Of course we care about the outcomes. This isn’t necessarily about having perfect predictive power or outplaying the predictor, it’s about winning Newcomb’s problem. 3-Condition Marion, when presented with Newcomb’s problem, runs the first two conditionals which is essentially a check to see how adversarial she can get away with being. If she predicted that she would be able to outgame the predictor at some point, she would take two boxes. However the Predictor is essentially perfect at its job, so the most she predicts being able to do is cause a non-halting recursion in her own decision tree, so that’s no good. That cuts off the option to try and get 1,001,000 out of the Predictor and Marion settles for her second best outcome, which is two conclude she should just take Box B. The Predictor correctly predicts that Marion will employ this algorithm and only take box B, and thus fills Box B. Marion can’t then decide to two-box, she’s already reasoned out that there’s no way for her to game the predictor.
3-Conditional Marion is interesting in part because something like adversarial play emerges from her decision algorithm simply from the fact she’s trying to model the other agent and conditionally respond to her predictions of them. The other agent of course wants to satisfy its own values and block Marion from adversarially going too far, so she wants to calculate exactly how extortionary she can be before the other party defects. She can’t get more than 1,000,000 out of the Predictor without losing 1,000,000 in her attempt for being too greedy and failing to cooperate. The same thing happens in Parfit’s Hitchhiker.
Of course we care about the outcomes. This isn’t necessarily about having perfect predictive power or outplaying the predictor, it’s about winning Newcomb’s problem. 3-Condition Marion, when presented with Newcomb’s problem, runs the first two conditionals which is essentially a check to see how adversarial she can get away with being. If she predicted that she would be able to outgame the predictor at some point, she would take two boxes. However the Predictor is essentially perfect at its job, so the most she predicts being able to do is cause a non-halting recursion in her own decision tree, so that’s no good. That cuts off the option to try and get 1,001,000 out of the Predictor and Marion settles for her second best outcome, which is two conclude she should just take Box B. The Predictor correctly predicts that Marion will employ this algorithm and only take box B, and thus fills Box B. Marion can’t then decide to two-box, she’s already reasoned out that there’s no way for her to game the predictor.
3-Conditional Marion is interesting in part because something like adversarial play emerges from her decision algorithm simply from the fact she’s trying to model the other agent and conditionally respond to her predictions of them. The other agent of course wants to satisfy its own values and block Marion from adversarially going too far, so she wants to calculate exactly how extortionary she can be before the other party defects. She can’t get more than 1,000,000 out of the Predictor without losing 1,000,000 in her attempt for being too greedy and failing to cooperate. The same thing happens in Parfit’s Hitchhiker.