Microsoft’s Cortana is an NFL game-picking machine, but still a machine

Deep Blue sounds like something it would be OK to get beat by. Even Watson, IBM’s answer-in-the-form-of-a-question brute, has a name that sounds like a winner.

But Cortana? Come on. It sounds like a Chrysler sedan out of the 1980s. But it’s an intelligent machine; Microsoft’s personal assistant in fact, the tech giant’s answer to Apple’s Siri.

And it beat me. And my wife, Alice, too.

Maybe you’ve read about Cortana’s sports prognosticating prowess. With the help of Microsoft programmers and the company’s Prediction Lab, the Windows phone digital assistant nailed its World Cup soccer predictions, calling 15 of 16 matches correctly. So naturally, Cortana turned its game-picking attention to the kind of football you don’t play with your feet — the National Football League.

The Microsoft machine has been predicting the outcome of every NFL game since the beginning of the season. It’s 69-37 or calling it right 65 percent of the time. I couldn’t resist. I had to go head to head with the machine.

I’ve been exploring the intersection of humans and machines and I couldn’t just stand on the sidelines. (Get it?) I decided, starting with Week 7, to go head-to-head with the disembodied voice. And for good measure, I dragged my wife, Alice, into the competition, given that she actually knows something about NFL football.

Cortana

Through six weeks, Cortana had proven itself a worthy opponent. Me? I keep an eye on football through the season, but I tend to focus on a couple of teams — my beloved Chicago Bears and my adopted San Francisco/Santa Clara 49ers. On the other hand, Alice, a lifelong 49ers fan, will watch anybody, and everybody, play — and not just watch, but fully dedicate herself to the unfolding drama. She knows football.

Cortana crunches data — wins; losses; artificial turf or grass; home or away; weather; strength of schedule; team statistics. Then it adds in public sentiment gathered from Facebook and Twitter, an attempt to inject some human wisdom into the process. The idea: If the “crowd” is really big on a team or really down on a team, it could be an indication that something is going on that isn’t reflected in the numbers.

It’s telling, isn’t it, that Cortana’s programmers would find a way, albeit a rather inexact way, to add the human element. I’ve written before about how the man vs. machine construct is outdated. Instead, it’s best to think of the ideal as human + machine.

In fact, it’s also significant that Microsoft turned to a team of humans to make their machine feel more human by mastering the art of chitchat. Now that Cortana is picking football games, they might want to move from chitchat to trash talk (maybe in the next release).

But back to picking NFL games. There are a few caveats. For one, Cortana doesn’t factor in the point spread when picking winners. Maybe that adds a level of complexity that the system isn’t ready for. (Microsoft declined an opportunity to talk to me about the thinking behind Cortana picking NFL games.) Gamblers typically can choose either an underdog, which is spotted points, or a favorite, which gives up points. If the point spread is 6.5, for instance, the favorite must win by seven or more for a bet on the favorite to pay off.

The failure to factor in the point spread left one professional bookmaker completely unimpressed with Cortana’s 65-percent performance.

“I could do that,” he said. In fact, he added, a chicken could do that. Just put the chicken on a paper with the games listed and watch where it pecks. It’s all about the spread; and picking a game with a spread attached takes some significant processing power or brain power.

And it requires something more; something that humans bring to the table and machines just can’t, says the bookie, who didn’t want his name published. (Can you believe it?)

 

“You have your empirical data, which is out there. There is so much data, so many statistics on sports. You could rate so many columns of data, coming down to anything,” he says, including which team wins when a given announcer is calling the game. “The thing that you can’t do, is essentially know what’s happening, to have that inside information. It’s that last little tidbit of information that skews (things). That’s the true element that tips the scale.”

Little things, like knowing which player’s marriage is on the rocks; or who’s battling addiction, are sometimes the keys to the game that don’t show up on television graphics, the bookie says. And the only way to learn about that stuff is to build relationships.

Yes, build relationships, the way humans do.

On the other hand, a machine isn’t going to pick the Bears just because they’re the Bears, which is what I did in my Week 7 contest against the machine. (Dolphins 27; Bears 14.) And Cortana isn’t going to take Cleveland over Jacksonville, based on a long-held bias against expansion teams. No, that would be me. (Did I mention that Jacksonville won 24-6?)

OK, so how did the humans do? I know you’re dying to find out. If you must know, Cortana and I were tied at 10-4 going into the Monday night Texans vs. Steelers game. Alice was at 8-6. Hey, she’s human.

And in the end, so am I. Cortana picked Pittsburgh in the Monday night contest. And wouldn’t you know it? The Steelers won by virtue of a late, second quarter scoring blitz, racking up 21 points in 73 seconds.

Who knew? Cortana apparently. But hey, I gave the machine a run for its money.

The only catch: As with any machine learning system, Cortana is no doubt getting smarter as the season goes along. Me? Not so much.

Cortana image courtesy of Microsoft

Mike Cassidy is BloomReach’s storyteller. Contact him at mike.cassidy@bloomreach.com; follow him on Twitter at @mikecassidy.