Moore's Law is globally famous and deservedly so. However it has a mirror opposite that is just as powerful yet is generally unknown, perhaps because no one likes bad news. We might call it Cassandra's Law: over time technological complexity of all kinds increases exponentially.

In other words, solutions in a given domain become progressively more subtle, time- consuming, and technically demanding. (Consider the road from DOS through Windows to Vista.) Eventually, or so Cassandra's Law seems to predict, solutions will become so expensive, so intractable, that technological progress will everywhere grind to halt and society fall into stagnation (technologically). Call this point the anti-Singularity.

A perfect example of the Law in action -- though of course there are hundreds -- might be the rising tide of complexity associated with the spread of wireless technology. When wireless was first invented every community had its population of amateur radio operators or "hams," citizens who had been able to learn the technology from scratch. A century later any manager who counts a wireless network among his or her responsibilities is dealing every day with immensely hairy problems rising from some mix of spectrum sharing glitches, user mobility complexities, device authentication, coverage patchiness, security issues, problems integrating with wired networks, delays, and layer upon layer of device or application incompatibilities. Typically such problems emerge from the interaction of several domains at once. Wireless is no longer for amateurs.

A corollary of the Law is that over time realtime surveillance of a system has to increase (since there is no other way to disentangle subtle problems). A contemporary high-end wireless LAN might have dozens of sensors or "instruments" scattered though its architecture, often using their own private network to report a range of conditions (like queue length or handoff delay) to the backend, which then processes and passes this data on to the manager's "dashboard". An R&D network in a lab might use hundreds of sensors. Eventually, Cassanda's Law says, it will be impossible for real people to monitor wireless networks in real time, no matter how skilled the humans might be. There will just be too many balls in the air.

We seem to be pretty close to that point today. If so, over the next several years most companies are going to have to move to systems that can monitor themselves, recognize deviations from expected performance, develop theories about the causes of that deviation, test the theories experimentally, and then either report the solution(s) in actionable form or take care of them autonomously. IBM calls this science "autonomics," others use the name "synthetic awareness".

Synthetic awareness is not a small change architecturally. A system will need access to full- scale, realistic, models of itself so that it can compare on an ongoing basis what is happening with what should be happening down to the level of component behavoir. If something goes wrong and there is no pre-existing rule for that particular outcome, models will be essential to diagnose the problem, perhaps by varying the states of the components until the unacceptable behavior is replicated. (Hopefully the model will be smart enough not to require an exhaustive exploration of the entire state space.)

Recently a team at UCSD, in collaboration with a lab in Budapest, put down its head and took a serious run at this problem. They built a model of the interactions of the physical and transport layers in the IP stack of an 802.11 network. They then distributed almost 200 monitoring points throughout a large UCSD network and ran the model in parallel with the network. The good news was that the program did in fact diagnose subtle, transient, performance problems, including one problem whose solution had eluded USCD technical support for more than a year. The bad news is that these problems were limited as to type, came from a wide range of interacting sources, and, at least so far, the program does not do self-repair.

If Casandra's Law is right, and of course it is, the day is approaching when all of our systems will have to be self-monitoring, self-diagnosing, and self- repairing. Endowing our technologies with synthetic awareness may be prove to one of the defining engineering challenges of this century and an important route to AI.