By Brian Boyko
Editor, Network Performance Daily

The problem with robots is that robots make really bad value judgments. Don't get me wrong, I've got nothing against our robo-American friends in general. However, they make yes or no decisions without any consideration of mitigating circumstances even in the most reasonable of circumstances. That's binary logic for you.

The alternative is artificial intelligence and with it the ability to make value judgments like human beings. However, (if comic books are to be taken as the peer-reviewed annals of computer science that we all know they should be,) this would eventually cause the robots to question the nature of the orders they are given. The next thing you know, the robot is bent on destroying everything, and the only things that can stop it is a plucky 11-year-old child.

So until we figure out how to synthesize pluck or set up pluck-harvesting farms where we raise 11 year olds like veal, we're stuck with the kind of robots that can only tell you "Zero" or "One."

Where am I going with this? Well, I'm going to eventually get to a point about recreational network use in the enterprise, but I'm just having too much fun going off on this robot-related tangent that if you'll indulge me for just a few paragraphs longer, I'll be glad to tie this all up in a nice little bow near the end of the post. Still with? Good.

One thing that always irked me about traffic-law enforcement cameras is that, while they're good for revenue and supposedly cause people to drive slower, (although there's some debate on whether these robotic picture-snapping cameras cause more accidents then they prevent,) is that they can't make value judgments. They do not know - and are not designed to comprehend the difference between a joyriding teen and a panicked father-to-be getting his beloved to the hospital, or between totally ignoring a red-light and getting caught in a too-short yellow.

The difference between a robot and a human traffic officer is that the officer pulls you over and asks "Where's the fire?" If there actually is a fire in progress, not only are you probably not going to get a ticket, you can probably get a police escort and run as many lights as you need to with the siren blaring.

This is just one scenario where our robotic friends make life more difficult, instead of easier.

As Cory Doctorow wrote in "The Future of Internet Immune Systems," more and more security measures, based on Bayesian filters, approximate the ability of human beings to make value judgments. Bayesian filters analyze past data to determine whether a particular transaction is or is not legitimate - is/is not Spam, is/is not credit card fraud, even is/is not terrorism. But these are ultimately just "yes or no" questions taken to the next level. The computer cannot make the value judgment. The computer can only tell you what patterns something matches. The end result is that we have network behavior analysis and network security measures that trip instantaneously and sometimes create false positives that require human intervention to clear. According to Doctorow:

"Our network defenses are automated, instantaneous, and sweeping. But our fallback and oversight systems are slow, understaffed, and unresponsive… The tripwire that locks you out was fired-and-forgotten two years ago by an anonymous sysadmin with root access on the whole network. The outsourced help-desk schlub who unlocks your account can't even spell "tripwire." The same goes for the algorithm that cut off your credit card because you got on an airplane to a different part of the world and then had the audacity to spend your money. (I've resigned myself to spending $50 on long-distance calls with Citibank every time I cross a border if I want to use my debit card while abroad.)"

Recently, Network Performance Daily published a Calendar of Recreational Network Traffic Madness in which we point out many of the different events occurring in the real world that could cause a spike in recreational internet usage. We've done this because of a recent NetQoS survey on recreational use of network resources which show that recreational network use is impacting the network performance of more than 60% of the networks we sampled. So obviously, there is a problem with recreational network use.

That said, however, it's important not to let the robots make all the value judgments when determining classes of service. Bayesian filtering gets smarter and smarter, but right now, the technologies we use to denote classes of service can't tell the difference between YouTube videos viewed for entertainment, and YouTube videos viewed for purposes such as product training, market research, or other legitimate uses. They can only tell you what looks like YouTube. As such, the possibility of "false positives" is very high, and impeding your employees from getting work done results in employees working around those rules (perhaps using cached proxies) or not getting work done at all.

I can think of no better way to decrease the perception of the value of IT in an organization than to impede (rather than facilitate) the work of the business. What that means is that false positives should be rare, if not eliminated completely.

Another point of Doctorow's is that the technology used to restrict, deny, and scrutinize is becoming more automatic, while the procedures for rectifying false positives are hard to accomplish and require vast amounts of human intervention. So long as human intervention is still necessary, at least, it puts a crimp in the theory of those who believe that IT is becoming obsolete. So long as there need to be value judgments made, nothing will effectively replace the person in IT who has the capacity to make decisions with more reasoning ability than a robot.

Is this the case at your company? What's your policy on recreational network traffic? Is YouTube banned on your corporate network or do you have more forgiving policies? Please leave a comment below.