So you dismiss the possibility that both the choice and the prediction share a common cause?
No, by causally independent, I mean the prediction doesn’t influence causally the choice and the choice doesn’t influence causally the prediction. It is still possible that both share a common cause. That cause would be causally influencing both.
You are still overthinking this. It has nothing to do with causation or determination, whether ontic or epistemic.
You and your clone can choose to either share or take $1,000. If you both choose to share it then you each win $500. If you both choose to take it then you each win $0. If one of you chooses to share it and the other chooses to take it then the one who chose to share it wins $0 and the one who chose to take it wins $1,000.
Your reasoning is: a) if my clone chose to share it then I win $500 more by choosing to take it, and b) if my clone chose to take it then it doesn’t matter what I choose; therefore, c) it is rational to choose to take it.
My reasoning is: d) if I choose to share it then I will almost certainly win $500, and e) if I choose to take it then I will almost certainly win $0; therefore, f) it is rational to choose to share it.
Even though premises (a) and (b) are true, conclusion (c) doesn’t follow; it is refuted by premises (d) and (e). You say this is the game “rewarding irrationality”; I say this is proof that conclusion (f) is true.
1 Like
You’re right that in the artillery case, if the round misses, the written prediction doesn’t change and the predictor is simply wrong. I acknowledge that the artillery analogy is imperfect. Let me try to make it more precise.
In the artillery case, the operator has the power to make the prediction false by firing badly but cannot change what the predictor wrote. In the Newcomb case, the structure is inverted. The agent lacks the power to falsify the predictor’s prediction through rational means (since the predictor is tracking their rational grounds), but they do have the power to ensure that the predictor acted as they would have wished (filling the opaque the box) by being the competent rational agent the predictor took them to be.
Where the parallel does hold is at the level of the cooperation between rational grounds and intended outcome. In the artillery case, the fulfillment of the prediction requires two things: the predictor’s calculation and the operator’s competent execution. The predictor who writes “hit” is relying on the operator to do what their shared understanding of the trajectory demands. In the Newcomb case, walking out with the $1,000,000 likewise requires two things: the predictor’s anticipatory action (filling the box) and the agent’s rational follow-through (one-boxing). The predictor who fills the box is relying on the agent to do what their shared rational grounds demand. In both cases, the “retro-causality” is nothing more than the forward-directed cooperation of two parties who share access to the same rational ground: one acting in anticipation, the other in execution.
Now, the predictor is fallible, and your counterfactual might be read as pointing to the cases where the prediction turns out to be wrong. But the one-boxer’s strategy doesn’t require an infallible predictor. It requires only that the predictor be reliable enough for one-boxing to maximize expected utility. And when the predictor does get it wrong (filling the box for an agent who ends up two-boxing) this isn’t a case of the agent outplaying the predictor and walking away $1,000 richer than they otherwise would have. It is simply a case of predictive failure. Adopting two-boxing as a strategy in order to capture such small and occasional windfalls is self-defeating, since a reliable predictor will mostly anticipate it.
In the Newcomb case, the structure is inverted. The agent lacks the power to falsify the predictor’s prediction through rational means (since the predictor is tracking their rational grounds), but they do have the power to ensure that the predictor acted as they would have wished (filling the opaque the box) by being the competent rational agent the predictor took them to be.
Nope. Look, I am not even sure we are talking about the same problem. That’s why I took a concrete scenario and asked you questions about it and you changed my questions to consider your own special counterfactual. From what I gather, you are picturing a perfect predictor case or a situation where the player doesn’t have agency.
If not then you are simply wrong, the agent can falsify the prediction. The agent probably won’t but it’s a possibility.
I’m not picturing a perfect predictor. I’ve been explicit throughout that the predictor is fallible and that my account requires only significant reliability, not certainty. And I’m not denying agency either. The agent is free to one-box or two-box.
What I am saying is that “the agent can falsify the prediction” is ambiguous in a way that matters. There is a “can” of mere possibility: yes, it is possible that the prediction turns out wrong, just as it is possible that a carefully aimed artillery round misses its target. The predictor is fallible, and on some occasions they will get it wrong. But there is also a “can” of agency. (See John Austin’s delightful paper Ifs and Cans). This is the “can” of something you are able to deliberately bring about. And the agent cannot deliberately falsify the prediction through rational means, because the predictor is tracking precisely those rational means. Any strategy aimed at exploiting the predictor’s fallibility (e.g. “I’ll two-box to cash in on the cases where the predictor wrongly predicted one-boxing”) is itself the kind of reasoning the predictor is sensitive to.
So the predictor’s fallibility is real but not actionable. It doesn’t open up a strategic opportunity for the agent to outplay the predictor. It simply means that the cooperation between the agent’s rational grounds and the predictor’s anticipation of them occasionally misfires, as any predictive process does. The one-boxer’s strategy doesn’t require this cooperation to be perfect. It requires only that it be reliable enough for one-boxing to maximize expected utility.
Let’s go back to our scenario where the agent deliberated and picked TB and won $1,000. I’ll ask a closed question, just yes/no:
If the agent changed their mind after deliberating and decided to pick OB, they would have won $0 instead, do you agree?
No.
(This is filler text since the forum only allows responses that are at least 20 characters long.)
Then we just disagree on what exactly is the Newcomb problem.
I don’t think we disagree on the problem. I think we disagree on the solution. But it’s worth noting that you earlier cited Nozick’s original paper as laying out the case for two-boxing decisively. Nozick himself didn’t think so. By The Nature of Rationality (1993), he had moved to a position where the agent’s decision-value is a weighted combination of causal and evidential considerations, explicitly allowing genuine practical weight to evidential factors. The philosopher who made the problem famous spent over two decades reconsidering his initial two-boxing inclination and never arrived at the confident verdict you attribute to him.
I mention this not as an argument from authority but because it suggests the problem is not as straightforwardly specified in favour of two-boxing as you seem to believe. The questions I’ve been raising about what the predictor is tracking, about the structure of the relevant counterfactuals, and about the kind of causation at work in rational agency, are not departures from the Newcomb problem. They are the problem.
I don’t think we disagree on the problem. I think we disagree on the solution.
No I don’t think so. In my version of the problem and the scenario I presented, the predictor put $0 in the box, then the agent is put in front of the boxes and deliberates. If (the actual scenario) the agent TB then they get $1,000. But if (that’s the counterfactual) they change their mind in the moment and OB then they get $0.
This is not a question of argument or anything, that’s just how the problem plays out in my version. You disagreed so I assume the problem doesn’t play out like this in your version.
Nozick himself didn’t think so. By The Nature of Rationality (1993), he had moved to a position where the agent’s decision-value is a weighted combination of causal and evidential considerations, explicitly allowing genuine practical weight to evidential factors. The philosopher who made the problem famous spent over two decades reconsidering his initial two-boxing inclination and never arrived at the confident verdict you attribute to him.
What verdict did I attribute to him? At best I said that Nozick argued for TB in the paper and that the arguments in the paper are pretty good. Does he explicitly argue for OB or reject TB in the book?
it suggests the problem is not as straightforwardly specified in favour of two-boxing as you seem to believe
It is, or at least the OB arguments haven’t been very convincing to me. To make me doubt would be to show that (at least) one of the claims below is false:
-
Once the prediction OB has been made, in 100% of cases, those who take two boxes win more money.
-
Once the prediction TB has been made, in 100% of cases, those who take two boxes win more money
-
The prediction has already been made and I can choose, therefore me taking two boxes will win me more money.
What is P-Determination?
Your earlier quote heavily suggested it was the eventhood of the relata itself that was the issue.
But fine, I will roll with your rolling definition. I will adjust to your latest reply.
My attempt at formalizing your relation, this P-determines, has the wrong “shape”. What do you mean by that? Let’s see.
Okay, so there are four properties of my earlier suggestion for P-determines that are being mentioned here, and any combination of them may be at fault. So, here they are:
- P-determines is a relation.
- P-determines is binary (it takes two arguments/inputs)
- P-determines is a kind of determination
- P-determines takes events as its arguments/inputs
Let’s look at each property. Is P-determines a relation, in your view?
You talk of “the relata of determination”, though you only object to them being events in that quote. Okay, so determination is something that has relata. What is a relatum? A thing that is related.
If you talk of the relata of some thing, denoted as R for example, then you are saying R has a relata… you are saying R is a relation. If you are making a predicate logic, you’d say R is a predicate with more than one argument.
So, “determines” is a relation. We say X determines Y, making X and Y the relata of the determination relation. Now, P-determines is a kind of determination, obviously. You have been saying all this time that there is some kind of determination between one’s action to TB/OB, and the contents of Box 2. If there was no determination whatsoever, then it’d be irrelevant, no? And to quote you, just in case you’ve forgotten your own stance, here it is:
(bold by me)
So, this determination of yours, what we are calling P-determines, short for both Pierre determines and practically determines… it is a kind of determination, and therefore a relation, by the above argument. It relates things, whatever those things are. Those things may not be events or even tangible, but they are relata nonetheless.
So, we uphold the first claim above. The first claim cannot logically be the claim you are rejecting regarding P-determines. Let’s move on to the second.
- P-determines is binary (it takes two arguments/inputs)
Well, since we know P-determines is a relation now, this reduces to the question: is P-determines a binary relation?
Well, let’s check the textual source here.
So here, you are saying that Thing 1 determines Thing 2. Specifically:
[Picking one box] determines [Content of Box 2]
I don’t see any other relata here, and my arithmetic may be lacking here, but I count one… two! Two relata for you determination relation. Now of course, maybe the above relation is that of mere epistemic determination, but you have already agreed that this is not sufficient alone. So if your response to Suny here had any chance of being relevant, it was because you were specifically talking about P-determination. So, steel-manning your position, you were saying this:
[Picking one box] P-determines [Content of Box 2]
Therefore, P-determines is not just a relation, it is a binary relation. So, let’s move on.
- P-determines is a kind of determination
Now, this one is pretty easy. Several of my quotes above establish this is meant to be a kind of determination. Also, if it isn’t a determination, how would it be relevant?
- P-determines takes events as its arguments/inputs
This one is tricky. You have both seemingly taken issue with this holding between events, but you have also said that isn’t really the issue. However, it seems like I have ruled out the other options, so what else could be the problem than this?
You say I have gotten the shape of P-determines wrong… okay, then map out the shape, please. Explain rigorously, clearly, logically: what this P-determines is, and what it holds, or does not hold, of/between.
The section below will probably inspire a lot of disagreement in you. But I recommend you, fully and decisively deal with this first section before you deal with the section below.
Let us make an account of the exact nature and definition of P-determines before we continue.
Practical Knowledge
To me, this practical knowledge stuff seems related to the Fichtean Tathandlung, or Act-Facts. I am quite partial to this concept, I think it absolutely essential to philosophy. I agree on distinguishing between this kind of self-realizing knowledge, and mere speculative knowledge. I actually study this in my formal system building.
When I am in the room, and I deliberate, I have not decided anything yet. There is no finality to my thinking. Each thought is an action with no other immediate consequence than the thought that follows it.
As I lean more towards TBing, the epistemic determination of Box 2’s contents goes futher and further towards “likely empty”. As my mind swings, and I lean more towards OBing, the epistemic determination of Box 2’s contents goes further and further towards “likely filled with $1M”.
This swinging epistemic determination happens all the while the money, or the air, sits in Box 2, completely unaffected. It is already ontically determined, after all. It cares not for my epistemic determination regarding it. I hate to quote Ben Shapiro, but I reject his monopoly on the following truth: “Facts don’t care about your feelings.” Ontic determination does not care about epistemic determination. The past does not care about what we learn about it.
The above preliminary swinging epistemic determination is never that strong, really. Merely thinking about TBing or OBing is not that strong evidence of what is in the box. If the predictor is 99% accurate, then my Phase 2 pre-decision deliberation will only swing the probabilities up and down within some range that is strictly below that 99% probability.
How do I achieve the 99% probability regarding the contents of Box 2? Through an Act-Fact, of course.
I TB. Or I OB. It is only when the action is done that I have achieved the highest achievable certainty regarding the box’ contents possible, save for actually opening it. Once I have made my decision in an unretracteable way, I know with 99% certainly what is in the box. When I open it, that 99% certainty becomes 100% certainty.
In 1% of cases, the 99% certainty of $0 / $1M being in Box 2 snaps to 100% that $1M / $0 is the case. There can be a flip, and I just though I’d mention that.
So, when I have just TBed/OBed, my action is both the ontic and epistemic determination of something, but they are not of the same thing.
My action to XB is to ontically determine that “I am a Phase 2 XB.” For me to be a Phase 2 XB is heavily correlated (but not causally so) with me being a Phase 1 XB. More precisely, it is correlated with the possibility that, “I was predisposed to XBing in Phase 1”. Not seemingly predisposed, because we can take the strong claim and say that the predictor actually saw my rational profile and absolutely certainly established what my predisposition was, not the mere appearance of what it was. This seems to be the FDT / LDT approach when talking about “tracking the right thing”.
If the predictor saw I was predisposed to XBing in Phase 1, and I YB in Phase 2 instead, we could say the predictor was correct in their assessment of my predisposition, but it just so happened that the time between the predictor’s assessment, and my action, simply changed my predisposition or somehow made me go contrary to it. This only happens 1% of the time in this scenario, but that’s not the point. I am just here modelling this situation like you seem to be: the predictor is not tracking my appearance in Phase 1, it is tracking my actual nature, my rational profile, my true disposition to XB.
So, we could say the predictor is not even primarily predictor, but rather an assessor. The predictor is always right in their assessment, but their assessment can be made outdated by the march of time. It is unlikely to be so, but it could.
Now, I cannot ontically determine my predisposition in the past. I cannot ontically determine anything in the past. I can merely epistemically determine it. I cannot practically determine it either. I can practically determine my actual behavior/categorization right now, through the Act-Fact that is XBing. By XBing in Phase 2, I become an XB in Phase 2, regardless of whether I was predisposed to XBing in Phase 1.
This Act-Fact changes the present, not the past. My predisposition in Phase 1, which is what ontically determined Box 2, is in the past. It cannot be ontically determined, and it cannot be practically determined.
It can only be epistemically determined, to a certainty of 99% for example, by my XBing in Phase 2. And mere epistemic determination is not good enough.
I too value the importance of self-realizing knowledge, of phenomena that are simultaneously facts and acts. But I am not delusional regarding the reach of their influence. They are stuck in the present/future, as far as cause, ontic determination, or practical determination, is concerned.
And how does your philosophy here account for the 1% of people who either do one of two things:
- They OB and get $0
- They TB and get $1,001,000
My CDT thinking fully accounts for how they’re possible, exactly how they came to be, and why they collectively only make up 1% of all participants in V1 of the Newcomb Game (specifically when we say the predictor is only 99% accurate, we could change it and then just change the 1% accordingly).
My thinking easily accounts for them by sticking to only ontic and epistemic determination, by understanding their difference and connection. I am thus able to set up the right links of causal correlations and non-causal correlations.
There seems to be a stubborn misconcepting among OBs. You all say that two thing sharing a common cause means they are causatively correlated.
That is not what causative correlation means. The term causative correlation is just a weaker, or perhaps more precisem version of cause. When we say X is causatively correlated with Y, we are really saying that X causes Y in the future, but it may have some failure rate. It is about letting there be some room for violating the causative connection, but it is still just one thing causing another.
When we merely learn a thing about the future or the past, we do not cause that thing to likely happen, or to likely have happened: we merely cause our knowledge thereof.
When we talk about non-causative correlation, it is to point out that the correlates do not actually impact each other. They do not exchange any energy or particles, they have no ontic determination on one another.
And, yet they are correlated.
Common cause correlation means that there is a third thing that causes them both to be likely.
X is non-causatively correlated with Y due to a common cause
\implies
There is a Z such that:
Z is causatively correlated with X
AND
Z is causatively correlated with Y
Applying this to V1 of the Newcomb Game, I have made a diagram for you to see this more clearly. You are always saying there is a common cause for the predictor’s prediction, and the player’s decision. I agree, but I also know this does change the correct move.
Here, we are modelling the case in which the Player’s predisposition to XBing is causes a 99% likelihood that they will XB in Phase 2. We then say this is the source of the predictor’s 99% accuracy, meaning the predictor actually perfectly measures the player’s predisposition, it is just that this predisposition does not absolutely guarantee this choice in the future. There must be imperfection somewhere, or else this reduces to the perfect scenario, for which everyone here agrees OBing is correct and rational.
Now, in this diagram, all ONTICALLY DETERMINES relations that do not have a probability specified are 100% likely. We are of course rejecting real-life possibilities like the predictor forgetting to put $1M in the box and such, so we see that those perfect ontic determinations are indeed 100% for this idealized thought experiment.
Also, the direction of time goes left-to-right in this diagram. Notice that ontic determination is ALWAYS going left-to-right, which is upholding the unidirectionaly of causality.
The epistemic determination is instead going both directions here, because it is symmetric across time in this scenario! Every epistemic determination here is mutual; symmetric.
The Shape of Your Disagreement
Notice that my above diagram is a collection of directed edges and vertices.
My diagram is a weighted, strongly connected digraph. Almost* all of the vertices who are not directly connected by ontic edges are nonetheless indirectly connected through a chain of ontic edges that are all weighted at 100%, meaning we could add a new ontic edge going directly between them, also weighted at 100%, with no change in the meaning.
Also, using that ontically comprehensive graph (you could call it the ontic transitive closure of the graph), we could draw reciprocal, fully weighted epistemic edges between all those vertices that became connected by ontic transitive closure. This would then create the comprehensive transitive closure of the graph, with no change in the meaning, as all of these extra edges were logically inferred.
In this sense, I have a near-complete, weighted digraph.
All the vertices in this transitive closure of the graph would be epistemically connected. But exactly one pair of vertices would be ontically disconnected:
The node of whether the player gets the $1M is not connected to the node of which choice the player makes in Phase 2. This is my account for their relationship.
And these are precisely the two vertices CDT derives to be ontically unconnected!
As such, this diagram COMPREHENSIVELY represents my entire account of Version 1 of the Newcomb Game. Therefore, if you disagree with my view, you can reduce it to something regarding my above graph, because the above graph captures the entirety of my view.
As such, there are three kinds of disagreements that you could possibly have with my view:
- One or more of the vertices are incorrect.
- One or more of the edges are incorrect.
- One or more vertices/edges are missing from the correct graph-theoretic representation of Version 1 of the Newcomb Game.
Your disagreement with me has to fall into one or more of those three categories.
This leads me to asking a question I think you should probably have the answer to. What two nodes are connected by your P-determines relation? None of them? And what direction does this connection have? Left-to-right? … Or, right-to-left?
@Sime , @FlannelJesus , @Michael , @noAxioms I don’t know if you agree with Pierre, but all of you are thinking somewhat similarly. You have some kind of special determination relation as well, or you don’t. In the former case, you should be able to add to my diagram. In the latter case, you should be able to change something about my diagram, without adding anything, I suspect.
Why don’t you OBs draw diagrams like I do? Why do you not map out the causations and the correlations like I do? Why do you not ascribe epistemic determination to some pairs of things, and ontic determination to (other) pairs of things?
The reason is that IF you tried, you would talk yourself out of a hypothetical $1M by changing your predisposition… and you would also prove to yourself you lost a discussion, which never feels good. But, I would argue that doing this is still very valuable. Having a truthful decision-theory will help you more generally than maintaining the most profitable predisposition for the Newcomb Game.
I mean, that predisposition that helps you so much in the Newcomb Game will harm you in the Anti-Newcomb Game, which I explained in this post. Any predisposition can harm you if the game is definitionally constructed to do so.
But a decision-theory is not judged by whether one can construct a game that, by its very definition, punishes merely the past predisposition towards believing in that decision-theory. If this was the basis for judgement, all decision-theories would be equally bad, because any decision-theory can be “shown to be bad” through this method.
It is the application of the decision-theory that matters to the judgement of its goodness/rationality. In both the Newcomb Game and the Anti-Newcomb Game, applying CDT wins you $1000 more, and not applying CDT wins you $1000 less.
In the Newcomb Game, being predisposed to CDT loses you $1M. Equivalently, in the Anti-Newcomb Game, being predisposed to a non-CDT theory loses you $1M. You are missing the forest for the trees. You are letting the tail wag the dog. You are comparing apples and oranges.
The $1M that OBs won was not attainable to me, as someone predisposed to TBism. Once I reached Phase 2, the $1M was already lost. But that $1000 was attainable, and I attained it. And on the off-chance that I was predisposed to OBism after all, then my Phase 2 TBing won my $1,001,000!
The $1000 that OBs left was a completely free gift for them, as someone predisposed to OBism. Once they were in Phase 2, they had already won the $1M. The $1000 was attainable at no extra cost, other than mental exertion and stress I guess. The $1000 was attainable to OBs in Phase 2, but they did not attain it (usually). And on the off-chance that OB was predisposed to TBism after all, then their Phase 2 OBing won them no more than $0.
When you are in Phase 2, TBing can only help you, and OBing can only harm you.
TBs in Phase 2 focus on the one thing they can control, the $1000. And the Newcomb Game is not constructed to punish this in Phase 2, but to have already punished for the predisposition of it in Phase 1, a phase over which the players had no control.
Looking at the Anti-Newcomb Game (ANG), we see that the uncontrollable variable gets flipped. TBs in Phase 2 of the ANG focus on the one thing they can control, the $1000. And the ANG is not constructed to punish this in Phase 2, nor to have punished the predisposition of it in Phase 1. Instead, it was rewarded in Phase 1.
The past predisposition to TB can be punished or rewarded, but this predisposition is nonetheless always uncontrollable in Version 1, by virtue of being in the past, and those games allowing for no pre-planning.
THE PAST IS THE PAST.
The Weak Anti-Newcomb Game
Some of you may say that the (strong) Anti-Newcomb Game is too unfair. I’d say the strong version is fair in that it lies to all players, but it is unfair in that it rewards TBs in Phase 1. But we have the exact same unfairness in the Newcomb Game.
But, you know what? I don’t even need that Phase 1 unfairness.
The Weak Anti-Newcomb Game:
- Participants do not know about this game, even in theory, before entering the room.
- Before the participants have entered the room, the predictor has predicted whether they will OB or TB.
- Box 2 is always empty.
- The predictor lies, and presents the players with the normal Newcomb Dilemma, with the visible $1000 in Box 1, and the opaque Box 2 that is said to contain $1M if the predictor predicted the player would OB in the past.
Here, all OBs get $0, and all TBs get $1000.
None of the players are given an unfair advantage in Phase 1. In Phase 2, they are presented with the same lie. TBs deal with this lie by recognizing the information it claims is irrelevant to their current decision-making.
OBs deal with this lie by fooling themselves into thinking it is relevant to their current decision-making, thus leaving a completely free $1000 on the table.
Here, the TBs win solely because applying CDT wins over not applying CDT. If you say that OBing is nonetheless rational in this game despite losing, then I ask, Why ain’cha rich?
Or maybe you are underthinking this. You just said my decision-making has nothing to do with causation, or ontic/epistemic determination.
What???
It takes two to share. In the non-communicative version of your Game 3, me-me has no causal effect on other-me.
Therefore, me-me cannot cause BOTH me-me and other-me to share.
Because it takes TWO to share. Me-me can only cause me-me to do something, and causing what only one party does is not sufficient to cause a party of two to share.
Please, I beg of you, stare at my determination diagrams! I control only who I am now, not who I was in the past.
Me-me controls only me-me, NOT other-me.
Your reasoning optimizes for good news. In a game constructed with uncontrollable variables that reward such a predisposition, this will gain you more than for those who didn’t have that predisposition. But once you have entered the decision-making part of the game, you could have gained even more by not optimizing for the news, but rather optimizing for money. But, on the whole, globally speaking, it’s not a bad deal to have this predisposition for those games, of course. It’s a good deal in the games constructed to reward that predisposition before the decisions begin. In other games, it can be detrimental, however.
BOTTOMLINE:
To CHOOSE to SHARE is RATIONAL AND BEST. But me-me cannot choose to share, because it takes two to share, and me-me cannot cause other-me to share. Me-me is a singular agent. Other-me is a different, also singular agent.
With no pre-planning, you are stuck with the locally best option, which is b).
Anyways, please look at my latest reply to Pierre. I tagged you in it. In that post, I lay out a comprehensive determination diagram of Version 1 of the Newcomb Game.
I agree with the P_N comment, but the player has agency since he is free to abandon rational means. It is thus rather simple to thwart the predictor by utilizing this ‘free will’ instead of rationality. The result will be the predictor being wrong half the time, and the player averaging a little over half a million in the game, not an optimal strategy, which is to be expected of a stragegy that isn’t rational.
Anyway, P_N doesn’t assert the predictor is perfect. He just says that rational means isn’t the way to go about it, and I agree with that part.
His reply doesn’t say this, but interpreting it that way is what makes me agree with it.
Clearly deliberation wasn’t complete yet then. The question is self-contradictory. It isn’t complete until the selected box(s) is opened.
Anyway, P_N doesn’t assert the predictor is perfect. He just says that rational means isn’t the way to go about it, and I agree with that part.
I do not care about ‘rational means’ or not. All I want to know is if we even agree on the problem.
Clearly deliberation wasn’t complete yet then. The question is self-contradictory. It isn’t complete until the selected box(s) is opened.
Right, I wanted to say that the deliberation continued and he changed his mind.
Thanks for this detailed argument. I will focus on what appears to me the main issue for now. Your diagram is helpful precisely because it makes explicit the assumption I’ve been challenging. Notice that “The Player’s Predisposition” and “The Player’s Choice” sit as two separate nodes, linked by an ontic determination arrow at 99%. This represents the agent’s rational character as a past state that produces the choice as a downstream effect much as one billiard ball produces the movement of another. The agent at the moment of decision then confronts their own predisposition as a given, a past fact they must reckon with, and the practical question becomes: “Given that my predisposition has already done its work in Phase 1, what should this temporally localized Phase 2 self do now?”
Under that picture, two-boxing is trivially correct, because the only thing the Phase 2 self “controls” is the $1,000. But this is a picture of agency in which the agent is always arriving too late to their own rational life. The predisposition did the real causal work — it caused the prediction, it (at 99%) caused the choice — and the agent at the moment of decision is left to optimize against the consequences of a character they can no longer influence. They are, as it were, pulling the lever of action from outside their own embodied rational dispositions rather than exercising those dispositions in the act of deciding.
This is also why your iterated version of the problem leads to a programme of self-manipulation. If your rational character is something that acts on you from the past rather than something you exercise in the present, then the only way to improve your Newcomb performance is to tinker with the predisposition in advance and to somehow make yourself into a one-boxer without arriving at one-boxing through genuine rational assessment. This is exactly what M.R. Ayers was diagnosing: the end of deliberation is not to screw oneself up to the point of acting, but to determine rationally which course of action to follow.
On the view I’m defending, the diagram needs to be redrawn, not by adding a new arrow, but by collapsing two nodes. The agent’s predisposition and the agent’s choice are not two events linked by a probabilistic causal arrow. They are related as a rational capacity to its exercise the way a skill relates to its use. The agent doesn’t first have a predisposition and then, as a separate event, make a choice that the predisposition probabilistically causes. The agent deliberates, and their cultivated rational character is the capacity they bring to that deliberation, not a prior event that determines its output from behind.
Once you see this, the Phase 1/Phase 2 boundary no longer marks a causal cut between what is fixed and what is open. It marks, at most, a temporal cut, and the question is whether that temporal cut has the practical significance you claim for it. On your diagram, the predisposition sits wholly in Phase 1, the choice sits wholly in Phase 2, and the 99% arrow between them is the only bridge. But from the agent’s own practical standpoint, there is no such gap. The agent who asks “what ought I to do?” is not estimating the probability that their predisposition will cause them to one-box. They are exercising their rational capacity, and the question they face is the actionable one: “What kind of an agent ought I to be in these circumstances?” and not the non-actionable one: “What ought I to be doing differently, given that I am already the kind of agent I am?”
The first question is the question of genuine practical deliberation. The second is the question your diagram forces upon the agent. And it is a question that has no satisfactory answer, because it asks the agent to step outside their own rational agency and optimize against it as though it were someone else’s.
The player’s predisposition in the past is a state-of-affairs time t_1.
The player’s choice in Phase 2 is a state-of-affairs/event at time t_2, where t_2 >t_1.
You want to collapse these two nodes. That would imply that t_1 = t_2, but this is impossible, because t_2 > t_1.
You could get around this with retro-causation, but that wouldn’t even necessarily correspond to collapsing the two nodes, but rather by making the ontic determination between them bi-directional.
Also, notice how you just gave up on your P-determines concept? Did it not exist, after all? I did tell you, if you’d go looking for it, you would find it never existed in the first place.
Now, you are instead talking about collapsing two states-of-affairs into one. This violates spacetime. If you are fine with that, just say so and we can continue on that basis. Also, I find it hard to see where the 1% failure rate would even come in, in your collapsed model.
Please, feel free to draw your own diagram.
@Michael , @noAxioms , @Sime , @FlannelJesus Please tell me, do you guys agree with the quote from Pierre? Do you really think it is truthful to collapse those two nodes into one?
You don’t need to control the other participant’s decision for it to be rational to share.
Just ask yourself these questions:
- If two clones of AlveK were to play this game, what do I predict they each choose and how much do I predict they each win?
- If two clones of Michael were to play this game, what do I predict they each choose and how much do I predict they each win?
You should accept that your clones are most likely to win $0 and that my clones are most likely to win $500.
And you want to insist that your clones are the most rational and that mine are being rewarded for irrationality? That’s irrational.
You are committing the fallacy of judging the rationality of CDT (versus non-CDT) on outcome divorced from which controllable elements were involved, and which uncontrollable elements were involved.
In both the Weak Newcomb Game, and the Strong Newcomb Game, we have that TBs win over OBs.
You insist that OBs are being rational in these games, no? And yet they lose. How is that possible, when rationality is judged on outcome alone, with no analysis regarding the (un)controllable mechanics behind the outcome.
Go to the ends of both the linked-to posts, analyze the games, and please tell me how the OBs are playing rationally, and yet are nonetheless losing.
In those games you lie to them about how it works. That’s a relevant difference.
So to be more precise; if I know all the rules of the game and if strategy A will almost always win me more money than strategy B then strategy A is more rational than strategy B.