Self-Driving Cars and IT’s Automation Problem

February, 2017

GoogleSelfDrivingCar - Self-Driving Cars and IT’s Automation Problem
Most IT departments are embracing automation in one form or another. Some use it in software-defined networks; for load balancing; self-service delivery of virtual machines; or to increase uptime. As we reported last year, automation is even entering into business applications, as Salesforce and other CRM makers are using artificial intelligence (AI) and machine learning to automate email composition. When done well, AI saves money, saves time, and frees humans from tedious tasks. However, there is an interesting and little-discussed downside to automation—human reaction time.

When you embrace automation, you need to factor in that when automation reaches its limit, and humans have to take over, people respond more slowly than they would if there were never any automation to begin with. If, for example, you are using automation to warn the network operation center of a problem that may lead to an outage, you may need to give more lead time than you would if you had no automation in place. This seems counterintuitive, as you’d think a fresh mind coming to a problem could respond more quickly. But that just isn’t how the mind works. To see why, you can look at self-driving cars.

Asleep at the Wheel
Self-driving cars are being developed in stages. Right now, most carmakers could introduce what is known as level 3 self-driving cars. This means that, for the most part, the cars can drive themselves with little human intervention. In fact, they’re so good that it has been reported that Ford engineers are having trouble staying awake while sitting in self-driving cars to evaluate the cars’ performance. But every once in a while, at level 3, when a situation is too complicated the cars need to transfer control to humans. In that case, it is expected that the driver will take over in about 30 seconds.

Here’s the problem—most people need more than 30 seconds to take control of a car they haven’t been driving. Because it is so difficult to take over a car you haven’t been driving, it is one of the reasons Ford, Volvo, and Alphabet are thinking of skipping level 3 altogether and waiting until they can produce fully automated cars. The reason is that humans simply can’t go from not paying attention at all to suddenly driving a car. Volvo research shows that, depending on their level of awareness before they are alerted, drivers can take anywhere from 10 seconds to a full two minutes to be able to take over a car. Response was particularly slow if they were asleep or engrossed in an email or a book. It makes sense: Imagine your alarm clock going off in the morning, and instead of hitting the snooze button, you get dropped into a car going 65 miles per hour.

Hopefully your IT staff isn’t sleeping, but automation means they will be engrossed in other things. And it isn’t just “waking up.” Our brains get on neurological “tracks,” and moving from one track to the next takes time.  A study published in Science Robotics showed that when drivers in self-driving cars were asked to take over, they did a very poor job.  Something as simple as a lane change required multiple “wheel oscillations,” meaning the human driver had to keep correcting because the car hadn’t gotten properly in the lane. They were not turning the steering wheel the appropriate amount, because the brain expected one thing and got another.

When the test drivers started, they were driving at parking-lot speeds before turning control over to the car. The next time the drivers had to take over, the car was at highway speeds. As any driver will tell you, you turn the wheel much less to change a car’s direction at highway speeds. This is the kind of situational awareness your brain needs while driving. In the control tests where drivers took the cars up to highway speeds themselves, the lane change was simple. When they didn’t take the car up to speed themselves, it was a problem. When the brain switches to do another task, it leaves, in effect, a “bookmark” to the first task. If you come back to the first task, it takes time for your brain to correct for the new realities of the situation.

IT Situational Awareness
What does this mean in terms of automation in IT organizations? It probably doesn’t matter a great deal with automated emails. Presumably, knowledge workers can be trained to think before they send emails that were composed with machine learning. But it means a lot in certain operational settings where responding to automation means the difference between an outage and staying online. It is normal for most network operations centers, for example, to have automated warnings for issues that can cause an outage. Automation is so common, that vendors are selling software with pre-existing automation scripts in them that can be customized. But that automation needs to be set up for the way humans interact with an automated system.

When using automation, it is a crucial though seldom-discussed step to build in extra cushion in your lead time for human intervention. Think of it this way: Your support staff is checking email, in a meeting, or even sitting in front of the dashboard but thinking of something else, because the system is on auto-pilot. They’ve “bookmarked” the situation that their network was last in. Suddenly they have to react. They need to regain situational awareness. They need to see all the new realities of the network before they can make adjustments.

It is as if someone blindfolded you, drove you halfway to work, and suddenly took off the blindfold and made you drive. Where are you? What is around you? Where do you need to go? Would you be able to respond quickly enough if, in those few seconds, a school bus were to cut in front of you?

And it isn’t just for major uptime decisions. Every time that simple, scheduled tasks and reports are changed, you must consider what it does to the larger picture in terms of resources, such as network bandwidth, storage or virtual machines. In other words, situational awareness, even when taking your hands off the wheel, is a requirement of IT. As automation increases, the situation will become more complicated, and the interdependencies of systems will become more important. Even as automation allows us to stop doing certain tedious tasks, it asks us instead to do more of the higher-level thinking. It is not a time to figuratively go to sleep because artificial intelligence is in control.

Automation is an essential and powerful tool of the modern data center. Just make sure that as you implement it, you ensure that the blindfold will comes off early enough for your team to avoid the bus.


Image Credit: Wikipedia