Skip to main content
Category: Paradoxes
Type: Decision Theory Paradox
Origin: Introduced by physicist William Newcomb in the 1960s and popularized by Robert Nozick in 1969
Also known as: Newcomb’s Problem, Newcomb’s Game
Quick Answer — Newcomb’s Paradox is a thought experiment in which a highly reliable predictor has already filled an opaque box based on what it thinks you will do: if it predicted you would take only that box, it placed $1,000,000 inside; if it predicted you would take both boxes, it left the opaque box empty and put $1,000 in a transparent box. At decision time, the transparent box visibly contains $1,000; you must now choose between taking just the opaque box (one-boxing) or taking both (two-boxing). One-boxers follow evidential reasoning—your choice is evidence about what the predictor already did—while two-boxers follow causal reasoning—your choice cannot now affect the past—and the puzzle tests which style of decision theory we should trust in high-stakes, prediction-heavy situations.

​
What is Newcomb’s Paradox?

Newcomb’s Paradox is a decision puzzle that exposes a deep conflict between two attractive principles of rational choice. You face two boxes: a transparent box that always contains $1,000 and an opaque box that contains either $1,000,000 or nothing. A super-reliable predictor has already simulated you and filled the opaque box: if it predicted you would take only the opaque box, it put $1,000,000 inside; if it predicted you would take both boxes, it left the opaque box empty. At the moment of choice, nothing physical about the past can change. Still, your decision seems tightly linked to what the predictor must already have done: almost everyone agrees that if you one-box, you almost certainly walk away with $1,000,000, while if you two-box, you almost certainly walk away with only $1,000. The paradox is that standard expected-value reasoning appears to support one-boxing, while standard causal reasoning appears to support two-boxing, even though they cannot both be right about what you should rationally do in a single, fully specified decision problem.
“Newcomb’s Problem is not about greed versus caution; it is about what it means for an action to be the rational one when our choices are entangled with accurate predictions of those same choices.”

​
Newcomb’s Paradox in 3 Depths

  • Beginner: Imagine a game show with two boxes. A nearly perfect AI has already guessed whether you will take one box or two. If it predicted you would take only the big opaque box, it secretly put $1,000,000 inside; if it predicted you would grab both, it put nothing there, but you can still see $1,000 in the clear box. If the AI is really that good, people who one-box tend to get $1,000,000, and people who two-box tend to get only $1,000—yet, at the last second, taking both boxes still feels tempting.
  • Practitioner: In real decisions, like whether to adopt a risky strategy your competitors can anticipate, your payoff often depends on how others predict your behavior. Newcomb-style problems formalize this: your current move is statistically linked to what others have already done in response to their model of you. Thinking only in terms of “my choice cannot change the past” misses that others’ past actions already encode assumptions about your present policy.
  • Advanced: In contemporary decision theory, Newcomb’s Paradox separates evidential decision theory (EDT), causal decision theory (CDT), and more recent formulations like functional or policy-based decision theories. EDT says you should one-box because doing so is strong evidence that the predictor filled the box; CDT says you should two-box because, conditioning on the predictor’s past action, your current move does not causally affect the money. Newer approaches attempt to reconcile these views by treating your choice as the output of an algorithm that the predictor has already analyzed, aligning Newcomb’s problem with issues in game theory, AI alignment, and Bayesian Thinking.

​
Origin

Newcomb’s problem originated with physicist William Newcomb, who proposed the basic setup in the 1960s as an unpublished puzzle. Philosopher Robert Nozick brought it to wide attention in a 1969 paper, using it as a challenge case for standard theories of rational choice and expected utility. Nozick himself did not offer a definitive solution; instead, he highlighted the clash between seemingly self-evident principles. The paradox is often described informally as a contest between “one-boxers” and “two-boxers.” One-boxers point to the empirical track record: if the predictor is almost always right, then agents who one-box systematically walk away richer than those who two-box. Two-boxers respond that at the moment of choice, the contents of the opaque box are already fixed, so taking the extra transparent box cannot possibly make you worse off in any possible world. In the decades since Nozick, Newcomb’s Paradox has become a central example in decision theory, alongside the Prisoner’s Dilemma and St. Petersburg Paradox. It appears in debates over Expected Value, rational choice under uncertainty, and how to model agents whose decisions are predictable by powerful observers—ranging from ideal Bayesians to superintelligent AI systems.

​
Key Points

Before using Newcomb’s Paradox as a lesson in strategy or ethics, it helps to isolate its structural features.
1

Predictive Entanglement Between Choice and Outcome

The predictor’s action is no longer causally influenced by your current choice, yet your payoff depends on how closely your choice matches what was predicted. This creates a predictive entanglement: an evidential link between what you now do and what was earlier done, even though causal arrows point only from past to future. Any satisfying account of rational choice must decide how much weight to give such correlations.
2

Evidential vs. Causal Decision Principles

Evidential decision theory recommends the action that would make it most likely that you are in a good world—here, a world where the box is full. Causal decision theory recommends the action that best changes outcomes, holding fixed the past. In Newcomb’s setup, EDT one-boxes and CDT two-boxes, even if both accept the same probabilities about the predictor’s accuracy.
3

Policy-Based and Functional Perspectives

More recent approaches, sometimes called functional, policy, or updateless decision theories, evaluate not individual acts but entire policies or algorithms. The idea is that the predictor has already simulated your policy; by choosing the one-boxing policy in general, you make it the case that accurate predictors tend to fill the box for people like you. This perspective tries to keep CDT’s focus on structure while recovering EDT’s success in Newcomb-style environments.
4

Real-World Analogues in Reputation and AI

Newcomb-like structures appear whenever your current action influences how future predictors, partners, or systems treat you. Building a reputation for “keeping your word even when it no longer pays locally” can change what deals are offered to you; similarly, an AI whose source code others can inspect may face situations where sticking to a pre-committed policy yields better outcomes than opportunistic deviations.

​
Applications

Although Newcomb’s Paradox is stylized, its core tension shows up in negotiations, AI design, and strategic planning wherever others successfully model your behavior.

Strategic Commitment and Reputation

In long-term relationships—business partnerships, alliances, or repeated games—your partners form expectations about whether you will “two-box” when you can get away with it. Adopting a policy of honoring commitments, even when short-term incentives flip, makes you look more like a one-boxer in Newcomb’s setup, encouraging others to offer you high-value deals in the first place.

AI Alignment and Predictable Agents

Advanced AI systems may be deployed in environments where other agents, or even the AI’s own training process, can accurately predict its policy. Newcomb-style reasoning helps clarify what it means for an AI to follow a decision theory that leads to globally good outcomes when its internal algorithm is inspectable, connecting to debates in Determinism and Free Will.

Contract Design and Incentive Architecture

When designing contracts, bonus schemes, or platform rules, you often rely on how people anticipate your responses to different behaviors. Newcomb’s Paradox reminds you that if participants can reliably predict you will enforce rules strictly—even when it looks locally costly—you may deter violations more effectively than if you always re-evaluate from scratch.

Personal Decision Habits

On a personal level, Newcomb’s problem encourages you to think about your “type” of decision-maker rather than isolated choices. Cultivating habits like honoring past commitments, avoiding convenient excuses, and resisting short-term temptations can make your future environment kinder, because people and institutions implicitly treat you as a reliable one-boxer instead of an opportunistic two-boxer.

​
Case Study

Consider a founder negotiating with a major investor over a term sheet that includes strict vesting and clawback provisions. The investor has studied the founder’s past behavior and spoken with previous partners; they have a reasonably accurate model of whether the founder will walk away from a deal later if circumstances change. Before the final negotiation meeting, the investor has already decided how generous the valuation and control terms will be, based on their prediction of the founder’s “type.” At the table, the founder faces a Newcomb-like choice. On the surface, they can either accept the current offer but signal a willingness to renegotiate aggressively later (two-boxing), or they can commit clearly to honoring the spirit of the deal even if future market conditions make it look locally suboptimal (one-boxing). The investor’s model has already shaped the term sheet now on the table: founders predicted to be reliable received more favorable terms earlier in the process. If the founder reasons purely causally—“the contract is already written; how I describe my future intentions cannot change how the investor behaved last month”—they may be tempted to preserve maximum future optionality. But over a portfolio of deals, investors update their predictive models: founders who act like two-boxers tend to see fewer generous offers and more protective clauses in subsequent negotiations. The case mirrors Newcomb’s Paradox: by committing to a policy that would look irrational in a single-shot causal snapshot, the founder can systematically end up in worlds where better offers appear in the first place.

​
Boundaries and Failure Modes

Newcomb’s Paradox is powerful but easy to misapply. Understanding its limits keeps it from turning into a slogan.
  1. Requires Highly Reliable Prediction: The puzzle assumes a predictor that is correct almost all the time. In real life, forecasters are fallible, and if the “predictor” of your behavior is very noisy, taking the extra $1,000 (two-boxing) may genuinely dominate. Treating every weak signal or stereotype as if it were an almost perfect predictor can lead to overfitting and self-defeating deference.
  2. Depends on How the Problem Is Framed: Small changes in how the predictor works or when you learn information can collapse the paradox. If you know the exact algorithm the predictor uses, or if the prediction is published in advance, the structure becomes more like a coordination game or a variant of the Prisoner’s Dilemma than the classic Newcomb setup.
  3. Misuse: Justifying Blind Fatalism: Some readers incorrectly generalize Newcomb-style reasoning into “my choices do not matter because everything important was already set by predictions or fate.” The original paradox keeps your choice decision-relevant: your policy still determines which kind of world you tend to inhabit, even if it does not rewrite the past. Ignoring this nuance can encourage passivity instead of thoughtful commitment.

​
Common Misconceptions

Because Newcomb’s Paradox is counterintuitive, it is often taken to prove more than it really does.
Reality: One-boxers are not necessarily greedier; they are following a different decision principle. Given a highly reliable predictor, one-boxers expect to walk away with much more money on average. The disagreement is about which rule—evidential, causal, or policy-based decision theory—best captures rational choice when others can accurately model you.
Reality: From a narrow causal snapshot, taking both boxes seems dominant because the money in the opaque box is already fixed. But when you zoom out to the full setup—where agents like you are predictable and the predictor has conditioned on your policy—systematically two-boxing can lead to predictably worse outcomes for people who share your decision rule. The paradox challenges the idea that local dominance is always the right criterion.
Reality: While stylized, Newcomb-like structures appear whenever behavior is forecast and priced in advance: in credit scoring, dynamic pricing, algorithmic moderation, and geopolitical deterrence. How you think about Newcomb’s problem shapes how you design systems that interact with predictive models, from loan algorithms to AI agents facing each other.
Newcomb’s Paradox connects to several other core ideas in decision theory and philosophy.

Expected Value

A central tool in decision analysis that multiplies outcomes by their probabilities. Newcomb’s problem forces you to ask which probabilities you should condition on—those before or after learning how the predictor works.

Prisoner's Dilemma

A game where mutual cooperation is globally better but mutual defection is the only Nash equilibrium. Like Newcomb’s Paradox, it exposes tensions between local and global rationality when others’ behavior depends on your policy.

Regret Minimization

A decision strategy that focuses on minimizing worst-case or expected regret rather than maximizing expected payoff. Different regret notions can support one-boxing or two-boxing, making Newcomb’s problem a useful test case.

Bayesian Thinking

A framework in which beliefs are updated by conditioning on new evidence. Newcomb’s Paradox probes how Bayesian conditioning should treat the fact that your own choice is strong evidence about what has already happened.

Free Will and Determinism

Philosophical debates about whether our choices are determined, and if so, how they can still be meaningful. Newcomb-style puzzles show that even in deterministic settings, the structure of prediction and dependence matters for rational action.

St. Petersburg Paradox

A classic paradox about infinite expected value and risk attitudes. Together with Newcomb’s Paradox, it shows how decision theory must grapple with both extreme payoffs and unusual dependency structures.

​
One-Line Takeaway

Newcomb’s Paradox teaches that when others can accurately predict your policy, the rational choice is not just about changing immediate outcomes, but about being the kind of decision-maker for whom good worlds are already prepared.