Talk:Newcomb's problem

Related articles/concepts

The unexpected hanging paradox likewise deals with knowledge and reasoning about the future.

Focal point (game theory) deals with "attractive" strategies (not very formally put). Elias (talk) 13:22, 7 June 2023 (UTC)

Y Added both to "See also". Thiagovscoelho (talk) 12:49, 27 August 2025 (UTC)

Probabilities

Instead of assuming vague general probabilities of "almost always correct" we can assign numbers to this to make the problem solvable. For example, let's say the predictor is an expert poker player and he has millions of hands of poker as experience/data and his prediction rate of whether someone is bluffing/not is 90%. So his expertise is guessing whether someone will do something they claim they will do.

If his success rate is 90%, then we can deterministically decide which option is optimal. Choosing A+B will result in exactly $1000 90% of the time, and $1001000 10% of the time. So each time we choose A+B we gain on average $100100 (if we choose A+B 100 times, this is the average result. Median result is $1000). If we choose B, the result will be $1000000 90% of the time and $0 10% of the time for a total of $900000 on average. Because $900000 > $100100 then B is the choice which maximizes gain for the chooser.

90% is not considered almost certain by most people, yet B is the correct choice for all values above 90. The truth is, flipping a completely random coin results in A+B = $501,000 while B = $500,000

Even a slightly better than random guesser of 51% accurate will make A+B = 491,000 and B = 510,000 Ergo, if the guesser is even the slightest bit better than random, B will be the correct choice. This is due to the large disparity between the values of $1000 and $1,000,000. If the assigned values are different, then the statistics will be different. — Preceding unsigned comment added by 198.241.159.102 (talk) 19:47, 24 June 2025 (UTC)

The fact that the argument works for "even a slightly better than random guesser" is why vague general words like "almost always correct" are used in the first place. There are actually different versions of Newcomb's problem, from the "limit case" where the predictor is 100% correct to less stark correctness cases.

What you gave is the expected utility argument. The dominance argument says you should always two-box regardless of the predictor's accuracy, and that's the puzzle. I'm not sure how to make this clearer in the article, at the moment. Thiagovscoelho (talk) 12:47, 27 August 2025 (UTC)

Move to "Newcomb's problem"

The following is a closed discussion of a requested move. Please do not modify it. Subsequent comments should be made in a new section on the talk page. Editors desiring to contest the closing decision should consider a move review after discussing it on the closer's talk page. No further edits should be made to this discussion.

The result of the move request was: Moved * Pppery * _{it has begun...} 21:36, 7 September 2025 (UTC)

---

Newcomb's paradox → Newcomb's problem – I propose moving this page to "Newcomb's problem" as that is the more common name in sources. For instance, one of the few books dedicated exclusively to the problem, Ahmed (2018), calls it "Newcomb's problem" in the title. So did the famous paper by Nozick that introduced it. This is regardless of older, now archived talk page discussions about whether it really qualifies as a "paradox". Thiagovscoelho (talk) 12:39, 27 August 2025 (UTC) — Relisting. Jeffrey34555 (talk) 15:37, 3 September 2025 (UTC)

The discussion above is closed. Please do not modify it. Subsequent comments should be made on the appropriate discussion page. No further edits should be made to this discussion.