David Weinberger

KMWorld Archive

KMWorld Archive

This column is part of an archive of David Weinberger's columns for KMWorld. Used with permission. Thanks, KMWorld!

Tomorrow there’s a 25% chance of rain.

The latest poll shows Smith at 52% and Jones at 48%, with a 3.6 point margin of error.

Why is it that when it comes to weather or to polling, we routinely expect predictions to come with an explicit expression of the probability that they’re correct, but not with most other predictions? And even with polls, the margin of error is often omitted.

Oh, sure, there are other instances where we expect probability estimates, or at least are not surprised by them. The prognosis for a serious disease might come with probabilities: You’ve got a 70% chance of full recovery, a 20% chance of recovery with some issues, and so forth. In less-fraught situations, we often only include a probability when it is at the extremes of certainty: definitely yes, definitely no, definitely maybe. So, we might say, “There’s a 10% chance I’m going to get a bonus this year,” or “I’m 99% certain I’m getting a bonus,” or “Could go either way.” It is rare that we’d say, “There’s a 35% chance I’m getting a bonus.”

**Predictions are probabilistic**

Yet every prediction is probabilistic. Google Maps (www.google.com/maps) tells you to the minute when it thinks you’ll arrive at your destination, even if your destination is hundreds of miles away. Amazon suggests books to you without explaining what the probability is that you’re actually going to like one. Airplane departure and arrival times usually don’t come with probabilities attached even though airlines track those statistics. Photo apps identify the baby version of you without telling you how confident they are that they’ve identified the photo correctly.

Sometimes, there’s no realistic way to calculate the odds, so we use vague phrases such as, “Even though the season has yet to start, my money is on Boston to win the championship” or, “Unless something goes very wrong, this time next year we’ll have moved to Boulder.” When it’s very vague, we’ll use a word such as “probably,” which is less an assessment of probability than a way of ducking responsibility for an error. We even use “probably” when an outcome is entirely up to our discretion: “I’ll probably just have a pizza delivered.”