Being Googly Is More Important Than Being Right

Googly. Yes, it’s a word, at least within Google. It’s an adjective describing someone who has the appropriate characteristics of a Google employee.

How do I know this? Because the last time I went through the Google recruitment mill, I got up to the point where we were discussing a trip to Mountain View for in-person interviews. That’s when they let me know that one of the in-person interviews was to determine if I was “Googly.”

James Damore was just fired for being insufficiently Googly. He rejected Google’s internal mythology, and worse, he did so with basic math, in a company where mathiness is supposed to be part of the culture.

He also rejected a piece of the general mythology so firmly that what he said was actively misreported — so blatantly that one has to conclude the reporters either can’t read the hard parts of the memo, didn’t bother to read the memo, or somehow managed to see things that weren’t there. (That last is my guess, based on the examples of Trump Trance we’ve seen over the last six months.)

My evidence for this is literally dozens of headlines like this one:

“A Google engineer said women may be genetically unsuited for tech jobs.”

I encourage you to read the whole memo in its unexpurgated form — the one Gizmodo published carefully deleted all his links to sources and all explanatory diagrams — but I think from the reaction, this bullet point is the key statement that got Damore fired:

Differences in distributions of traits between men and women may in part explain why we don’t have 50% representation of women in tech and leadership. Discrimination to reach equal representation is unfair, divisive, and bad for business.

Damore went on to explain this with references to the literature — and sure enough, a number of social scientists concerned with the field of gender differences have supported him publicly, not that it seems to have mattered. But we can actually see that the first sentence is almost certainly true, just from basic statistics. Yes, there’s going to be math, but I promise to try to make it clear even if you’re not a math geek.

The theory of probability, and the whole notion of statistics, got started mathematically in the 17th century, but a lot of the major concepts were developed in the late 18th and early 19th centuries. One of the key notions was the Gaussian distribution — the old familiar bell curve — and the Central Limit Theorem. Both of these are demonstrated in this video of falling pachinko balls.

https://youtu.be/epq-dpMJIxs

As you see, the bouncing pachinko balls, being knocked around by colliding with a bunch of pins, naturally fall into the bins below in a way that makes the piles form the familiar bell curve.

Why? Imagine you’re one of those pachinko balls falling through the pins. Every time you collide with a pin, you either bounce to the right or the left, and the direction is pretty random, like flipping a fair coin. So, to end up far to one end or the other, you have the get a long run of heads or tails, and you wind up in the middle if you get closer to the same number of heads and tails. Now, there is a bunch of math that goes with this that is so cool I can barely resist talking about it more, but that’s not what the article is about. The point is, natural processes tend to have that Gaussian distribution — so much so it’s also called the normal distribution.

The Central Limit Theorem then establishes mathematically that it doesn’t matter what the exact random events are, they will (almost) always sum up to a normal distribution.

What all that means is that if you measure any trait of any population, from individual scores on an IQ test to height to the exact diameter of a bunch of one-inch ball bearings, your measurements are going to have approximately a normal distribution.

Damore said, explicitly, that “differences in distributions of traits between men and women may in part explain why we don’t have 50% representation of women in tech and leadership.” The notion that there is any difference in the distribution of any trait that makes a difference between males and females is the heresy for which Damore had to be burned.

I’m using the word “heresy” purposefully. That notion is a religious belief, like the resurrection of the dead or the 72 virgins or grapes or whatever awaiting the martyr in Paradise. In fact, even if it were true, when you look at a real collection of individuals, it is (almost) always going to have different distributions for the males and females. To see why that’s true, imagine we paint our pachinko balls pink and blue, and have exactly as many pink and blue balls, thoroughly mixed, when we send them through our machine.

Now imagine you are a pink or a blue ball. You don’t know what’s going on with any of the other balls, you just know that you flip a coin to go right or left. But the only way that you will get exactly as many pink balls and blue balls in each bin is if the blue balls get exactly the same sequence of heads and tails. And that’s not the way randomness works.

In other words, even if Google got as many female applicants as males, getting as many females and males with the same traits is extremely improbable.

So now, let’s make this a little more like reality: males and females do have different distributions for traits Google cares about. Now, our pachinko balls do the same little pachinko dance, but the coins aren’t quite fair: the pink balls get coins that come up “bounce left” more often, the blue balls get “bounce right.” The analogy here would be if males really were better at programming than females — with that magic phrase on average. Now, let’s follow a particular pink ball: it starts through the track, and even though its coin is more likely to flip left, this particular time it flips right-right-right … every time. Sure enough, this pink ball will wind up all the way in the right-most bin.

So, assuming you all are still here, what’s the point? Simply that “on average” doesn’t tell you anything about an individual’s traits; all it tells you is how many individuals you’ll find with those traits if you examine lots of them.

Which brings us to that second sentence. “Discrimination to reach equal representation is unfair, divisive, and bad for business.”

Let’s say, first, that the distribution of traits really does mean that men on average are more likely to have that trait than women. (And remember, the science for that is pretty strong.) That means in any population of applicants, more men than women will have that trait. At that point, Google has two choices:

1. They can relax their standards for women only.
2. They can interview as many people as they can, but hire only women.

Yes, both options are discriminatory. The last few days have shown how divisive they are. And it’s pretty hard to argue that turning down qualified candidates or hiring less-qualified candidates can possibly be good for business.

Now, it turns out that roughly 18 percent of all Computer Science graduates are female, and about 20 percent of Google’s programmer hires are women. Objectively, that means that Google is succeeding in hiring in roughly the same proportions as the population of Computer Science students. So, now let’s turn our pachinko game on its head: now all the pachinko balls, pink and blue, have the same fair coin, but there are four times as many blue balls as pink ones. What would the expected outcome be then?

Of course, we would expect there to be one pink ball for every four blue balls in every bin; in particular, the bins far to the right would have roughly four blue balls to every pink ball. The Google situation actually is consistent with the idea that men and women are essentially equal on that trait — you have as many women in the selected group as you have in the source population. But that isn’t the 50/50 split the zealots demand.

This isn’t going to be the end of the argument, and I firmly expect to be denounced for daring to support the math over the feels. But the math is not mocked, and any attempt to beat the math is doomed to fail. The only real question is how many people have to be hurt before we finally realize it.