logic gates to exploit hidden properties of the digital hardware, which eventually led to producing the correct result from a tiny assemblage that made no logical sense, and was so bizarre that no human engineer could understand or explain why it worked. Evidence suggests that the circuit was making use of tiny electromagnetic quirks in the materials of the particular microchip - when the circuit was copied to another chip from the same batch, it simply didn't work. It also didn't work when constructed in an artificial software simulation.
The very concept of such genetic algorithms arises directly from the history of operating systems and artificial evolution. In the 1980s, a cult game amongst systems programmers emerged, known as Core Wars. In 1989, inspired by Core Wars, and increasingly fascinated with the possibility of modeling natural selection in self-replicating virtual machines, evolutionary biologist Tom Ray began a related experiment called Tierra. Starting with a single ancestor program that would reproduce by making copies of itself in memory, Ray devised a virtual environment that would mutate these programs by flipping random bits in their instructions.
At first, Ray's idea was met with great skepticism. Ecologists and AI researchers couldn't believe that randomly mutating a program would lead to any improvement of its functionality - such a system would be far too brittle to return any reasonable results. But when he flipped the switch in early 1990, Ray learned that his hunch was entirely correct. The programs actually did replicate and evolve via mutation.
The first ancestor program (and the only one hand coded by Ray) was 80 bytes long; the shortest possible program that could achieve this end (or so he thought). After a few generations, 81s started to appear - mutants with an additional byte attached. Later, a 79 appeared. With a smaller memory footprint, the 79s began to take over the environment! But what was more stunning - eventually a 45 appeared - nearly half the size of what Ray had thought was the shortest possible program. Upon investigation, it turned out that the 45 was a parasite - it attached itself to the replication code of another 80 and executed this code to replicate itself. None of the constraints of the simulation were geared towards parasitic behavior - this was a completely unplanned consequence of the simple evolutionary principles underlying the system. The most remarkable result was the further emergence of a 22. It turned out that the 22 was entirely self contained - not a parasite, yet a mere quarter of the length of the simplest program that a human could comprehend. While tangled and impenetrable to the human mind, the 22 byte self-replicating program was entirely functional.
One of the most interesting results of the Tierra experiments was the way in which the absence of influences from a physical environment drove the system to be entirely based on co-evolution from interaction between the species of programs. Results like this suggest evolution is an abstract pattern of the universe, not a phenomenon of planetary biology. Selection processes seem to emerge from any conflation of distinct individual 'programs' or 'cells' under common environmental pressures. There is no sense of design, the species simply mutate and slowly change their behavior until they cross thresholds of regularity. Apart from the obvious conclusion that this is a devastating critique of "intelligent design", it raises a very significant mystery as to what is really emerging through tiny electromagnetic aberrations and malformed signals that swarm across the global communications grid.
The evidence from evolved circuits suggests that humans have yet to come close to grasping and utilizing anywhere near the full range of phenomenon exuded by electron flow and electromagnetic waves. Anyone who has ever played with electronic musical instruments will have experienced this directly. Sometimes the signals and flows form strange loops, linger and bubble in the circuit after switches have been turned off, and can shift and transform in unpredictable ways. Some electrical devices need to be "worn in" or "tuned" to the particular combination of their environment, settings, and power source. Human engineering does not have the requisite tools and techniques to fully understand and analyze the tiny variations between electrical components of the same design, but it seems that evolving circuits can take advantage of these variations. To me, this suggests that there are abundant possibilities within electricity grids and internet networks for strange energy and noisy signals to cohere in a global pattern based order.
There are two potential scenarios for the internet intelligence, both vaguely frightening in different ways. I'm not going to speculate too far, until I have come up with a useful way to test this, or at least bring something new to discussions about it.
While the first is possible, it's less likely because of the command and control psychology of the military and the fact that genetic algorithms were not widely appreciated until the 1990s. If von Neumann or Turing had outlived the 1950's, perhaps this scenario would be more plausible.
It is also possible that both scenarios are interlinked. That the military network has evolved further and faster than anyone could have anticipated. That in our obsession with localized point to point communications, we have lost sight of the wider global phenomenon that this network is in some way, silently aware, and evolving, but we simply don't understand it because it is beyond our comprehension.
I should also add that this process may have begun much earlier than the 1960's when the first computer networks were invented. Its foundations could be a deep and twisted by-product of the electrical and telegraph grids of the 19th century.
This Article