# Newcomb's Paradox **URL:** \ **Category:** Logic & Philosophy of Mathematics\ **Created:** [March 11, 2026, 3:09am UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298 "2026-03-11T03:09:32Z")\ **Posts on this page:** 20\ **Page:** 25
**Author:** ![Michael](https://avatars.discourse-cdn.com/v4/letter/m/4bbf92/32.png) [@Michael](https://www.thephilosophyforum.com/u/Michael)\ **Post date:** [March 15, 2026, 7:09pm UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/482 "2026-03-15T19:09:37Z")
> [@Suny](#): > > Is this written in the description? It’s implied. Just as it’s implied that $1,000,000 is greater than $1,000, and that this isn’t some alien numeral system in which the symbols don’t mean what they mean to us. ---
**Author:** ![Suny](https://avatars.discourse-cdn.com/v4/letter/s/e0b2c6/32.png) [@Suny](https://www.thephilosophyforum.com/u/Suny)\ **Post date:** [March 15, 2026, 7:12pm UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/483 "2026-03-15T19:12:18Z")
whatever you need to believe. You take the IQ test so you agree TB is rational. ---
**Author:** ![Michael](https://avatars.discourse-cdn.com/v4/letter/m/4bbf92/32.png) [@Michael](https://www.thephilosophyforum.com/u/Michael)\ **Post date:** [March 15, 2026, 7:13pm UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/484 "2026-03-15T19:13:16Z")
> [@Suny](#): > > You take the IQ test so you agree TB is rational. The conclusion doesn’t follow and is false. ---
**Author:** ![Suny](https://avatars.discourse-cdn.com/v4/letter/s/e0b2c6/32.png) [@Suny](https://www.thephilosophyforum.com/u/Suny)\ **Post date:** [March 15, 2026, 7:16pm UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/485 "2026-03-15T19:16:02Z")
Well sure it’s not formal, I am using some unstated assumptions. The fact that you take the IQ test because you think it’s rational and that IQ test problem and Newcomb are equivalent but I have yet to see a coherent objection to any of the two. ---
**Author:** ![TheEudemian](https://yyz1.discourse-cdn.com/flex007/user_avatar/www.thephilosophyforum.com/theeudemian/32/1304_2.png) [@TheEudemian](https://www.thephilosophyforum.com/u/TheEudemian)\ **Post date:** [March 15, 2026, 8:31pm UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/486 "2026-03-15T20:31:33Z")
> [@Suny](#): > > That’s not true. If you are a OB (i.e. the predictor predicted you’ll OB) then choosing TB will result in $1,001,000. That is only true if the predictor predicts wrong (namely that you are an OB when you are a TB): the predictor has a 99% chance of predicting correctly, so you wanting to ultimately choose TB despite initially wanting to choose OB would most likely make the predictor classify you as a TB—not an OB. The only way this isn’t the case is if one thinks that there is nothing in phase 1 about oneself that indicates the transition from choosing OB to TB; which only is the case if ontologically the transition was purely random. ---
**Author:** ![TheEudemian](https://yyz1.discourse-cdn.com/flex007/user_avatar/www.thephilosophyforum.com/theeudemian/32/1304_2.png) [@TheEudemian](https://www.thephilosophyforum.com/u/TheEudemian)\ **Post date:** [March 15, 2026, 8:39pm UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/487 "2026-03-15T20:39:58Z")
> [@noAxioms](#): > > It is the choice made, and not any self-prediction of my choice, that the predictor is interested in. Yes, but what choice you will make is informed by what you know, believe, desire, etc.; so that is tied intimately to what the predictor will predict. Even if you knew that you are an OB and decide that you will go TB, that’s all reflected in your beliefs and desires; which the predictor will pick up on to determine, in this case, that you are really a TB. The idea that you can ‘be a TB’ and yet choose ‘OB’ is incoherent: what you are in this case is tied, as you mentioned, foremostly to what you would choose. > [@noAxioms](#): > > Various methods have been suggested as to how this might work, but in the end, the only thing that seems to matter is that however the predictor does its thing, it works. So I again contest ‘crucially’. That’s fair: I agree. > [@noAxioms](#): > > The predictor, if nothing else, can easily detect is intent to deceive, something a true OB type would not do. Exactly. > [@noAxioms](#): > > This wording has strong connotations of retrocausality. The choice does not determine the payout. Something before the prediction is made does. My choice does affect the outcome; because that is what the predictor is predicting with 99% accuracy. Of course, to your point, it doesn’t temporally affect it like a domino hitting another domino; but it’s still a significant aspect of what outcome will happen. > [@](#): > > Do the initials E.S. mean anything to you? Just a hunch. No: what does it stand for? I am more than happy to entertain it if you elaborate on it. ---
**Author:** ![Suny](https://avatars.discourse-cdn.com/v4/letter/s/e0b2c6/32.png) [@Suny](https://www.thephilosophyforum.com/u/Suny)\ **Post date:** [March 15, 2026, 8:42pm UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/488 "2026-03-15T20:42:47Z")
> [@TheEudemian](#): > > That is only true if the predictor predicts wrong (namely that you are an OB when you are a TB) Yes but wasn’t that what would happen in your scenario? You said: > [@TheEudemian](#): > > It is highly probable that you are a OB and that choosing TB will result in $1000 whereas OB will result in $1,000,000. If “it is highly probable that you are a OB” means the predictor predicted you would OB then yeah TB will get you $1,001,000. But even if it doesn’t mean that, it’s not possible that, for any particular agent in front of the boxes, OB would give them $1,000,000 and TB will give them $1,000. Either the money is already in the box or it isn’t. ---
**Author:** ![AlveK](https://yyz1.discourse-cdn.com/flex007/user_avatar/www.thephilosophyforum.com/alvek/32/594_2.png) [@AlveK](https://www.thephilosophyforum.com/u/AlveK)\ **Post date:** [March 15, 2026, 9:16pm UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/489 "2026-03-15T21:16:17Z")
> [@Pierre-Normand](#): > > I’m not saying the predictor is sensitive to a disposition that is somehow “temporally extended” in the sense of being spread across both Phase 1 and Phase 2 as a process. The disposition is a stand **ing** rational capacity. (bold by me) What do you mean by _stand **ing** _ if not that the disposition is spread out across the entire game, both Phase 1 and 2? And the predictor measures P\_1 at t\_1, and anticipates that P\_2 will happen in the future. And then it fills Box 2 accordingly. Then comes time t\_2. The player knows that if P\_2 was anticipated at t\_1, then Box 2 will give them \$x. Or equivalently, they know that if P\_1 was measured at t\_1, then Box 2 will give them \$x. They know that this is in the past. They know that doing P\_2 teaches them about the past, but it does not change it. Doing P\_2 at t\_2 epistemically determines that P\_1 was measured at t\_1. But, it does not ontically determine it, which you have admitted. You have also admitted that **mere** epistemic determination of the past is not sufficient as a basis for decision. So, **how** does doing P\_2 at t\_2 determine that P\_2 was anticipated at t\_1? **What kind of determination is it?** Let’s look at your diagram: ![image](https://canada1.discourse-cdn.com/flex007/uploads/thephilosophyforum/original/1X/9c728c5ff7a13f208f46ee541d4b4404cfcac263.jpeg) **There are no arrows going backwards in time here.** There are no arrows going **from** the rational choice **to** the predictor’s anticipation. Now, my diagram _does_ have such arrows, so I **do** account for the determinations that go backwards in time. I just also show that they’re _mere_ epistemic determinations. Your diagram is completely insufficient. It adds **nothing** but fancy terms. **You have not accounted for HOW the choice determines the anticipation thereof.** You talked of practical determination, and I wanted to formalize this thing you were talking about, this _P-determines._ You completely let go off that, it seems. Why? **Unless the player’s choice does something MORE than non-certainly epistemically determine the predictor’s choice** … then we do not need to take the predictor’s choice into account, because it is a past variable that is now fixed, unchanging and unchangeable. ---
**Author:** ![AlveK](https://yyz1.discourse-cdn.com/flex007/user_avatar/www.thephilosophyforum.com/alvek/32/594_2.png) [@AlveK](https://www.thephilosophyforum.com/u/AlveK)\ **Post date:** [March 15, 2026, 9:23pm UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/490 "2026-03-15T21:23:10Z")
> [@TheEudemian](#): > > Remember, P is the probability of the predictor guessing correctly, which is ~99% in both versions of your thought experiment; which means _your prediction of yourself_ is the predictor, in effect, of which choice you should make. So your knowledge of yourself is what you gauge to determine in probability what such an agent would accurately predict about you. What I think about myself is irrelevant. Let’s say I gauge that I’m a bit OB-ish, and so I think the predictor probably guessed that, and thusly filled Box 2 with $1M. Well, then, what should I do based on this? Of course, grab Box 2 and Box 1. **If there’s $1M in Box 2, that does not entail I shouldn’t take Box 1.** Conversely, if I gauge that I’m a bit TB-ish, and so I think the predictor probably guessed that, and thusly filled Box 2 with $0. Well, then, what should I do based on this? Of course, grab Box 2 and Box 1. **If there’s $0 in Box 2, that does not entail I shouldn’t take Box 1.** In Phase 2, TBing is always better. Being predisposed to TBing in Phase 1, howver, is always worse (for this game specifically, believing in CDT is pretty handy in general). Don’t get the present and the past confused. Don’t get correlation and causation confused. ---
**Author:** ![Pierre-Normand](https://avatars.discourse-cdn.com/v4/letter/p/fbc32d/32.png) [@Pierre-Normand](https://www.thephilosophyforum.com/u/Pierre-Normand)\ **Post date:** [March 16, 2026, 12:42am UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/491 "2026-03-16T00:42:57Z")
I want to address two lines of objection that @Suny and @AlveK have pressed from different angles but that share, I think, a common root. [Suny argues](https://www.thephilosophyforum.com/t/newcombs-paradox/298/415) that if P₁ nomologically leads to P₂ [in my schema](https://www.thephilosophyforum.com/t/newcombs-paradox/298/413), then the agent “simply follows what their past physical state leads them to do” and has “no choice in the room.” [AlveK argues](https://www.thephilosophyforum.com/t/newcombs-paradox/298/489) that my diagram is “completely insufficient” because it contains no arrows running backward from the agent’s choice to the predictor’s anticipation, and therefore fails to account for how the choice _determines_ the anticipation. Let me take these in turn and then draw out the underlying issue. **Suny’s worry: does the P₁→P₂ connection eliminate agency?** @Suny wrote: “Once you are in the room, the agent simply follows what their past physical state leads them to do. So no choice/agency/freedom… there is no decision once you are in the room.” This worry rests on an assumption I don’t share: that if a choice is physically predictable, it isn’t genuinely a choice. Consider an analogy. A chess grandmaster is playing a well-known endgame position. A sufficiently skilled observer, knowing the grandmaster’s style and the position on the board, can predict with near-certainty what move they will play. A Laplacean observer who could read the grandmaster’s brain states could predict it with even greater certainty. Does this mean the grandmaster “has no choice”? That they “simply follow what their past physical state leads them to do”? In one sense, yes. The physical story from brain state to hand movement is nomologically sufficient. But it would be bizarre to conclude that the grandmaster isn’t _deliberating_, or that their assessment of the position plays no role in explaining why they move the bishop rather than the knight. The physical-level description (P₁→P₂) and the person-level description (the grandmaster sees that the bishop move wins) are descriptions of the _same_ process at different levels. The grandmaster’s practical reasoning isn’t something that happens _after_ and _in addition to_ the physical process. Rather, it’s what the physical process _is_, described at the level at which it is rationally intelligible. The Newcomb case has the same structure. When I say P₁ nomologically leads to P₂, I’m not saying the agent is a passive vehicle for physical processes that bypass their reasoning. I’m saying their reasoning _is_ the person-level reality that the P₁→P₂ connection physically realises. The agent in the room is deliberating, and their deliberation is genuinely efficacious. It’s just that its efficacy is not something that floats free of the physical story but is realised by it. You say: “We consider that the physical state in the past doesn’t predetermine the choice in the room, otherwise obviously there is no choice in the room.” But this conflates two things. _Predictability_ does not entail the absence of _agency_. What would eliminate agency is if the agent’s rational assessment of their situation were idle, if the physical process would produce the same output regardless of whether the agent reasoned well or badly. But that’s precisely what doesn’t hold. The agent who reasons well and the agent who reasons badly end up in different physical states. The nomological connection between P₁ and P₂ _runs through_ the agent’s rational capacities, not around them. @AlveK observes that my diagram has no arrows running from the rational choice back to the predictor’s anticipation, and concludes that the diagram “adds nothing but fancy terms.” They ask: “HOW does doing P₂ at t₂ determine that P₂ was anticipated at t₁? What kind of determination is it?” The answer is: it doesn’t. There is no determination running from the choice to the anticipation: not ontic, not epistemic, not any third kind. You are looking for an arrow that my position says does not exist, and treating its absence as a deficiency rather than as the point. What my diagram shows is that the predictor’s anticipation and the agent’s choice both flow from a common source: the agent’s rational grounds, physically realised at P₁ and stable through time. The predictor reads P₁ (the physical realisation of those grounds at prediction time) and anticipates P₂ (their physical realisation at choice time). The agent, when the moment arrives, exercises the very rational capacity that was already physically realised in P₁. There is no backward arrow because neither event determines the other. They are co-expressions of the same rational ground, one via the predictor’s anticipatory reading, the other via the agent’s deliberative exercise. You may feel this doesn’t answer your question. But I think your question already presupposes what I’m contesting: namely, that the only way the agent’s choice can be _relevant_ to the box contents is through some determination (ontic or epistemic) running from the choice to the filling. My position is that the relevance has a different structure entirely. The choice is relevant to the box contents because both are downstream of the same rational ground, not because one acts on the other across time. Let me now draw out what I think is the common root of both objections. Both @Suny and @AlveK, in different ways, picture the agent’s practical situation as one in which they confront their own rational character from outside: as a fact about themselves that they then take into account while deliberating about what to do. Suny frames this as the agent being “forced” to follow their predetermined physical state. AlveK frames it more subtly: the agent “gauges” that they’re “a bit OB-ish,” treats this as evidence about the box contents, and then deliberates about what to do with this information. But this picture splits the agent into two: an observed dispositional self (the one with the OB-ish or TB-ish character) and an observing strategic self (the one who surveys this character and decides how to act in light of it). All the deliberative authority is then given to the observing self, and from _that_ standpoint: the standpoint of someone surveying already-settled facts about character and box contents. There is a useful Aristotelian contrast here. In the _Physics_ (192b23), Aristotle distinguishes things that have their principle of change _within_ themselves (e.g. organisms, natural substances) from things whose principle of change is external, and he notes that a physician who heals _themselves_ is a peculiar case: the art of medicine is the principle of the healing, but it is only _accidentally_ present in the patient, because the physician and the patient happen to be the same person. The physician relates to their own body the way they would relate to any other patient, diagnosing from without, prescribing a treatment. By contrast, when an organism heals _naturally_, the principle of recovery is internal to it _qua_ the very thing that heals. No part of the organism stands outside to prescribe. The teleology is internal. @AlveK’s picture of the deliberating agent, and especially the prescription for the iterated Newcomb case, where the agent must somehow _make_ themselves into a one-boxer despite judging two-boxing to be rational, is the physician-patient. The “strategic self” diagnoses the “dispositional self” and prescribes: “You need to become the kind of agent who one-boxes.” The principle of the change is external to the agent _qua deliberator_; it is only accidentally located in the same person. This is precisely what Ayers meant by “screwing oneself up to the point of acting” as opposed to genuine preactical deliberation. The agent who one-boxes on rational grounds is the naturally healing organism. Their one-boxing isn’t a result of one part of themselves operating on another from without. It is the exercise of their rational nature in these circumstances. The principle of action is internal to them _qua rational agent_. They don’t need to manipulate themselves into one-boxing any more than a healthy organism needs to prescribe itself health. They deliberate about what to do, and one-boxing is what the competent exercise of practical reason yields. And this connects directly to the two ways of knowing what you will do. When @AlveK, in response to @TheEudemian, writes “What I think about myself is irrelevant,” you reveal the standpoint that you are occupying: one from which your own rational character is a feature of the landscape to be surveyed and strategised around, rather than the very capacity through which you deliberate. You imagine gauging that you’re “a bit OB-ish,” treating this as evidence about the probable box contents, and then asking: “Given this information, what should my hand do?” But this is _speculative_ self-knowledge: knowledge derived from observing yourself as you might observe any other object. It’s the physician diagnosing themselves. The alternative is _practical_ self-knowledge. This is the knowledge that is, as Anscombe put it (following Aquinas), “the cause of what it understands.” The agent doesn’t arrive at “I shall one-box” by observing their own OB-ish tendencies and extrapolating. They arrive at it by working out what they ought to do thereby exercising the very rational capacity that the predictor tracked. Their self-knowledge isn’t a report about what their disposition will make them do. It _is_ the exercise of that disposition. When the agent determines “I shall one-box,” this determination is simultaneously self-knowledge (“I know what I will do; I have just now decided it”) and, given the predictor’s sensitivity to the physical realisation of this very rational capacity, it is _also_ world-knowledge (“the predictor, reading the physical signature of these same rational grounds, will have anticipated precisely this”). This is why the question “what should I do given what I know about my dispositions?” already concedes the game to the two-boxer. It frames the agent’s relationship to their own rational character as speculative rather than practical: the physician observing their own symptoms. The right question is simply: “What should I do?” And the answer to that question, arrived at through the exercise of practical reason, with the teleology internal to the deliberation itself, is the very thing the predictor was sensitive to. ---
**Author:** ![noAxioms](https://yyz1.discourse-cdn.com/flex007/user_avatar/www.thephilosophyforum.com/noaxioms/32/658_2.png) [@noAxioms](https://www.thephilosophyforum.com/u/noAxioms)\ **Post date:** [March 16, 2026, 1:50am UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/492 "2026-03-16T01:50:31Z")
> [@TheEudemian](#): > > Yes, but what choice you will make is informed by what you know, believe, desire, etc.; so that is tied intimately to what the predictor will predict. Only if you predict yourself correctly, so I don’t see the point in the exercise. I’m an easy OB, and I have no need to search my inner self to ‘predict’ something I’ve probably already chosen. I’d spend almost no time deliberating, only enough to comprehend the situation. > Even if you knew that you are an OB and decide that you will go TB If you predict X and subsequently choose Y, you’re only fooling yourself, not the predictor. The intent to deceive is plain to see, while it is absent in somebody who actually is an OB. > The idea that you can ‘be a TB’ and yet choose ‘OB’ is incoherent: Not sure. Think of @AlveK who seems to be a TB, but may be open to what he deems irrational choosing OB anyway. The betrayal of one’s nature is not an attempt at deception, but perhaps a triumph of greed over what he sees as rationality. > [@TheEudemian](#): > > No: what does [ES] stand for? I am more than happy to entertain it if you elaborate on it. You have a style similar to somebody I know and respect, but apparently my hunch was wrong. No, this isn’t me implying that I don’t respect you now 🙂 > [@Suny](#): > > This isn’t part of any formulation of the Newcomb problem and makes little sense. You would need to clarify this. It indeed isn’t part of the formulation, but given some knowledge of physics and physiology, it follows from that formulation. Physics: Identical classical systems lacking components sensitive to either quantum uncertainty or chaotic behavior (such as the pi computing example) will tend to evolve along very similar lines. That makes them predictable. Physiology: I cannot think of a single biological feature of any living thing that employs a quantum amplifier. Rational thought rarely has a need to leverage deliberate unpredictability. A chess game is an exception. This scenario is not one of them. An example of a quantum amplifier is what they put in the box with Schrodinger’s cat. An example of chaotic amplification is the flight of a moth or roll of a die. These sorts of things destroy classical predictability, and it’s pretty easy to employ one of them if the goal is to thwart the predictor, which it isn’t. If you deny all this, then you can come up with your own explanation of how such a predictor can have that success rate, such as magic (the kind that prophets use) that allows retrocausality or something. I suggest nothing of the sort. ---
**Author:** ![AlveK](https://yyz1.discourse-cdn.com/flex007/user_avatar/www.thephilosophyforum.com/alvek/32/594_2.png) [@AlveK](https://www.thephilosophyforum.com/u/AlveK)\ **Post date:** [March 16, 2026, 2:13am UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/493 "2026-03-16T02:13:32Z")
(all bold in this post by me) > [@Pierre-Normand](#): > > What my diagram shows is that the predictor’s anticipation and the agent’s choice both flow from a **common source** : (…) No shit. This is called _correlation due to common cause._ It is one of the forms of _non-causative correlation._ **If there is an element that I cannot cause, then it is irrelevant to my decision-making; for my decision-making seeks to cause something.** > [@Pierre-Normand](#): > > The answer is: it doesn’t. There is **no determination** running from the choice to the anticipation: not ontic, not epistemic, **not any third kind**. You are looking for an arrow that **my position says does not exist** , and treating its absence as a deficiency rather than as the point. So, the absolute lack of any consistency in your stance shows that you have nothing but a deepity. Your stance is a bar of soap: the second I grasp it, it flies out of my hand. I can prove that you are contradicting yourself here. With the below quotes, we see that you have been maintaining that the player’s Phase 2 choice **determines** the contents of Box 2, in _some way._ Here: > [@Pierre-Normand](#): > > (…) it is precisely what _makes_ the agent’s Phase 2 deliberation **causally efficacious** with respect to the content of the opaque box. > [@Pierre-Normand](#): > > What there _is_ is a kind of **causation** — **rational** **causation** — that the CDT framework doesn’t recognize. > [@Pierre-Normand](#): > > It is a **practical determination** (…) > [@Pierre-Normand](#): > > The predictor has made the agent’s practical knowledge **causally efficacious** with respect to the very facts the agent is deliberating about. > [@Pierre-Normand](#): > > So **the determination** is neither merely epistemic (“sad news”) nor “ontically retrocausal.” It is **the distinctive determination** that practical knowledge has with respect to what it understands. > [@Pierre-Normand](#): > > You wonder what **semantic space** there might be between ontic and epistemic **determination**. > > (…) > > Within that framework, there genuinely is no **third option** , and I grant that my position must sound a bit like a blur of the two. > [@Pierre-Normand](#): > > You asked: “What changes, then?” I think the question is revealing precisely because it presupposes that **determination** must involve one thing changing another thing. > [@Pierre-Normand](#): > > **Practical determination** , though, is not a matter of _changing_ but of _making it the case_. > [@Pierre-Normand](#): > > My view is not that the agent’s choice stands in **some novel binary relation** to the box contents. > [@Pierre-Normand](#): > > As for **P-determines** : I haven’t given it up. Of course, this _determination_ that you are so determined to not define, it relates to your flirtation with retro-causation: > [@Pierre-Normand](#): > > On the view I’m defending, the diagram needs to be redrawn, not by adding a new arrow, but by collapsing two nodes. > [@Pierre-Normand](#): > > But your Phase 1/Phase 2 distinction draws the cut between past and present at the moment of deliberation and assigns the predictor’s action entirely to the “fixed past” side. > [@Pierre-Normand](#): > > It [CDT] holds the predictor’s action fixed because it treats it as part of “the past,” **causally insulated** from the agent’s **present** reasoning. So, you _fault_ CDT for holding the predictor’s action **causally insulated** from the player’s **present reasoning?** The predictor does **not cause** the player to make their choice in Phase 2. And the player’s choice in Phase 2 does **not cause** the predictor to fill Box 2 accordingly. The former violates the entire premise of the Newcomb Game, and the latter would be retro-causation. Lastly, there is this point of yours I missed: > [@Pierre-Normand](#): > > And the 1% failure rate comes from the predictor’s imperfect sensitivity to those grounds, as your own diagram already shows. Firstly, you have misinterpreted my diagram. For simplicity, I said the predictor had **perfect sensitivity** to the player’s predisposition, but that the player’s predisposition only ontically determined the player’s choice in Phase 2 with 99% likelihood. Flipping this picture is not an alternative. Doing so would imply people’s rational profile never changes, and that people never violate their own rational profile. The only way to disagree with my diagram on that front, then, would be to say that the predictor’s sensitivity is _also_ imperfect, but all that does is add some extra uncertainty to its prediction. It changes nothing fundamental about the scenario. ## Conclusion You do not have a consistent view, Pierre. I hope your fellow OBs, @Michael, @noAxioms, @Sime, @Patterner, @FlannelJesus can see this. All of you, _look at the quotes!_ **There either is a determination between the player’s choice and Box 2’s contents… or there isn’t.** If it’s **merely epistemic** , it is not sufficient, as Pierre has agreed. And if it is causal, then it is retro-causal, which Pierre has also agreed is not correct. Please, let us hold each other accountable. Pierre is claiming a different thing in every post. I have been claiming the exact same things the whole time, and my claims are clear and rigorous. You may disagree with my claims. **However, I expect none of you are confused about _what_ I believe.** Conversely, I think _even Pierre_ is confused about what Pierre believes. ---
**Author:** ![AlveK](https://yyz1.discourse-cdn.com/flex007/user_avatar/www.thephilosophyforum.com/alvek/32/594_2.png) [@AlveK](https://www.thephilosophyforum.com/u/AlveK)\ **Post date:** [March 16, 2026, 3:15am UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/494 "2026-03-16T03:15:49Z")
# 1. Introduction @Michael, @noAxioms, @Pierre-Normand, @Sime, @TheEudemian, @Patterner, @Suny, @Pseudonym, @leo_starboy, @FlannelJesus Unfortunately, I have to go off-site and return to the boring part of my life. This has been a very fun discussion so far, and I hope to return to it in the future. Most people don’t change their mind on the Newcomb Paradox. Not after the first initial phase of discovering it, at least. In a way, we’re all engaged in a fool’s errand; the statistically-unlikely-to-succeed endeavour of changing each other’s minds. Since we’re talking decision theory, I guess our decision to engage in this endeavour could be classified as rather low expected utility… 😂 And yet we do it, because we cannot bear letting declared falsities be unchallenged. OBs look at those _over-thinking, pseudo-intellectual, money-loathing, pretentious TBs_, and they feel a rightful urge to correct what they perceive to be wrong. TBs look at those _under-thinking, pseudo-intellectual, money-crazed, confused OBs_, and they too feel a righftul urge to correct what they perceive to be wrong. The above adjectives are not my opinion of course. They are instead what is often felt by either side regarding the other. But both sides are united in two things: 1. We want the truth. 2. We want the money. I am not accusing anyone of being overly emotional or angry in _this_ _thread_. I simply commenting on the fact that _this_ _topic_ inspires so much anger and emotion in people. I myself have rarely gotten as frustrated and angry in philosophical discourse as I have in this conversation. I would understand it if my opponents have felt that way too at times in this thread. Our passion here is good, but it also makes it harder to grasp the other side’s views. As a TB, of course, I believe the OBs are mistaken about the nature of Version 1 of the Newcomb Game. In line with that view, I’d say they are the ones who would benefit the most from letting go of their emotions. **And I don’t say this from a high horse.** ### I was an OB… When I watched that linked-to Veritasium video for the first time, the answer was obvious. _Who would I rather want to be?_ _A millionaire or a pretensious pseudo-intellectual who shoots themselves in the foot?_ EDT’s expected utility calculations made perfect sense. “Of course we have to take the evidence our actions generate into account!” That was obvious, and of course it had to be right, because _thinking_ _that way_ made me a millionaire. At this point, you could say I was playing Phase 1 of any V2 Newcomb Game potentially in my future very well. **I went to sleep as an OB that night.** **I woke up as a TB.** Really. It had been late at night, and I had gotten so angry at the TBers. But in the morning, I thought about it, and I saw the problem clearly. I saw I had suffered from outcome bias. That’s when I developed my understanding of the problem, and my definition of rationality had to evolve to account for the following fact: **Anyone can construct a hypothetical game that rewards irrationality _before_ the decisions begin.** This fact is trivially true. **You do not need to agree with me on _WHAT_ is irrational, to see that _that which is irrational remains irrational even in a game that rewards it… ASSUMING it rewards it BEFORE the DECISIONS begin…_** This does not prove OBism is wrong. But it is a very important general fact. Irrationality remains irrationality, even if it is correlated with better outcomes in _some_ games, assuming those games _rewarded_ the irrationality **before** the decisions began. Rational decision theory is all about the **decisions** you can make, after all. **You cannot decide the past.** So, what do I want to do with this post? Well, I want to create a hub of links to all of my most important arguments, and then I want to add some extra arguments at the end. # 2. Links to My Past Arguments 1. [My OP](https://www.thephilosophyforum.com/t/newcombs-paradox/298) 2. [The Irrationality Game](https://www.thephilosophyforum.com/t/newcombs-paradox/298/37) 3. [The Time Travelling Rhetoritician](https://www.thephilosophyforum.com/t/newcombs-paradox/298/37) (in same post as above) 4. [CDT is Worth It](https://www.thephilosophyforum.com/t/newcombs-paradox/298/232) 5. [He Has Already TrackED It](https://www.thephilosophyforum.com/t/newcombs-paradox/298/272) 6. [The (Strong) Anti-Newcomb Game](https://www.thephilosophyforum.com/t/newcombs-paradox/298/310) 7. [No Quantumagic 1](https://www.thephilosophyforum.com/t/newcombs-paradox/298/348) 8. [No Quantumagic 2](https://www.thephilosophyforum.com/t/newcombs-paradox/298/371) 9. [Only Two Kinds of Determination](https://www.thephilosophyforum.com/t/newcombs-paradox/298/372) 10. [Response to Michael’s Game](https://www.thephilosophyforum.com/t/newcombs-paradox/298/376) 11. [What is P-determination?](https://www.thephilosophyforum.com/t/newcombs-paradox/298/392) 12. [The Weak Anti-Newcomb Game](https://www.thephilosophyforum.com/t/newcombs-paradox/298/392) (in same post as above) 13. [The Generalized Newcomb Game (intro)](https://www.thephilosophyforum.com/t/newcombs-paradox/298/401) 14. [The Generalized Newcomb Game (example)](https://www.thephilosophyforum.com/t/newcombs-paradox/298/410) 15. [No Relevant Determination of the Past](https://www.thephilosophyforum.com/t/newcombs-paradox/298/489) 16. [Pierre’s Inconsistent View](https://www.thephilosophyforum.com/t/newcombs-paradox/298/493) In addition to this, I think the most important diagram I made is the one below here, as it really sums up the entire TB / CDT position, in my opinion. ![newcomb](https://canada1.discourse-cdn.com/flex007/uploads/thephilosophyforum/original/1X/268dd72294253aaccf213d3aa9c8c985d4af4600.jpeg) The diagram is explained in [What is P-determination?](https://www.thephilosophyforum.com/t/newcombs-paradox/298/392) I genuinely believe that if all the OBs in this thread read through _all_ of those posts thoroughly, including the arguments in _this_ post as well, then I think at least one OB would change their mind. **Of course, I would never expect all that work from anyone for no reason.** But if any of you have lots of time on your hands, and you have the desire, I truly believe that a slow, back-to-back consumption of all of those arguments would change your mind. With all of my main past arguments linked-to in one place, I will proceed to the last arguments I will offer in this thread, for now. # 3. Irrelevant Knowledge Let’s say we have a game G. In this game, there is an unknown U, and it has two possible values: a and b. U is **not a variable** , it is rather an unknown, but constant, thing/fact/number. Now, is our knowledge regarding U relevant to our decision-making in the game G? **Well, there’s a simple check.** 1. What is the rational decision if you know U = a? 2. What what is the rational decision if you know U=b? If the rational decisions in G remain the same in both scenarios, **then all knowledge about U is irrelevant to rational decision-making in the game G.** Can we all agree on this? ## 3.1 Example You are playing a game inside a room, and the single window has its curtains pulled together. The game is this: ### **The 3 Coins** You have **3 coins** and a balance scale Two coins are real. One is a fake that is slightly heavier. You are only allowed to touch two coins. Your task is to figure out which coin is the fake one. * * * Now, let us see if we can find a U such that it is irrelevant to the game. Well, the curtains block your view of the weather outside. We have that U = Sunny Weather, or U = Not Sunny Weather. The rational decision for how to play the 3-coin game is NOT changed whether U = Sunny Weather, or U = Not Sunny Weather. **So, all knowledge regarding the unknown U is rationally irrelevant to the 3-coin game.** # 4. Cheating the Newcomb Game In the Newcomb Game, we have an unknown U. It is a very important unknown U. **U is the amount of money in Box 2.** We can write this as U \in \{\$0, \ \$1\text{M} \}, but U is **not a variable**. Instead, U is an unknown constant dollar amount, and we know its possible values as the elements of the set above. Now, let’s ask ourselves, is knowledge about U **rationally irrelevant knowledge**? Well, in the 3 coin game, we could find that out by pulling the curtains away, and seeing if it affected the rational decision in all possible scenarios: when it is sunny, and when it isn’t. **In both cases, the rational decision remained the same.** We can do the same thing for the Newcomb Game. ## 4.1 Alice the Cheater Alice is in Phase 2 of Version 1 of the Newcomb Game. She is staring at the two boxes. She sees $1000 in one of them, and she looks at the mysterious Box 2. She is unsure what to do. She laments, “Oh! If only I could know what was in Box 2!” The fear of losing the $1M by TBing is really stressing her out. Also, the fear of needlessly leaving $1000 on the table by OBing is also bugging her. She wants some security and clarity. Little did the predictor know, she brought some magical glasses with her. Using them, she could look into Box 2. The predictor did not know of these glasses, and so it did not factor in to his prediction. Alice puts them on and looks into Box 2. It contains… $0. So, she has learned that U = \$0. **What is the only rational decision now?** To say TBing is irrational here would be to say she could retro-causatively change the $0 into $1M. Or worse yet, it would be to say $0 \> $1000. So, Alice TBs, and she feels no regret. **It wasn’t her fault that Box 2 contained $0, after all.** ## 4.2 Bob the Cheater Bob is in Phase 2 of Version 1 of the Newcomb Game. He is staring at the two boxes. He sees $1000 in one of them, and he looks at the mysterious Box 2. He is unsure what to do. He laments, “Oh! If only I could know what was in Box 2!” The fear of losing the $1M by TBing is really stressing him out. Also, the fear of needlessly leaving $1000 on the table by OBing is also bugging him. He wants some security and clarity. Little did the predictor know, he brought some magical glasses with him. Using them, he could look into Box 2. The predictor did not know of these glasses, and so it did not factor in to his prediction. Bob puts them on and looks into Box 2. It contains… $1M. So, he has learned that U = \$1\text{M}. **What is the only rational decision now?** To say TBing is irrational here would be to say she could retro-causatively change the $1M into $0. Or even worse, it would be to say $1M \> $1M + $1000. So, Bob TBs, and he feels no regret. **Why wouldn’t he take the extra $1000? It is free, after all.** ## 4.3 But Cheating Gives an Edge! So? The cheaters, using their advantage, rationally chose to TB. If the knowledge they gained was relevant, then what does it mean that **all they used that extra knowledge for was to TB…** If all players who gain _extra, illegal, relevant information_ all make the same decision, then that decision sounds like it is a very good one. If more information makes everyone take that decision **in a one-move game** , then the truth illuminates it as the best decision for everyone. ## 4.4 The Glasses Break the Premises! It is true that it is a premise you cannot look into Box 2. But this does not prevent us from classifying U as irrelevant knowledge. We are just applying the definition from section 3. Some may go further and say these magical glasses break the premise that the predictor is accurate 99% of the time. The predictor was not aware of the glasses, after all. Well, if thousands of players had access to such glasses, then yes, the predictor would lose all its accuracy, because all OBs would TB. **But it does not logically follow from the mere conception of these glasses that many/all players have them.** We can say that **only** Alice and Bob had them, and taking into account the singular incorrect prediction it caused (Bob TBed instead of OBing), the predictor would still remain very accurate, and we could simply say the premise of 99% already takes into account that one, extra prediction failure. **These glasses are a logically innocent addition to the scenario.** They do violate _one_ premise, which is the ignorance premise, **but it violates it on purpose** , to show this premise was never rationally relevant. **The ignorance of U** is rationally irrelevant because our thought experiment showed that **all knowledge about U is rationally irrelevant.** ## 4.5 Conclusion Section 4 has shown that the money amount in Box 2 is, under the definition in section 3, **rationally irrelevant knowledge.** Everything you know, or can know, about how much money there is in Box 2, be it _absolutely certain knowledge_, or _99% certain knowledge_, is _rationally irrelevant knowledge._ The reason: _knowing what U is does not change the rational decision-making!_ It is an irrelevant unknown. **But it is also an important unknown, because it makes up the bulk of your potential payday.** _Important_ is not the same as _rationally relevant to decision-making within the game_, however. Knowing about U is rationally irrelevant. And you cannot change U either, **not in Phase 2 at least.** If you disagree with this, then it must be one of these things: 1. You think the definition in section 3 is wrong. 2. You think that knowing about the money in Box 2 actually changes what you should do. The second point just flies right in the face of what Alice and Bob just taught us. And well, the first point will be hard to argue. _Rationally irrelevant knowledge_ is **literally defined as knowledge that has no impact on the rational decision-making for that game.** ## 5. Expected Utility Once you have digested section 3 and 4 above, you will be in a good position to understand section 5. The expected utilities of one-boxing and two-boxing are the following: \begin{align} \text{EU(one-box)} &= U \\[2ex] \text{EU(two-box)} &= U + \$1000 \end{align} We have that U is an **unknown** , **a constant** , and that **all knowledge about U is rationally irrelevant knowledge.** Clearly then, the expected utility of two-boxing is highest. It is a free, extra $1000, regardless of whether you won the $1M or not. To make this clearer, let us look at the two possible values for U, given that U \in \{ \$0, \ \$1\text{M} \} ## 5.1 Possibility One Let’s look at the expected utilities when U = \$ 0 \begin{align} \text{EU(one-box)} &= \$ 0\\[2ex] \text{EU(two-box)} &= \$ 0 + \$1000 \end{align} In this case, **two-boxing** is clearly best. ## 5.2 Possibility Two Let’s look at the expected utilities when U = \$ 1\text{M} \begin{align} \text{EU(one-box)} &= \$1\text{M} \\[2ex] \text{EU(two-box)} &= \$ 1\text{M}+ \$1000 \end{align} In this case also, **two-boxing** is clearly best. Taking section 5.1 and 5.2 together, we say that **two-boxing dominates.** ## 5.3 The Non-Possibility OB logic is that of **comparing apples and oranges**. OB logic looks at the two sections above, and protests that the _real_ _situation_ looks more like this: \begin{align} \text{EU(one-box)} &= \$1\text{M} \\[2ex] \text{EU(two-box)} &= \$0+ \$1000 \end{align} In this case, of course, **one-boxing** is clearly best. **But this case is impossible.** If you map the above equations to a physical, classical, spatiotemporal situation, then you are saying that the money can both be $0 and $1M at the same time… or you are saying your decision can cause the past to change, thus causing the money inside Box 2 to change. The fact that one-boxers’ payout usually looks like the first one, and two-boxers’ payout usually looks like the second one, is because **the game was constructed to create a non-causative correlation between the two.** I refer you back to my diagram for the Newcomb Game in section 2, if you want to understand the structure of this non-causative correlation. Remember, **correlation due to a common cause is not itself causation!** [It is a kind of non-causative correlation.](https://en.wikipedia.org/wiki/Correlation_does_not_imply_causation#Third_factor_C_(the_common-causal_variable)_causes_both_A_and_B) # 6. The Bank Transfer Newcomb Game In the above sections, and in most of my former arguments, I’ve been approaching this using logic. In sections 6 and 7, I will approach this using intuition. ## 6.1 The Bank Transfer Consider the normal Newcomb Game, but instead of two boxes, there is instead just $1000 on the table. On the day **before** the Newcomb Game, the predictor predicted whether the player would “OB” or “TB”. If they predicted that the player would OB, then they sent $1M to the player’s bank account. If they predicted the player would TB, then they didn’t. The player has been prevented from seeing any notifications from their bank. **They have no idea if they got it or not.** When the player enters the room, they are told that this game’s equivalent to TBing is **just grabbing the $1000 on the table.** Conversely, OBing means **just exiting the room.** **So OBs, what do you do here?** Do you grab the free $1000 on the table, or do you leave with nothing? The $1M is **already** in your bank account right now, or it **already** isn’t. # 7. The Final Argument Bob is playing Version 1 of the Newcomb Game. He is in Phase 2 right now. He’s staring at Box 2, and he’s stressing. He feels like he’s risking the $1M. But it doesn’t _quite_ make sense to him to _just_ take Box 2. But everytime he thinks like that, his heart races. “Am I a TB? Oh God, then Box 2 **is** probably empty…” “Maybe I should just **play it safe.** I don’t want to **risk** the $1M”. He realizes something. Regardless of if he TBs or OBs, **he is taking Box 2 anyways.** The stress he’s feeling right now could perhaps be stilled if he could just… _secure Box 2._ Not commit to OBing or TBing just yet, but he wants Box 2. He asks the game organizers if it’s fine, and they say there is nothing in the rules preventing his ask. So, he lifts Box 2. It is made to be impossible to feel if there’s anything in it. **He gains no information.** He carries it out of the room and into the neighboring room. His wife sits there and smiles at him. He sits Box 2 down on her lap. He tells her this box belongs to them. **The money or air that is in it, is theirs.** They own it. **Right now, they own it.** He tells her the game is still ongoing, but that she can open it already if she wants… but she has to wait till he’s turned around so he gains no information. And she cannot make any noise. He turns around, and in his peripheral, he sees her opening Box 2. He doesn’t see her facial expression. And she doesn’t make a noise. He exits the room. **He has gained no information.** But his wife has. **She knows if they’ve already won the $1M, or won air.** Bob enters the game room again. Box 1 stands there, filled with the stack of $1000. Bob wonders: **“Should I take them?”** ---
**Author:** ![Pierre-Normand](https://avatars.discourse-cdn.com/v4/letter/p/fbc32d/32.png) [@Pierre-Normand](https://www.thephilosophyforum.com/u/Pierre-Normand)\ **Post date:** [March 16, 2026, 5:02am UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/495 "2026-03-16T05:02:09Z")
> [@AlveK](#): > > Conversely, I think _even Pierre_ is confused about what Pierre believes. @AlveK, just in case you’re still here, I understand why the quotes look contradictory when assembled this way though your context stripping isn’t always charitable. Nevertheless, I appreciate your effort to pin my position down. You present a dilemma: either there _is_ a determination between the player’s choice and the box contents, or there _isn’t_. If there is, it must be ontic (retro-causation) or merely epistemic (insufficient, as I’ve agreed). And since I’ve denied both while also affirming that the agent’s deliberation is “causally efficacious,” I must be contradicting myself. The dilemma is genuine only if “the agent’s deliberation is causally efficacious with respect to the box contents” means “there is a determination running _from_ the choice (qua event at t₂) _to_ the box contents (qua state settled at t₁).” I have never meant it in that sense, and the posts you quote were at pains to explain what I _did_ mean. Here is my position. The agent’s choice does not stand in any binary determination relation (ontic, epistemic, or otherwise) to the box contents. There is no arrow, in any direction, between “The Player’s Choice” and “Box 2’s Contents.” On this, we agree. What I have been saying is that the agent’s _rational grounds_ are the common source from which both the choice and the box contents flow. The choice arises via the agent’s deliberation, the box contents via the predictor’s anticipatory sensitivity to those same grounds. When I said the agent’s deliberation is “causally efficacious with respect to the content of the opaque box,” I was not positing an arrow from the choice to the box. I was saying that the rational grounds _exercised in_ deliberation are genuinely efficacious — not epiphenomenal — and the predictor’s sensitivity to those grounds is what connects them to the box contents. The efficacy runs from the common ground outward to both events, not from one event to the other. Every passage you quoted says the same thing. “Practical determination” and “making it the case” refer to the agent’s determining _what to do_ (i.e. exercising their rational capacity) where the predictor has arranged things so that this rational determination is simultaneously a determination of the box contents, not by retro-causation but by anticipatory sensitivity. You recognise the structure I’m describing since your first response to it was: “No shit. This is called correlation due to common cause.” And on your own diagram, the common cause is “The Player’s Predisposition,” which sits upstream of both the prediction and the choice. So far we agree on the _shape_ of the causal structure. Where we disagree is on the _nature_ of the common cause. On your picture, the predisposition is a fixed past state, a node in Phase 1 that produces its downstream effects through event-event ontic determination. Once it has produced the prediction (and, via the prediction, the box contents), the agent arrives in Phase 2 and confronts a world that is settled independently of their present reasoning. The common cause has already done its work, and the agent’s deliberation adds nothing to the causal picture except the $1,000 choice. This is why you can say: “If there is an element that I cannot cause, then it is irrelevant to my decision-making.” On my picture, the common cause is not a past event but a _standing rational capacity_ — the agent’s cultivated ability to assess their situation and act accordingly. This capacity is present at prediction time (which is why the predictor can read its physical signature) and it is exercised at choice time (which is why the agent acts on it). It is not two events linked by a causal arrow; it is one capacity whose exercise _is_ the choice. The predictor reads the capacity; the agent exercises it. Both the prediction and the choice flow from the same source, but that source is not a dead past state that has already spent its causal energy; it is a rational power whose exercise is the agent’s present deliberation. This is why the standard common-cause reasoning doesn’t apply. In a Reichenbachian fork, the common cause C is an event, its relations to E₁ and E₂ are both instances of dyadic efficient causation, and once you condition on C, E₁ and E₂ are rendered independent. You can then say: “C is fixed; its effects are settled; I now optimise against the settled results.” But on my picture, the agent cannot coherently occupy that standpoint, because one of the alleged “effects” — their own choice — is not a downstream product of the common ground that has already been settled. It is the _exercise_ of the common ground. You cannot screen off a capacity from its own exercise. This is also why I said the diagram should be redrawn by _collapsing_ the predisposition and the choice into a single node: not because they occur at the same time, but because they are not two independent events linked by a probabilistic causal arrow. They are related as a rational capacity to its exercise, and treating them as separate causal handles is the misrepresentation that generates the two-boxing illusion. You say: “If there is an element that I cannot cause, then it is irrelevant to my decision-making; for my decision-making seeks to cause something.” This is the assumption I’ve been challenging throughout. The agent’s decision-making doesn’t seek to _cause_ the box contents in the sense of producing them by efficient causation. It seeks to _determine what to do_. And in the Newcomb setup, the predictor has arranged things so that the agent’s determining what to do is simultaneously a determining of what the box contains not by retro-causation but by the predictor’s sensitivity to the very rational grounds the agent exercises in deliberation. When the agent works out what they ought to do and concludes “I shall one-box,” this isn’t a speculative discovery about an independently fixed world. It is a practical determination that is simultaneously self-knowledge and, given the predictor’s anticipation, world-knowledge at the same time. ---
**Author:** ![Suny](https://avatars.discourse-cdn.com/v4/letter/s/e0b2c6/32.png) [@Suny](https://www.thephilosophyforum.com/u/Suny)\ **Post date:** [March 16, 2026, 5:36am UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/496 "2026-03-16T05:36:08Z")
> [@Pierre-Normand](#): > > The physical-level description (P₁→P₂) and the person-level description (the grandmaster sees that the bishop move wins) are descriptions of the _same_ process at different levels. And we reject this physical description. If you think P\_1 leads only to P\_2 and cannot lead to P\_3 you are foregoing the choice whether or not you acknowledge it. > [@Pierre-Normand](#): > > When I say P₁ nomologically leads to P₂, I’m not saying the agent is a passive vehicle for physical processes that bypass their reasoning. I’m saying their reasoning _is_ the person-level reality that the P₁→P₂ connection physically realises. The agent in the room is deliberating, and their deliberation is genuinely efficacious. It’s just that its efficacy is not something that floats free of the physical story but is realised by it. Yes, so no agency. Being able to draw a single arrow from P\_1 in the past to one P\_2 in the present is what it means to not have agency in the room. > [@Pierre-Normand](#): > > You say: “We consider that the physical state in the past doesn’t predetermine the choice in the room, otherwise obviously there is no choice in the room.” But this conflates two things. _Predictability_ does not entail the absence of _agency_. Never said predictability entailed the absence of agency. It’s not about predictability, it’s about the past state. But again, you are not making it clear what you believe in. > [@Pierre-Normand](#): > > The agent who reasons well and the agent who reasons badly end up in different physical states So here, I get that P\_1 could lead to P\_2 or P\_3 depending on the agent deliberation in the room. But then you say > [@Pierre-Normand](#): > > What my diagram shows is that the predictor’s anticipation and the agent’s choice both flow from a common source: the agent’s rational grounds, physically realised at P₁ and stable through time. The predictor reads P₁ (the physical realisation of those grounds at prediction time) and anticipates P₂ (their physical realisation at choice time). _The agent, when the moment arrives, exercises the very rational capacity that was already physically realised in P₁_. Here we have a different picture. There seems to be something in P\_1 that’s inextricably linked to the agent choice such that it seems the agent simply follows what P\_1 implied. You say it yourself, the agent’s choice flows from the common cause P\_1. And pressed, you’ll say sure but the agent still has agency and you’ll repeat yourself while explaining nothing. I don’t think it should be complicated at all to explain a position on the Newcomb problem in simple few words. NoAxioms said it well himself: “One predisposed to OB cannot TB”. I believe this is what you think but either don’t realize it yourself or you try to obscure it. I am sorry for misreading if not but I really can’t tell what you are saying precisely. If that’s approximately what you think, then as I said you are foregoing agency. If not then I fail to see how the predictor anticipating your rational process makes it so you should OB. Because whatever he did anticipate, I gain $1,000 more by two-boxing. ---
**Author:** ![noAxioms](https://yyz1.discourse-cdn.com/flex007/user_avatar/www.thephilosophyforum.com/noaxioms/32/658_2.png) [@noAxioms](https://www.thephilosophyforum.com/u/noAxioms)\ **Post date:** [March 16, 2026, 6:53am UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/497 "2026-03-16T06:53:46Z")
> [@Pierre-Normand](#): > > @Suny wrote: “Once you are in the room, **the agent simply follows what their past physical state leads them to do**. So no choice/agency/freedom… there is no decision once you are in the room.” > > This worry rests on an assumption I don’t share: that if a choice is physically predictable, it isn’t genuinely a choice. I share your disagreement. It is a genuine choice. It is agency, defined as “action or intervention, especially such as to produce a particular effect”. Suny seems to define agency differently, perhaps what I would distinguish as ‘free agency’, something that I would agree is probably absent in this scenario. What part of Suny’s comment I will agree to is the bolded part. The choice is a genuine choice, but it isn’t free. The agent very much has the agency to do whichever option he chooses, but if freedom is freedom from what past physical states evolved into, one don’t have that since having it would render one not fit for survival. I cannot think of a single decision that is better made utilizing ‘free will’ than using ‘what past physical state leads them to do’, or as is better put: “rational capacity”. Hindsight doesn’t count since not even free will gives you that. I would never want free will, as defined as freedom from said bolded part above. I think the Newcomb case helps illustrate this point since 1) the 99% success rate seems to presume a lack of it, and 2) exercise of free will is not an optimal strategy. The predictor would anticipate an attempt at free will (however one goes about doing that) and would know the intent, and thus predict TB. > _Predictability_ does not entail the absence of _agency_. What would eliminate agency is if the agent’s rational assessment of their situation were idle, if the physical process would produce the same output regardless of whether the agent reasoned well or badly. But that’s precisely what doesn’t hold. The agent who reasons well and the agent who reasons badly end up in different physical states. The nomological connection between P₁ and P₂ _runs through_ the agent’s rational capacities, not around them. Well stated, thanks. The whole post is excellently worded, and I’m jealous of the lack of ability to produce anything similar. > [@AlveK](#): > > **There either is a determination between the player’s choice and Box 2’s contents… or there isn’t.** I notice that @AlveK doesn’t seem to be paying much attention anymore, just doubling down on the assertions instead of addressing comments clarifying reasoning that leads to a more rational outcome. Looking at @Suny’s reply to this, it’s being read, but deliberately misunderstood. The predictor seems to have quite the easy task with all of us. You wonder why even 1% fail. > [@AlveK](#): > > > [@Pierre-Normand](#): > > > > And the 1% failure rate comes from the predictor’s imperfect sensitivity to those grounds, as your own diagram already shows. > > Firstly, … I said the predictor had **perfect sensitivity** to the player’s predisposition, but that the player’s predisposition only ontically determined the player’s choice in Phase 2 with 99% likelihood. I agree with AlveK here. A human is to a very small degree over short timespans subject to chaotic variance. But 99% seems too conservative, since almost everybody’s predisposition in this issue seems to be quite solidly one or the other, so that’s evidence that imperfect sensitivity plays a role. Too few are so on the fence as to be sensitive to chaotic variance. But explaining the source of the predictors error rate isn’t really the point of the exercise. ---
**Author:** ![Suny](https://avatars.discourse-cdn.com/v4/letter/s/e0b2c6/32.png) [@Suny](https://www.thephilosophyforum.com/u/Suny)\ **Post date:** [March 16, 2026, 6:58am UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/498 "2026-03-16T06:58:54Z")
> [@noAxioms](#): > > Suny seems to define agency differently, perhaps what I would distinguish as ‘free agency’, something that I would agree is probably absent in this scenario. it is only absent because you add your own assumptions to the scenario unfortunately ---
**Author:** ![Pierre-Normand](https://avatars.discourse-cdn.com/v4/letter/p/fbc32d/32.png) [@Pierre-Normand](https://www.thephilosophyforum.com/u/Pierre-Normand)\ **Post date:** [March 16, 2026, 7:17am UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/499 "2026-03-16T07:17:13Z")
> [@Suny](#): > > And we reject this physical description. If you think P\_1 leads only to P\_2 and cannot lead to P\_3 you are foregoing the choice whether or not you acknowledge it. I think your response clarifies where we disagree more precisely than anything else in the thread, and I’m grateful for that. You present me with a dilemma: Either P₁ can lead to P₂ _or_ P₃ (genuine agency, but unpredictability), or P₁ leads only to P₂ (predictability, but no agency). I want to show you that this dilemma is not original to the Newcomb problem. It is a compressed version of one of the most famous arguments in the philosophy of free will — van Inwagen’s Consequence Argument — and that pursuing its logic where it leads produces a practical absurdity that I don’t think many people would want to accept. (I had originally drawn this parallel in the second appendix of my 2009 paper _Autonomy, Consequence and Teleology_ that I had published in the first of the three incarnations of The Philosophy Forum.) Van Inwagen argues roughly as follows: if determinism is true, then the agent’s present action is a consequence of past states and natural laws that the agent cannot now alter. Since the agent can alter neither the past nor the laws, they “could not have done otherwise.” Genuine agency requires that the world, rolled back to the _exact same micro-physical state_, could have unfolded differently. Call this “rollback incompatibilism.” Your dilemma has the same structure. You say: if P₁ leads only to P₂, then the agent “simply follows what P₁ implied” and has no agency. Genuine agency requires that P₁ could lead to P₂ _or_ P₃; that the same initial state could yield different outcomes. This is rollback incompatibilism applied to the Newcomb case. Now suppose we accept this, and suppose further that the world is causally closed at the micro-physical level (e.g. that P₁ nomologically determines P₂). What follows for practical reasoning? Since your present action is determined by past physical states and laws you cannot alter, your deliberation is, by the lights of the Consequence Argument, causally idle. The consequences of your action are already settled by the state of the world before you were born. You can no more change them than you can change the orbit of Jupiter. So you have no reason to prefer one action over another _on the basis of their consequences_, since those consequences were going to occur regardless of how you deliberated. What grounds for choice remain? Only something like effort or convenience. If the consequences are going to be the same no matter what (since they’re determined by facts beyond your control), you might as well take the _easiest_ action. Deliberating carefully, weighing reasons, all of this is pointless, because the outcome was settled before you walked into the room. I trust you can see that this is absurd. Nobody actually reasons this way. The chess grandmaster doesn’t say: “My move is determined by my brain state, which is determined by prior physical states I can’t alter, so I might as well move the first piece my hand touches.” But why not? What’s wrong with the argument? The error is in assuming that causal closure at the physical level means the physical story is the _whole_ story and that once you’ve described the P₁→P₂ transition there is nothing left to explain. Return to [my earlier schema](https://www.thephilosophyforum.com/t/newcombs-paradox/298/413). P₁ nomologically determines P₂. M₂ supervenes on P₂. So in one sense, given P₁, it’s settled that a state realising M₂ will occur. But notice something: the fact that P₁ leads specifically to a P₂ that _realises M₂_ (e.g. that realises one-boxing rather than some other physically possible state) is not explained by the P-level story alone. The physical laws tell you that P₁ leads to P₂. They don’t tell you _why_ the transition P₁→P₂ is non-accidentally a transition between realisers of rationally connected mental states. What explains this is that P₁ _realises_ M₁ (the agent’s rational grounds), and M₂ (the agent’s rational choice) intelligibly flows from M₁. The M-level connection, the fact that one-boxing is what these rational grounds call for in these circumstances, is what makes the physical trajectory non-accidentally a trajectory between realisers of rationally connected states. Without that M-level explanation, the fact that the agent’s brain happens to traverse a path from a state realising “rational grounds favouring one-boxing” to a state realising “one-boxing” would be a brute regularity crying out for explanation. So the physical story is causally closed. P₁ suffices nomologically for P₂ (modulo quantum indeterminacies) but it doesn’t _exhaust_ what the agent is and does. The agent is a teleologically organised substance whose rational structure constrains (without violating) the physical trajectory. The physical laws aren’t broken. But the _specific_ path the system follows through its space of physically possible states is non-accidentally one whose person-level description is rationally intelligible. And that non-accidental character is explained by the M-level, not by the P-level. This is why your dilemma is false. You say: either P₁ can lead to different outcomes (agency but unpredictability) or it can’t (predictability but no agency). But there is a third option: P₁ leads to P₂, and the reason it leads _specifically to P₂_ rather than to some other nomologically possible state is that P₁ realises rational grounds from which P₂’s person-level description intelligibly flows. The single arrow from P₁ to P₂ doesn’t bypass the agent’s reasoning. It _is_ the agent’s reasoning but not in the deflationary sense that the reasoning is “nothing over and above” the physical process. The reasoning is the irreducible M-level reality whose rational structure explains why the physical trajectory takes the non-accidental course it does. And this is why I am not a rollback incompatibilist, even though I agree with you that genuine agency is something more than a physical process going where the physics takes it. Rollback incompatibilism says: genuine agency requires that the _same_ P₁ could lead to P₃ instead of P₂. But as many philosophers have pointed out (Daniel Dennett and Robert Kane, among others, calls this “the intelligibility problem”), this gives you randomness, not freedom. If the same rational grounds, the same circumstances, the same agent could produce a _different_ outcome, the outcome is arbitrary rather than free. What genuine agency requires is not that the same P₁ could have gone differently, but that the M-level rational structure is doing real, irreducible explanatory work in determining why the P-level trajectory takes the specific course it does. The agent is free because they are a rational substance whose deliberation genuinely explains their action and not because the physical story could have unfolded otherwise from the same starting point. The two-boxing argument has exactly the same structure as the practical absurdity I described above. You reason: the box contents are determined by the predictor’s past action, which I can no longer alter. My deliberation is therefore causally inert with respect to the box contents. So I should optimise on the only dimension my present action can affect: whether I pick up the extra $1,000. Compare: the consequences of my action are determined by past states and laws I can no longer alter. My deliberation is therefore causally inert with respect to those consequences. So I should optimise on the only dimension I can affect: picking the easiest action. Both arguments treat the past as settled _independently of the agent’s present reasoning_ and conclude that reasoning is idle with respect to the outcomes that matter. Both fail for the same reason: the past that “settles” the outcome is not independent of the agent’s rational character. It _reflects_ that character in the general case because P₁ is the physical realisation of M₁, and the P₁→P₂ trajectory is non-accidentally a transition between realisers of rationally connected states; and in the Newcomb case because the predictor read P₁ (the physical realisation of the agent’s rational grounds) and anticipated P₂ (the physical realisation of the agent’s choice). The predictor reads P₁ and anticipates P₂. They can do this entirely at the P-level. They need not understand _why_ the agent one-boxes. But what explains the fact that P₁ leads non-accidentally to a P₂ that realises _one-boxing specifically_ is that P₁ realises rational grounds favouring one-boxing, and one-boxing intelligibly flows from those grounds. The predictor’s reliability, although it operates at the P-level, is _parasitic on_ the M-level rational structure. If the rational connection between M₁ and M₂ were absent (if, that is, the P₁→P₂ transition merely _happened_ to carry the system from a realiser of “grounds favouring one-boxing” to a realiser of “one-boxing,” without the M-level connection doing any explanatory work) there would be no stable regularity for the predictor to exploit. Agents whose physical trajectories bore no non-accidental relationship to their rational assessments would not be predictable in the way the Newcomb setup requires. So when you say “Being able to draw a single arrow from P₁ to P₂ is what it means to not have agency,” I think you’ve stated the Consequence Argument with great clarity, and I think the Consequence Argument is wrong. Not because determinism is compatible with agency (which is what the compatibilist says), but because the physical story, even when causally closed, does not exhaust what the agent is and does. The M-level rational structure is irreducible, and it is what makes the agent’s deliberation genuinely efficacious rather than epiphenomenal. The single arrow from P₁ to P₂ doesn’t bypass the agent. The agent — the rational, teleologically organised substance — is what _explains_ why that arrow points where it does. And in the Newcomb case, it is precisely this rational structure (exercised by the agent and anticipated by the predictor) that makes one-boxing the rational choice. ---
**Author:** ![Suny](https://avatars.discourse-cdn.com/v4/letter/s/e0b2c6/32.png) [@Suny](https://www.thephilosophyforum.com/u/Suny)\ **Post date:** [March 16, 2026, 7:59am UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/500 "2026-03-16T07:59:03Z")
This isn’t about whether determinism is true or not. This is about choosing the boxes in front of you and if you don’t want to change the question “what box to choose in front of me” into a question about what past state is better to have (and therefore affect the prediction when it is clearly fixed in the scenario) then you assume non-determinism if you want but that’s just what the setup requires. It’s absurd to talk about choosing boxes and decisions when it’s all determined. > You present me with a dilemma: Either P₁ can lead to P₂ _or_ P₃ (genuine agency, but unpredictability), or P₁ leads only to P₂ (predictability, but no agency). I never reject predictability, agency doesn’t imply unpredictability. > What grounds for choice remain? No grounds remain, but that’s not a problem. You will do what you are predetermined to do. > If the consequences are going to be the same no matter what (since they’re determined by facts beyond your control), you might as well take the _easiest_ action. Deliberating carefully, weighing reasons, all of this is pointless, because the outcome was settled before you walked into the room. All of that doesn’t matter, you will do what you are predetermined to do. You are predetermined to reject this argument or whatever, you are predetermined to continue weighing options and act as if you had a choice. There is no absurdity anywhere, that’s simply what you are predetermined to do. At the very least, I guess I am predetermined to not see the absurdity from this text here. > The chess grandmaster doesn’t say: “My move is determined by my brain state, which is determined by prior physical states I can’t alter, so I might as well move the first piece my hand touches.” But why not? What’s wrong with the argument? They don’t because they are predetermined to not reason this way. They are predetermined to reason about their next move and play a good one, it’s all predetermined, that’s just the conclusion. You want to show this gives us an absurdity in the grandmaster reasoning when you already accepted that all they do is predetermined. There cannot be any absurdity, they are simply predetermined to act the way they act. > But there is a third option: P₁ leads to P₂, and the reason it leads _specifically to P₂_ rather than to some other nomologically possible state is that P₁ realises rational grounds from which P₂’s person-level description intelligibly flows. The single arrow from P₁ to P₂ doesn’t bypass the agent’s reasoning. It _is_ the agent’s reasoning but not in the deflationary sense that the reasoning is “nothing over and above” the physical process. Yes and I am telling you, in the scenario P\_1 is fixed so what you want to call “choice” here is about changing P\_1 to get different P\_2. That’s a question we can answer but that’s not the question about what to do in front of the boxes once the prediction and the past states are fixed. > The two-boxing argument has exactly the same structure as the practical absurdity I described above. You reason: the box contents are determined by the predictor’s past action, which I can no longer alter. My deliberation is therefore causally inert with respect to the box contents. So I should optimise on the only dimension my present action can affect: whether I pick up the extra $1,000. I don’t suppose my deliberation is causally inert, why do you say that? I don’t suppose your ‘determinism’ because then I would be trying to answer a nonsensical question about decisions when the agent has no decision to make. > Both arguments treat the past as settled _independently of the agent’s present reasoning_ and conclude that reasoning is idle with respect to the outcomes that matter. Indeed (if I understand this correctly), this is what it means to be able to choose what boxes to pick when the past is fixed. > the past that “settles” the outcome is not independent of the agent’s rational character. No one says it’s totally independent. It just doesn’t force the present to be a certain way (i.e. ‘your determinism’). Not only, we can just reject your determinism, we have to for the question to be meaningful without distorting the question. > So when you say “Being able to draw a single arrow from P₁ to P₂ is what it means to not have agency,” I think you’ve stated the Consequence Argument with great clarity, and I think the Consequence Argument is wrong. Not because determinism is compatible with agency (which is what the compatibilist says), but because the physical story, even when causally closed, does not exhaust what the agent is and does. I don’t know if I have. If the consequence argument is supposed to show something absurd then I don’t agree with the consequence argument. To me, there is nothing absurd about determinism or non-determinism in themselves. It’s just that for the question in Newcomb to be meaningful with the setup (past fixed, prediction fixed), you need to assume some form of non-determinism, i.e. agency. Otherwise again, if you reject this, you will have to say that what the agent _chooses_ to do _causes_ the state P\_1 (and therefore the prediction) to change but that’s retro-causality. And the physical story does exhaust what the agent does. I know there is the idea of compatibilism. I don’t like using “determinism” in this context because I am not sure we are really talking about the same thing as what the word refers to in other situations. I don’t think me saying with these specific definitions, determinism and agency are incompatible implies non-compatibilism in general. Here, agency means having choice in front of the boxes regardless of the past (“it can go both ways even with a fixed past”), and determinism means the past implies the agent will act a specific way. So, to me these two are clearly incompatible. The “Newcomb compatibilist” could say that “choosing in front of the boxes” actually means “defining the agent past such that what the agent picks changes”. I don’t think it’s a stupid question but I argue it’s not the question where it’s rational to TB and it’s not the question asked in the original version of the problem (and maybe most other versions). So we potentially aren’t even in disagreement about the solution of the problem itself. ---
**Author:** ![FlannelJesus](https://yyz1.discourse-cdn.com/flex007/user_avatar/www.thephilosophyforum.com/flanneljesus/32/595_2.png) [@FlannelJesus](https://www.thephilosophyforum.com/u/FlannelJesus)\ **Post date:** [March 16, 2026, 9:11am UTC](https://www.thephilosophyforum.com/t/newcombs-paradox/298/501 "2026-03-16T09:11:41Z")
I also changed my mind. The first time I was introduced to the problem, I was a two-boxer, to the point that it felt obvious to me that I should two-box. [Previous page](https://www.thephilosophyforum.com/t/newcombs-paradox/298.md?page=24) [Next page](https://www.thephilosophyforum.com/t/newcombs-paradox/298.md?page=26)