Rogue algos and straying servers need human babysitters


If an automated trading system or an algorithm had a theme song it would be George and Ira Gershwin's "Someone to Watch Over Me."

On a tour of IBM's Innovation Center in Hawthorne, New York I saw a demonstration of IBM servers in action. The Polo clad IBMer pointed to a screen and told me to imagine that this was running the back office of a major bank and here are the servers in the back office humming along. All except for one. The IBM system silently glided into action: It identified the lagging server, moved its data and instruction set to other servers in the enterprise and then performed an array of diagnostics to determine the cause of the high-tech flu. According to my guide, the enterprise did not crash or even experience a hiccup as it healed the server without any human intervention.

Cool? Absolutely. Creepy? Kinda. Totaling replacing human beings? Nope, not a chance.

If Wall Street CIOs learned one lesson it's that machines can hum along fine almost 99 percent of the tim, but they need a human overlord to make sure they are doing the right thing at the right time. Sometimes technology built by experts goes awry.

This is where a human comes in.

Algorithms that can go rogue and technology being stubborn - they do what they are programmed to do, dammit - will complete the task they were created to perform. Even if a set of instructions do not make sense or it encounters a variable (i.e. every day life) financial technology can go haywire. It's like a scene in a 60s movie when a robot is kissed by a pretty girl. Steam may not shoot out of your trading platform but something clearly does not compute.

I learned this on a visit to a market maker on the floor of the New York Stock Exchange last fall. This was before Hurricane Sandy so the biggest storm was a rogue algo unleashed by a major investment firm. I asked the market maker - a friendly guy who sounded like he should be shirtless in the stands of a Jets game - what he did when he saw the rogue algo unfold.

"I picked up the phone and called my customers. That's what we do," he told me.

Meanwhile, several other algorithms saw the original rogue algo and assumed that this was a positive market trend and not a potential disaster. The aglos followed the rogue algo until they were reigned in by the men and women on the trading floor.

This is why humans will never leave the trading floor. -Phil