A rule specifies an actionable constraint under stated conditions. A principle supplies a reason or standard for making, interpreting, evaluating, and revising such constraints.

This is a functional distinction rather than a law of vocabulary. Legal systems, companies, technical standards, and ordinary speech often use rule, principle, policy, standard, and norm differently. The useful question is therefore not merely what a document calls something, but how that statement guides judgment and action.

Rules reduce the available actions

A practical rule often has a conditional structure:

when condition C applies,
agent S must, may, or must not perform action A

For example:

If the user has not authorized disclosure,
the system must not publish the user's private information.

A well-specified rule identifies:

• the agent or role to which it applies;
• its triggering conditions;
• the required, permitted, or prohibited conduct;
• its jurisdiction or operational scope;
• any recognized exceptions;
• the evidence used to determine compliance;
• the authority that enforces or changes it.

Rules make coordination possible because they narrow the space of acceptable conduct. They reduce the need to reopen a foundational argument every time a recurring situation appears.

Regulative, constitutive, and procedural rules

Not every rule performs the same function.

The distinction between regulative and constitutive rules is central to social ontology. A regulative rule governs an antecedently intelligible activity, while a constitutive rule participates in defining what an institutional act is. A familiar form is:

X counts as Y in context C

Games, contracts, elections, offices, and money all depend partly on rules that make certain moves, documents, or objects count as socially significant acts or statuses.Stanford Encyclopedia of Philosophy: Social Institutions

Principles operate as reasons

A principle is a comparatively general normative consideration. It tells decision-makers what should matter across a range of cases.

Examples include:

• equal cases should be treated alike;
• exercises of power should be accountable;
• restrictions should be necessary and proportionate;
• claims should be answerable to evidence;
• affected people should retain meaningful agency;
• access should be limited to what a task requires.

A principle can do several kinds of work:

justify a family of rules
guide interpretation when a rule is unclear
identify a defect in an existing rule
direct judgment in a case not yet covered
provide a reason in conflicts among norms

A principle is therefore not simply a rule written at a higher level of abstraction. Its distinctive role is justificatory and evaluative.

Values, principles, policies, standards, and procedures

These terms are often arranged as if every organization had one fixed hierarchy. Real systems are less tidy. A functional map is still useful:

The same normative idea can move through several layers:

value: privacy is worth protecting
principle: processing should preserve user control
rule: disclosure requires authorization
procedure: verify authorization before publication
control: reject publication when the authorization record is absent

The label on a document does not settle its function. A “policy” may contain rules, principles, procedures, and standards in the same text.

A useful distinction, not an absolute dichotomy

Rules and principles can be compared along several dimensions:

This contrast became especially influential in jurisprudence through Ronald Dworkin’s criticism of rule-centered accounts of law. On his view, adjudication also invokes principles of justice, fairness, and due process that function as reasons within legal interpretation. The broader Hart–Dworkin debate concerns whether and how such moral considerations form part of law.Stanford Encyclopedia of Philosophy: Legal Positivism

The distinction should not be exaggerated. Rules can be vague, defeasible, or sensitive to proportionality. Principles can be institutionalized as strong constraints. “Rule” and “principle” do not name two natural kinds whose members always behave identically.

Exceptions require governance

Rules cannot anticipate every circumstance. A new case may fall outside the original design, two rules may conflict, or literal compliance may defeat the reason for which the rule exists.

This does not grant every decision-maker a general license to invoke a principle and ignore a rule. A defensible exception should identify:

1. why the case falls outside ordinary application;
2. which reason or principle is at stake;
3. why the existing rule produces an unacceptable result;
4. how the departure is limited to what is necessary;
5. who has authority to approve it;
6. how it will be recorded and reviewed;
7. whether the rule itself should be revised.

Principles can justify exceptions, but accountable institutions turn those exceptions into an explicit process.

Principles can conflict

Transparency, privacy, safety, autonomy, fairness, and efficiency may all be defensible principles. They do not always point in the same direction.

Full transparency can expose private information. Respect for individual choice can conflict with duties to protect people who cannot understand a serious risk. Uniform treatment can perpetuate disadvantages produced by unequal conditions.

Moral philosophy often uses prima facie or pro tanto duties to describe considerations that have genuine normative force but may be outweighed in a particular case. W. D. Ross’s account is influential precisely because it does not treat every principle as absolute or assume that a fixed ranking mechanically resolves every conflict.Stanford Encyclopedia of Philosophy: Moral Reasoning

Reasoning through a conflict therefore requires more than naming two principles. It requires an account of:

• the people and interests affected;
• the kind and magnitude of possible harm;
• necessity and proportionality;
• available alternatives;
• decision authority;
• residual duties such as explanation, compensation, or review.

Compliance, legitimacy, and effectiveness are separate

A rule can be followed perfectly and still produce a bad result.

The rule may rest on false assumptions, omit an affected group, reward a proxy rather than the real objective, or persist after its environment changes. Several locally reasonable rules can also interact to create a system-level failure.

Imagine a support center requiring every call to end within three minutes. The rule is measurable and consistently enforceable. Agents may meet it by ending difficult calls before the customer’s problem is solved.

Four judgments must remain separate:

compliance with the rule
≠ legitimacy of the rule
≠ effectiveness of the intervention
≠ desirability of the outcome

Compliance evidence shows whether a requirement was followed. It does not by itself establish that the rule was justified, that it served its purpose, or that its effects were acceptable.

Principles without rules, rules without principles

Principles without operational rules invite inconsistent interpretation. Responsibility becomes difficult to audit, recurring questions are repeatedly reopened, and public commitments remain disconnected from actual behavior.

Rules without principles create a different failure. People optimize for literal compliance, new cases have no intelligible direction, and an expanding rulebook substitutes for judgment. “The procedure was followed” becomes an answer to questions about fairness, truth, or consequences.

A workable normative system joins several components:

principles provide reasons and direction
rules establish actionable boundaries
procedures organize execution
controls prevent and record departures
feedback reveals effects
revision changes rules when evidence warrants it

From principle to enforceable system

The security principle of least privilege provides a concrete example. NIST defines it as restricting users, or processes acting for them, to the minimum authorizations and resources needed to perform assigned functions.NIST CSRC: Least Privilege

That principle does not implement itself. It can generate rules and controls such as:

• administrative access is denied by default;
• elevated access expires after a stated interval;
• ordinary work uses non-privileged accounts;
• privileged actions are logged;
• roles receive only the transactions required for their duties.

The principle explains the direction and the threat model. Rules specify the constraints. Access-control mechanisms enforce them. Audit logs provide evidence. Incident reviews test whether the design was effective.

Rules and principles in AI systems

A language model produces plausible continuations from probability distributions. Plausibility does not guarantee truth, logical validity, authorization, or acceptable consequences.

An AI system therefore needs a path from normative reason to technical enforcement:

principle: system action must remain within user authority
rule: no external write without explicit authorization
control: reject tool calls lacking an authorization state
verification: inspect the target identity and observed result
record: preserve an auditable action log

Other executable rules may include:

identity unresolved → stop
evidence unavailable → mark unknown
validation failed → do not claim success
irreversible effect → require stronger review

Principles remain necessary because no finite rule set anticipates every context. They can guide how a novel case is escalated, how uncertainty is represented, and which harms should be minimized. But a model that writes a persuasive explanation of a principle has not thereby complied with it. Reliable constraint requires permissions, schemas, tool boundaries, validation, logs, and human accountability.

Designing a rule-and-principle system

Identify what is at stake

Name the affected people, protected values, intended outcomes, possible harms, and responsible authority.

State a small set of principles

A principle should explain a family of judgments. Renaming every instruction as a principle only hides the operational work.

Derive rules with complete conditions

For each rule, specify:

subject or role
+ trigger
+ requirement, permission, or prohibition
+ scope
+ exception
+ evidence
+ authority

Implement procedures and controls

Important rules should appear in checklists, access controls, validation code, review workflows, and audit records rather than relying on memory alone.

Evaluate different layers separately

Ask:

Was the rule followed?
Did the rule adequately serve its principle?
Did the intervention achieve its objective?
What side effects appeared?

Create a revision path

Repeated exceptions usually signal an incomplete rule, a changed environment, or an unresolved conflict of principles. Revision should be governed rather than improvised.

Conclusion

Rules and principles belong to the same normative architecture but perform different jobs.

Rules state actionable requirements, permissions, and prohibitions under defined conditions. Principles provide the reasons and standards used to create, interpret, evaluate, and revise those rules.

Their relationship can be summarized as:

principles guide judgment
→ rules constrain action
→ procedures and controls implement rules
→ outcomes test effectiveness
→ feedback supports interpretation and revision

Rules cannot replace principles because no rulebook covers every future case or justifies its own authority. Principles cannot replace rules because general reasons do not produce consistent, auditable conduct by themselves.

A responsible system must explain both why a consideration matters and who must do what under which conditions. It must then examine actual consequences rather than treating either noble language or formal compliance as the end of judgment.

References

• Stanford Encyclopedia of Philosophy: Social Institutions
• Stanford Encyclopedia of Philosophy: Legal Positivism
• Stanford Encyclopedia of Philosophy: Moral Reasoning
• Stanford Encyclopedia of Philosophy: Deontological Ethics
• NIST CSRC: Least Privilege

That is why pricing, bargaining, voting, traffic, military deterrence, shared resources, and workplace coordination can all be studied as games. They need not be playful, and they need not have a single winner. Their common structure is strategic interdependence.

In a strategic interaction, I cannot evaluate my action without considering how others may act and respond.

From play to a formal model

The English word game has long referred to play, amusement, contest, and structured activity. Game theory abstracts from these familiar cases. It keeps the elements that make strategic reasoning possible—participants, permitted moves, information, consequences, and rules—while discarding the requirement that the situation be entertainment. Merriam-Webster: game

The abstraction is useful because the same strategic pattern can appear in very different settings. Two firms choosing prices, two governments considering an agreement, and two drivers approaching a narrow bridge may face structurally similar problems even though the subject matter differs.

The “game” is therefore not the social event itself in all its richness. It is a representation designed to isolate interdependent choice.

A decision problem becomes a game when another agent responds

Choosing whether to carry an umbrella is mainly a decision under uncertainty. The weather affects the outcome, but it does not observe your choice and revise its plan.

Negotiating a salary is different. The employer can infer how urgently you need the offer, conceal its limit, respond to your counteroffer, and change tactics when you change yours.

This gives a practical boundary:

Decision theory: choose under uncertainty about the world.
Game theory: choose under uncertainty about responsive agents.

Real problems can involve both. A company deciding whether to enter a market faces uncertain demand and strategic responses from competitors. Game theory does not remove ordinary uncertainty; it adds other decision-makers to it.

The anatomy of a game

A formal game specifies enough structure to compare strategies.

An action is a move. A strategy is a plan specifying what to do in every relevant situation. “Lower the price today” is an action. “Match any rival’s price cut, otherwise hold” is a strategy.

A payoff need not be money. It represents how a player ranks outcomes within the model. Safety, fairness, reputation, loyalty, risk, and the welfare of other people can all affect that ranking. Modeling an agent as payoff-maximizing does not by itself imply selfishness; it says that the model represents the agent as choosing according to an ordering of outcomes.

Best responses and Nash equilibrium

A strategy is a best response when it gives a player the most preferred available result given what the other players are doing.

A Nash equilibrium is a profile of strategies in which every player’s strategy is a best response to the others. No player can improve their modeled payoff by changing strategy alone while everyone else stays fixed.

This is a stability condition:

Given what everyone else is doing,
does any one player benefit by deviating alone?

If the answer is no for every player, the profile is a Nash equilibrium.

Nash showed that every finite game has at least one equilibrium when mixed strategies—probability distributions over pure strategies—are allowed. John F. Nash Jr., “Equilibrium Points in N-Person Games”

Existence does not tell us which equilibrium people will reach, whether they know the game, or whether the equilibrium is desirable.

The prisoner’s dilemma separates stability from collective value

Suppose two players can cooperate or defect. Higher numbers are preferred:

For each player, defection produces a higher payoff regardless of what the other does. Both therefore defect and receive 1, even though mutual cooperation would give each 3.

The result is not a claim that people are naturally selfish. It follows from a particular arrangement of choices and payoffs. Change the possibility of repeated interaction, reputation, monitoring, communication, or enforcement and the strategic structure changes.

The model reveals a general possibility:

Choices that are individually defensible within a set of rules can combine into an outcome every participant ranks below another feasible outcome.

That shifts attention from blaming individuals to examining institutions.

Information, commitment, and credibility change the game

Strategy depends on what players know and what they can credibly promise.

A threat matters only if carrying it out would remain rational when the time comes. A promise matters only if the other side expects compliance. Cheap statements without costs or enforcement may fail to change beliefs.

Several institutional mechanisms can alter behavior:

• repeated interaction makes future retaliation or reward relevant;
• reputation carries information across encounters;
• contracts change the consequences of breaking a promise;
• transparency changes what can be observed;
• voting rules change which coalitions matter;
• property rights change the payoff structure around a shared resource.

The point is not that every institution improves outcomes. It is that rules are part of the game. What appears to be a fixed conflict between personalities may be produced or intensified by incentives, information, and enforceability.

Rationality inside the model is conditional

Game theory commonly uses instrumental rationality: choose an effective means to an end, given beliefs about others and the available strategies.

This form of rationality does not determine which ends are worth pursuing. It also does not guarantee accurate beliefs, complete information, unlimited computation, or freedom from bias.

A strategy can be instrumentally rational relative to a cruel objective. A cooperative strategy can also be rational if the agent values fairness, relationship, or long-term trust. The mathematics operates on the preferences represented in the model; it does not supply those preferences from outside.

Behavioral and experimental work further shows that real people use heuristics, punish unfairness, reciprocate, misperceive probabilities, and differ in how they understand the situation. These findings do not make formal game theory useless. They clarify when its assumptions approximate behavior and when the model needs enrichment.

Equilibrium is not efficiency, consent, or justice

An equilibrium can be wasteful, coercive, unequal, or sustained by fear. Stability against unilateral deviation establishes none of the following:

• that total welfare is maximized;
• that the distribution is fair;
• that the rules are legitimate;
• that participation is voluntary;
• that no better institution exists.

Ethics and political philosophy therefore use game theory in at least two ways. It can explain how norms and institutions emerge or remain stable. It can also contribute to arguments about which institutions people could justify. But an explanatory equilibrium does not become a moral justification merely because it is mathematically stable. Stanford Encyclopedia of Philosophy: Game Theory and Ethics

Models clarify by leaving things out

A useful game makes assumptions explicit. That is also its limitation.

The result may change when the model changes:

• players may misunderstand one another;
• preferences may be unstable or socially formed;
• identities and relationships may matter;
• power may determine which actions are available;
• rules may be disputed rather than fixed;
• the relevant game may be embedded in a larger game;
• people may change the institution instead of optimizing within it.

Game theory is strongest when it identifies the consequences of a specified strategic structure. It is weaker when a simplified payoff table is treated as a complete description of human motivation or political legitimacy.

A practical way to analyze strategic interaction

When a problem feels like a “game,” ask:

1. Who can respond to whom?
2. Which actions and strategies are actually available?
3. What does each participant know or believe?
4. Which outcomes does each participant prefer, and how do we know?
5. Which promises or threats are credible?
6. Is the interaction repeated?
7. What equilibrium or pattern is stable under the current rules?
8. Is that stable result efficient, fair, or legitimate?
9. Which rule change would produce a different game?

The final question is often the most useful. Game theory does more than advise a player inside a fixed contest. It can show why changing the rules may matter more than choosing a cleverer move.

Sources

• Stanford Encyclopedia of Philosophy: Game Theory
• Stanford Encyclopedia of Philosophy: Game Theory and Ethics
• John F. Nash Jr., “Equilibrium Points in N-Person Games”
• Merriam-Webster: game