Resisting conclusions

Arguments provide reasons to believe their conclusions. For conclusions we already believe, it's easy to say, "that's a great argument." For conclusions we find unlikely, it takes a strong argument to influence our beliefs.[1]

The strength of evidence required in an argument depends on the likelihood of its conclusion. It doesn't take much to convince an environmentalist that more recycling is valuable. It takes a very strong argument to convince a fiscal conservative that a tax increase is worthwhile. The perceived likelihood of any conclusion, which varies by audience.

We do this without realizing it, in our own heads. We form a hypothesis, estimate its probability, weigh the evidence (and opposing evidence, if we notice it), and then react.

For example, pretend you're a regular attender of a user group for functional programming. A new person walks in to the meeting. Hypothesis: "A developer has come to hear the talks." Likelihood? high. Evidence: presence at user group. Opposing evidence: Dockers and a polo shirt with a company name. Action: welcome him and ask where he works.

Now suppose the person who walked in is also woman. Hypothesis: "A developer has come to hear the talks." Likelihood? low. Evidence: presence; dress is jeans and a tee. Action: say hello and ask her who she recruits for.

It's completely reasonable! I'm not offended. It's reasonable because our internal estimation of "How likely is it that this person is a developer?" is based on all the developers we've seen before. In a k-means based on available information -- appearance and presence -- she's pretty far from the "developer" cluster.

Fortunately, "this person is a developer" is not a strong conclusion, and it doesn't require any action. When the woman says she's a developer, people believe her.

Now consider the hypothesis "This person is a very good developer." That's a stronger conclusion,  requiring actions of respect and listening. Therefore, this conclusion demands stronger evidence. Because that subconscious likelihood estimation still includes information about appearance and gender, it's gonna take much stronger evidence before we believe the woman is a very good developer, compared to the evidence needed to believe the man is a very good developer.

We can't help it, our brains do this, our language does this. Appearance and race and gender and subtle class indicators -- these are determiners of that subconscious likelihood estimation.

The only way to combat this, the way to be fair and relatively impartial, is to pull the evaluation of this hypothesis out of the subconscious. To carefully and methodically evaluate the evidence with our reasoning brain[2]. To ignore our instincts and impressions. To work at making a decision about whether this person is a good developer, instead of going with what feels likely.

This is why people who think they have a sexist bias are less likely to exhibit one: they take precautionary measures.[3] Anyone who thinks they aren't biased may skip looking closely at the data and go with their gut. Our gut is always biased toward what it has seen before.

------------------
[1] I'm talking about inductive arguments, not deductive. This information comes out of the Coursera course "Think Again: How to Reason and Argue"
[2] System 2 in "Thinking Fast & Slow"
[3] there was a study showing evaluations of applicants, and male-named applicants were evaluated higher by both men and women. It's in here somewhere I think