Maybe there is no intelligent life on earth. After the first 25 games of this month’s NCAA basketball tourney, absolutely no one in the $1 billion Quicken/Warren Buffett challenge produced a totally correct bracket. Every single one of the contest’s 8.7 million human entrants made at least one wrong call in trying to predict the games’ winners.
What went wrong? The sports pages are full of rueful comments about various underdogs’ unexpected wins. Scads of basketball-savvy pool participants were caught flat-footed when Mercer (seeded No. 14 in its regional) beat Duke (No. 3) The same goes for the unlikely early triumphs of teams such as North Dakota State (No. 12), Harvard (No. 12) and Dayton (No. 11) over supposedly superior opponents.
But upsets happen. They are part of every NCAA tournament — and in fact they provide much of the populist appeal of watching so many schools’ basketball teams compete. Spend a little time looking at why humanity did so poorly in this particular contest, and an embarrassing truth emerges. A simple, randomized computer program probably would have done a better job of sizing up the upsets than we did.
The word “we,” by the way, is not a rhetorical device. I filled out this rather regrettable bracket myself in what’s officially known as the Quicken Loans Challenge. (Quicken Loans is sponsoring the contest, but the person getting the most brand identification out of the action is Berkshire Hathaway chairman Warren Buffett. He underwrote the insurance policy that provides for a $1 billion payment to anyone picking the right teams to win all 63 games.)
In my case, after two days of play, I’ve been sent to the sidelines, too. Color me lucky to pick Stanford and Harvard as upset winners. Write me off as hopeless for guessing Syracuse would be knocked out in a freakish first-round upset and Duke would survive. Wrong — and wrong again. Overall, I ended up with a truly mediocre 22 out of 32 right in the first two days’ action of March Madness.
Odds-makers early on had figured that no human was likely to make it all the way to, say, the Final Four with a 60-0 winning streak. But seeing everyone obliterated before the end of Day 2 of action — with just 25 of the initial 32 games having been played, is rather pitiful. Here’s why.
If people had filled out their brackets totally randomly, it would have taken 2 to the 25th, or 34 million people, to cover every single possibility. So with 8.7 million entrants, a simple computer program (call it MoronBot) would have enjoyed about a 26% chance of coming up with at least one surviving contender. Take the simple step of steering the program toward the 8.7 million most likely possibilities, based on this historical data about upsets — and MoronBot is likely to do even better.
But we couldn’t improve on random in either its raw or adjusted forms. In fact, maybe our individual efforts to out-think the odds just made it worse. Instead of working as a giant team — entering 8.7 million differentiated brackets and agreeing to split the prize — people by and large acted as solo players, trying to come up with the single bracket that they thought was most likely to win. The biggest team effort that I’ve been able to spot involved just 8,017 people acting in concert.
The result: an enormous amount of duplication, and lots of semi-plausible brackets that never got entered.
The important lesson here is that when data sets get enormous — as is happening in many areas of human endeavor — our classic human desire to be solo pattern-hunters who scour small batches of information may be counterproductive. Sometimes, the best way to get problems solved is to let the algorithms crunch methodically through all the possibilities.
Having just seen all of us do worse than MoronBot, I’m reluctantly starting to think that maybe artificial intelligence deserves a bigger shot at a lot of what we do. Because without it, we’re a pretty dumb planet.