Box A Has $1,000. Box B Has either a Million or Nothing.

Or, yet another “Which box would you pick?” problem

I’ve got a thing for thought experiments. They force us to stop and contemplate, really contemplate. While I find thought experiments involving physics (Einstein’s Train¹, for example) the most fascinating, there are plenty of equally intriguing ones outside of hard science. Laplace’s demon is one of the most popular examples.

Last week, I came across another fascinating thought experiment—Newcomb’s Problem and Two Principles of Choice in the excellent book When Einstein Walked with Gödel. It explores the conflict between rationality and decision-making strategies, and I felt compelled to share it.

So, here it goes.

Newcomb’s Problem started in the 1960s, created by physicist William Newcomb. It wasn’t widely known until philosopher Robert Nozick brought it into the spotlight in 1969.

Newcomb wanted to challenge how we think about making decisions, especially when predicting outcomes gets involved.

It goes like this. There are two closed boxes on the table, Box A and Box B.

Box A has $1,000 in it. Box B has either $1,000,000 or absolutely nothing. You have two choices:

1. Take both boxes

2. Take only Box B

But here’s where it gets interesting (and a little weird).

Imagine a Being—you can think of it as a genie, or a superior intelligence from another planet, or God, or the Entity—the powerful rogue AI from Mission Impossible: Dead Reckoning (this is more realistic). This Being is really good at figuring out what you’re going to do. Like, creepy good. It’s guessed your decisions right every single time before.

Yesterday, the Being made its call: it predicted whether you’d take both boxes or just Box B. And its prediction determined what’s in Box B. If it thought you’d take both boxes, it put nothing in Box B. But if it thought you’d take only Box B, it put a million bucks in there.

It’s that kind of like that situation from Friends. You know this. The Being knows you know this. You know the Being knows you know this.

So… what do you do?