Everyone's talking about Volkswagen this week, and I get why, its a genuinely wild story. EPA drops a notice of violation on the 18th, it turns out something like 11 million diesel cars worldwide have this "defeat device" baked into them, and by this morning Martin Winterkorn is out as CEO. Thats a fast unraveling for a company that size. But most of the coverage Ive read treats this as a corporate scandal or an environmental story, which, fine, it is both of those things. What I keep getting stuck on, and what nobody seems to want to talk about directly, is that this was software. Somebody wrote it.
Not "a company decided to cheat" in some abstract boardroom sense. An actual person, or more likely a small team of people, sat down at a desk and wrote code whose entire job was to detect a testing environment and behave differently than it behaves on the road. From what's been reported, the logic checked things like steering wheel position, vehicle speed, the duration of the drive, barometric pressure, basically the fingerprint of a dyno test versus real driving. When it recognized those conditions it would switch the car into a mode with full emissions controls on. The rest of the time, out on actual roads, it would let nitrogen oxide emissions run up to something like 40 times the legal limit. Thats not a bug. Thats a feature, in the ugliest possible sense of that word.
Ive spent enough years writing code for a living to know that nothing like this ships by accident. Somebody had to define the detection heuristics. Somebody had to test that they worked reliably across driving conditions, because a defeat device that gets caught by triggering incorrectly is worse than no defeat device at all. Somebody had to review it, or deliberately not review it, and somebody had to decide the variable names and the comments (if there were any) needed to stay vague enough that a stray audit wouldnt raise flags. That's not one rogue engineer having a bad week. Thats an organization with a QA process pointed at making sure the lie held up under pressure.
I dont know what I'd have done in that engineer's chair, honestly, and I think anyone in this field who claims theyd have obviously refused is being a little smug about it. You've got a manager telling you the targets are the targets, the emissions numbers arent hitting spec any other way, and the company needs this to work. Its a lot easier to say "quit or blow the whistle" from a blog on a Wednesday night than to actually do it when your mortgage depends on the paycheck. Doesnt make it right. Just makes it human, which is somehow worse to think about than if it were pure cartoon villainy.
What bugs me most is how invisible the deception was by design. A mechanic popping the hood wouldnt see it. A regulator running the standard test wouldnt see it, because the standard test is exactly the condition the software was built to recognize. The only way it got caught at all was researchers at West Virginia University running road tests with portable equipment instead of the lab protocol, because they were curious why European "clean diesel" cars performed so well on paper and wanted to see real-world numbers. That's basically a case study in why you dont trust the metric, you trust the thing the metric is supposed to be measuring, and go check yourself when you can.
Anyway. I dont have a tidy takeaway here, other than that I'll be paying closer attention next time some feature demo looks a little too perfect in exactly the conditions its being demoed in. And if you're an engineer reading this and someone hands you a spec that only needs to work "for certification purposes," maybe ask a follow-up question or two before you start typing.
In lighter news, the iPhone 6s review embargo lifted Monday and half my Twitter timeline is now debating whether 3D Touch is the next big thing or the next 3D TV. Ill believe it's useful when I've actually used it for a week, not from a hands-on demo at a press event. My 5s isnt going anywhere yet.