Operationalizing Newcomb’s Problem

The stan­dard for­mu­la­tion of New­comb’s prob­lem has always both­ered me, be­cause it seemed like a weird hy­po­thet­i­cal de­signed to make peo­ple give the wrong an­swer. When I first saw it, my im­me­di­ate re­sponse was that I would two-box, be­cause re­ally, I just don’t be­lieve in this “perfect pre­dic­tor” Omega. And while it may be true that New­comblike prob­lems are the norm, most real situ­a­tions are not so clear cut. It can be quite hard to demon­strate why causal de­ci­sion the­ory is in­ad­e­quate, let alone build up an in­tu­ition about it. In fact, the clos­est I’ve seen to a real-world ex­am­ple that made in­tu­itive sense is Nar­ra­tive Bread­crumbs vs Griz­zly Bear, which still re­quires a fair amount of sus­pen­sion of dis­be­lief.

So, here I’d like to pro­pose a thought ex­per­i­ment that would (more or less*) also work as an ac­tual ex­per­i­ment.

A psy­chol­o­gist con­tacts you and asks you to sign up for an ex­per­i­ment in ex­change for a pay­ment. You agree to par­ti­ci­pate and sign all the forms. The psy­chol­o­gist tells you: “I am go­ing to ad­minister a poly­graph (lie de­tec­tor) test in which I ask whether you are go­ing to sit in our wait­ing room for ten min­utes af­ter we finish the ex­per­i­ment. I won’t tell you whether you passed, but I will give you some money in a sealed en­velope, which you may open once you leave the build­ing. If you say yes, and you pass the test, it will be $200. If you say no, or you fail the test, it will be $10. Then we are done, and you may ei­ther sit in the wait­ing room or leave. Please feel no obli­ga­tion to stay, as the re­sults are equally use­ful to us ei­ther way. The poly­graph test is not perfect, but has so far been 90% ac­cu­rate in pre­dict­ing whether peo­ple stay or leave; 90% of the peo­ple who stay for ten min­utes get $200, and 90% of those who leave im­me­di­ately get $10.”

You say you’ll stay. You get your en­velope. Do you leave the build­ing right away, or sit in the wait­ing room first?

Does the an­swer change if you are al­lowed to open the en­velope be­fore de­cid­ing?

*I don’t know if poly­graphs are ac­cu­rate enough to make this test work in the real world or not.