Trusting the Machine

Though I don’t think paper-trail voting machines will fully solve our voting problems, I agree with many voting activists that today’s unverified voting machines are a potential security disaster waiting to happen. That said, it’s become clear to me (and many other voting researchers) over the last few years that non-computer-scientists see the world very differently than computer scientists do.

Most voting officials can’t possibly imagine how a voting machine might incorrectly count votes. After all, it’s been tested! They know, in the abstract, that computers can be “hacked”, but that, in their mind, would require sophisticated technology that only blockbuster movie budgets can afford. In the HBO Bev Harris documentary (75% great, 25% painfully misleading), a voting machine is hacked in front of an election official’s eyes: ballots are scanned, and the wrong total shows up. The official actually breaks down in tears. It’s as if a fundamental law of physics, a pillar of her life, has been shattered: oh my goodness, the machine cannot be trusted!

This is hard for computer scientists to understand. Really, it is. I’m still trying to understand why that voting official broke down in tears. Really, it was that surprising?

And then tonight, I turned on the TV to watch American Idol results (yes, I’m a fan, go David!) I was a few minutes early, and found myself watching the end of some awful Fox show where normal people are strapped to a lie detector, asked brutally personal questions in front of their family and friends, then told whether or not that answer was truthful. If the lie detector claims the contestant lied, she loses all, and the show is over. “Have you ever slept with someone in order to further your career?” “No!” Sorry, the machine says you did, you lose $100,000.

And suddenly, I got it. As a kid, I remember movies with lie detector tests, and I clearly remember that, when a person and a lie detector disagreed, of course I thought the lie detector was right. Sure, now I’m a scientist, and I know not to trust my feeble human instincts: lie detectors can be better than random, but not by much.

In the end, I think human instinct goes like this: the machine has no emotions, the machine is logical, thus the machine cannot be biased and cannot be wrong when it comes to factual issues like “did she lie?” or “how many people voted for Bob?” Picture it for a second, even if you’re a computer scientist: if a lie detector test says someone lied, and they vehemently deny it…. your first reaction is to believe the lie detector, isn’t it?

That’s why computer-based security is so hard. Because, by default, we trust the machine. Even as computer scientists, we quickly forget that it’s just another human being behind the curtain.