(Something I wrote more than 2 years ago.)
The real issue is simply whether we assume that the predictor has real predictive power. It seems many people get caught up with discussing the possibility of such a predictor, leading to questions of determinism, reverse causality, etc., all of which are really irrelevant to the problem of whether the best decision is to “one-box” or “two-box”.
This is what appears to be the fatal flaw of the two-boxer’s argument:
(1) Newcomb’s problem already assumes that the predictor has real predictive power.
(2) As a necessary consequence to (1), box contents are dependent (regardless of mechanism!) on the choice between one-boxing or two-boxing, because the prediction is dependent on that choice.
(3) Effectively, the two-boxer assumes that since the contents have already been decided, choosing one-box or two-box would not affect the contents, and concludes that two-boxing is optimal.
Unbeknownst to the two-boxer, the assumption in (3) actually violates (2) & (1), and therefore the two-boxer’s argument is unsound.
Simply put, the only thing paradoxical about Newcomb’s paradox is the assumption (1) that the predictor has real predictive power. If we simply accept that assumption (and stop distracting ourselves by questioning that assumption), then one-boxing is the obvious, incontrovertible solution!
PS. To be a logically consistent two-boxer, one simply needs to deny assumption (1), which then implies a denial of (2). Perhaps the predictor’s incredible record of correct past predictions were just lucky guesses after all.