were using field tests

where we take a leaderless group and give them an assignment, a complicated assignment that involves a telephone pole and a wall, and they've got to go from one side of the wall to the other while following some rules.

And there were always two of us.

There were some officers or majors, that sort of rank, who were on their way out of the army and they were assigned to that unit for a while.

And then there were some, quote, psychologists, and I was a psychologist.

So the two of us would examine those people, would observe what was happening.

All they carried, you didn't see ranks.

one.

... have seen?

You see it all over the place that when, you know, one very obvious situation is in job interviews.

So you know, you, you interview someone, and you have a very clear idea of how they will perform.

And even when you're told that your ideas are worthless because in fact you cannot predict performance or can predict it only very poorly, it doesn't affect it.

Next time you interview the person, you have the same confidence.

Uh, interview somebody else.

I mean, that's something that I discovered very early in my career.

Hm.

Well, in the middle of our conversation with him, he volunteered this remarkable comment.

The, the part that's appealing to the public, nudges, that's, that's social psychology, I mean, clearly.

And, you know, Cialdini, uh, should be a father of, uh, of behavioral economics, and he's considered as such.

And so now social psychologists who want to do applied work have to call themselves behavioral economists, so that's the, that's the ultimate irony because that was not what Dick had in mind to begin with.

Could you describe that work?

Well, in that insurance company, the topic came up, actually.

Do their underwriters operate in the same way? So we ran an experiment.

We called those experiments the noise audits, and the idea was they produced some cases, presented those cases to several dozen underwriters, and had each underwriter selected a premium that was appropriate to that risk.

What made the issue interesting was that I also asked the executives in that company how much of a difference they expected, and the question was put as follows.

Suppose you take two underwriters at random.

You compute the average of the two premiums they set, and you divide the difference by the average.

Like, I think that's another case where I think I, I view that as a little bit like the criminal justice system, where, where if I benefit as much from the good stuff as I'm hurt by the bad stuff, then I, I tend not to worry about noise.

I would love to.

If you really want to improve the quality of decision making, use algorithms.

I mean, (laughs) whenever, wherever you can, uh, if you can replace judgments by, by rules and algorithms, they'll do better.

Uh, there, there's big social costs to trusting, allowing algorithm to make decisions, but, but the decisions will likely to be better, so that's one thing.

If you can't use algorithms, then you slow yourself down.

And then there are things that you can do for certain types of problems, and there are different types of problems.

So one class of problems are forecasting problems.

I would love to.