Give us this day, our daily algorithm adjustments

 

Have you ever wondered about averages and statistics and probabilities?

I know. Of course not. That stuff, if you took any classes involving it, is back in the textbooks with other stuff you knew you were never going to use again once you got out of school. (Take that, geometry!)

Actually, some of it is fascinating. When collected over time, it can deliver some amazing things. It’s the foundation of how Netflix and Amazon make suggestions to you about shows you’ll enjoy and products you should try. It’s also why you look up a propane grill on a home goods site and end up miraculously seeing ads for propane grills all over your social media pages for the next week.

(Well, it’s more than that, truth be told. You know it’s the averages and statistics and probabilities and cookies and snooping around while sharing information you didn’t know you consented to allow the sharing of… but what’s a little privacy invasion between friends when there are some terrific programs to binge. Right?)

One problem I’ve always had with numbers though is that they lie.

Ok, relax.

Please. Hold on for a minute. Stop typing and close your e-mail. When well defined and properly used, numbers do not necessarily lie. We agree on that. No need to send me angry messages.

The problem I’m actually thinking of is created by slightly tilting, definitely twisting, and almost never explaining when the numbers aren’t incorporating the entire picture.

Consider an example.

Let’s say you absolutely love Chinese food. Love it so much that you eat it five to ten or more times for dinner every month. Twice a week on average sounds about right. Every day if offered would not upset you at all. If we’re trying to develop a program that suggests dinner options for you, we’d definitely expect to see some egg rolls and fried rice as the top possibilities.

But what if we started our data collection today? Not yesterday. Not last week. You made pizza today. Tomorrow tacos. Steak will be followed by lasagna to be followed by hamburgers with the last two meals of the week set for chicken on the grill and shrimp scampi.

There’s a really good chance that with this menu in place, the dinner selection offerings from data collection are going to be heavy on Italian cuisine and comfort foods with no Chinese dishes offered at all. If breakfast and lunch orders are tracked, you might find yourself more likely to see Count Chocula high on the list and not a single dumpling to be found.

Over time, the numbers will balance out. Like a probability of flipping a coin that starts off with tails appearing on the first ten tosses, the realities eventually take over the outliers. Harmony restored.

(Of course, tell that to Netflix when you open an account because a friend recommended a really great documentary to you, and now the streaming giant thinks you’re obsessed with serial killers.)

Friend of ours is back out in the dating world. For whatever reason, we were catching up one day and the idea of why we kept getting streaming recommendations for floral design and baking competitions came up. I ended up mentioning the time it took for fake numbers to be balanced out with actual facts. Her response was to make a joke about how it would be nice for people to balance out the personalities on display during a first date, where best behavior and nerves can create a disguise for the jerks that often show up on date number two.

She has a point. Tangent of this thought, totally different essay.

For now though, the reality remains. Numbers lie. Doesn’t mean they can’t be useful. Doesn’t mean they won’t make sense over time. Just means you need to understand how they were obtained.

 

If you have any comments or questions, please e-mail me at Bob@inmybackpack.com