Are You Human?

My sign is vital, my hands are cold

And I'm on my knees looking for the answer

Are we human or are we dancer?

That chorus from the song "Human" by The Killers front man Brandon Flowers was based on a quote from gonzo journalist Hunter S. Thompson: "We're raising a generation of dancers, afraid to take one step out of line." And Thompson would know. If there was anyone who was not a dancer, who not just stepped out of line but leapt, it was him. Indeed, he is perhaps the epitome of that most human of hallmarks: not to follow directions, but rather the compulsion to flaunt them. On the one hand we're taught from an early age to imitate and replicate what our parents and teachers do. On the other we're taught to follow our heart and our mind, which can lead us away from the crowd. In both situations you can get rewarded and punished, dammed if you do and dammed if you don't.

As we enter the AI era it's easy to see how, in the first instance, technology can best us flesh and blood beings. Computers are way better than we are at following orders, at putting one block on top of another, endlessly and forever. Add to that the massive power and scale that comes with access to an essentially infinite source of information and energy, and the race is quickly over. The technology will write for us, solve problems for us, see patterns we can't see and come up with paths we might never have discovered. If what we want is a tireless soldier, this is one of the best.

The flip side, however, is that creativity is not something computers do well. They can replicate, not originate. Sure, they can copy and blend disparate parts to make a new item, but they do it by certain rules and strictures. Our impetuousness to not do what is expected is not something easily coded into an app or a program. It was Picasso that decided that the nose could go there, in spite of all the guidance, history, evidence and common sense that said it went here. As the state-of-the-artificial-intelligence-art keeps getting better, perhaps nonconformity and impulsiveness may turn out to be the "tell" that helps us to differentiate between all it can do and an actual person. 

You see that in the evolution of how we differentiate flesh and blood from silicon and data. The first iterations were the various challenges that popped up when you tried to access your bank account or shop for groceries or book a doctor's appointment. Developed in 1997, those tests were called CAPTCHA, an acronym for "Completely Automated Public Turing Test To Tell Computers and Humans Apart." Pick out all the cats, or write out the squiggly letters, or click on the images that look like turkeys. The thinking was that only people could discern those distinctions, and that would be the hill on which computers would fail.

But the machines kept getting smarter. A study by Google suggested that modern artificial intelligence technology could solve even the most distorted of texts with 99.8% accuracy, and numbers in images with 90% accuracy. And so the quizzes have moved away from those pattern matching benchmarks to what looks to be a simpler approach, the "Click here if you're not a robot" prompt. It's not about the click, however. What you don't see is that once you do that, the system starts to record what you do and how you do it. How you scroll, how you hover, how fast you type or click. That reveals a pattern which is scored for human imperfection and indecision vs. machine efficiency.  Paradoxically, the more you wobble the higher you rank as likely a person, as faking indecision is hard. It's what makes you you and not a robot.

Work is underway to come up with other ways of differentiating between that which moves bits vs. those that bite. One approach is to create a so called "honeypot" form to fill in, where some of the fields are hidden on screen, but the code creating them is still there. As such, a human completing the form would skip over them, while a machine would complete everything, giving away its true nature. Again, the idea is to key in on the inefficiencies and shortcomings that being human is all about, thereby allowing the perfect machines to reveal themselves.

The old standard for judging how smart computers were was the previously referenced Turing test. If a blind panel of observers couldn't tell if they were talking to a human or a machine, the test was passed. Now it turns out that the real test of whether artificial intelligence can imitate a human may not be how smart it is, but if it can pretend to be as dumb as a person. By that metric, I could pass for an android.

-END-

Marc Wollin of Bedford is studying to be seen as a human. His column appears weekly via email and online http://www.glancingaskance.blogspot.com/ and https://marcwollin.substack.com/, as well as via Facebook, LinkedIn and Twitter.