AI safety coordination requires global cooperation—but cooperation raises questions of fairness. Who should bear the costs of AI safety? Should wealthy nations with leading AI companies contribute more than others? Do nations that take risks owe compensation to those exposed? The question of international distributive justice—how benefits and burdens should be distributed globally—directly affects whether AI safety governance can achieve broad participation or will be undermined by perceived unfairness.

The Distributive Question in AI Safety

AI safety governance creates winners and losers:

  • Cost bearers: nations that restrict AI development, invest in safety research, accept slower growth
  • Beneficiaries: all nations protected from catastrophic AI risk
  • Concentrated costs: AI-leading nations face largest economic constraints
  • Diffuse benefits: risk reduction helps everyone including non-participants

This creates distributive questions:

  • Should costs be distributed equally, proportional to benefit, or proportional to contribution to risk?
  • Do wealthy nations owe assistance to poorer nations for AI safety capacity?
  • Should AI-leading nations compensate others for the risks they create?
  • What do non-participants owe to the cooperative scheme?

Rawls and International Justice

John Rawls' theory of justice revolutionized political philosophy—but it applies differently to domestic and international contexts.

Domestic Justice: The Difference Principle

Within a state, Rawls argues that inequalities are justified only if they benefit the worst-off. The "difference principle" requires that any inequality in primary goods (rights, liberties, opportunities, income, wealth) must be arranged to maximize the position of the least advantaged.

This creates strong distributive requirements: mere efficiency or aggregate welfare doesn't justify inequality. What matters is whether those at the bottom benefit.

International Justice: The Law of Peoples

Yet Rawls explicitly limits these principles to the domestic sphere. Internationally, he proposes:

  • Peoples are free and independent, their freedom to be respected
  • Peoples are equal parties to international agreements
  • Non-intervention: states shouldn't interfere in others' domestic affairs
  • Human rights: minimum protections for all persons
  • Duty of assistance: help "burdened societies" become well-ordered

Notably absent: any international analogue of the difference principle. International inequality is not inherently problematic. Rich nations may remain rich; poor nations may remain poor—as long as each is "well-ordered" with functional institutions respecting basic rights.

Why the Difference?

Rawls argues that distributive justice applies only within certain institutional contexts:

  • The basic structure: principles of justice govern the major institutions of a society
  • Coercive authority: the state coercively imposes rules in citizens' names, requiring justification
  • Cooperative scheme: citizens participate in shared institutions creating mutual obligations

Internationally, Rawls sees no comparable basic structure. The international system lacks:

  • A unified coercive legal system (anarchy, not hierarchy)
  • Shared institutions determining life prospects
  • Political community justifying egalitarian demands

Therefore, distributive equality is not required internationally—only that each nation be minimally well-ordered.

The Cosmopolitan Critique

Many philosophers reject Rawls' distinction. Thomas Pogge and Charles Beitz argue:

1. Nationality Is Morally Arbitrary

"Nationality is just one further deep contingency (like genetic endowment, race, gender, and social class), one more potential basis of institutional inequalities that are inescapable and present from birth."

If being born rich doesn't justify inequality within a nation, why does being born in a rich nation justify inequality globally? The same logic that rejects domestic caste should reject international caste.

2. Global Institutions Exist

International trade, financial systems, and regulatory regimes constitute a global basic structure. These institutions:

  • Allocate advantages and disadvantages
  • Structure interactions among peoples
  • Determine distributive outcomes

Where there is a basic structure, there should be principles of justice governing it—including distributive principles.

3. The Global Difference Principle

If Rawls is right domestically, cosmopolitans argue he should accept international egalitarianism. Global inequalities should be justified by their benefit to the globally worst-off—not just the nationally worst-off within each state.

Application to AI safety: If cosmopolitanism is correct, AI safety costs should be distributed to benefit the globally least advantaged. This might mean wealthy nations bearing most costs while benefits flow to all.

The Institutional Response

Right institutionalists defend Rawls' distinction. Michael Blake, Samuel Freeman, and Thomas Nagel argue:

1. Coercion Requires Justification

Domestic law coercively structures citizens' lives. This coercion demands justification through egalitarian principles. But the international system doesn't coercively govern individuals in the same way—states remain the primary legal authorities.

2. Political Community Matters

Co-citizens share political institutions that determine their life prospects. They participate in creating the rules that bind them. This creates special obligations absent between strangers in different states.

3. Institutional Capacity

Even if global equality were desirable, there are no institutions capable of achieving it. The coercive, hierarchical legal system that makes domestic redistribution possible doesn't exist internationally.

Application to AI safety: If institutionalism is correct, AI safety costs should be distributed according to principles states accept—not according to an ideal global difference principle. National consent matters; forced redistribution doesn't.

The Nationalist Alternative

David Miller and others argue that nationality itself has moral significance:

1. Special Obligations

We have special obligations to fellow citizens that we don't have to strangers globally. These arise from:

  • Shared history and culture
  • Reciprocal cooperation
  • Common political project

2. Community Preservation

Nations are valuable communities that deserve preservation. Demanding global equality might undermine the national communities people value.

3. Responsibility for Choices

Nations make choices about savings, education, and economic policy that affect their wealth. Requiring rich nations to compensate poor nations for the consequences of their own choices is unfair.

Application to AI safety: If nationalism is correct, nations can prioritize their own interests in AI safety coordination. There's no obligation to accept costs that primarily benefit other nations—though humanitarian duties may require minimum assistance.

Implications for AI Safety Governance

1. The Distribution Question Cannot Be Avoided

Any AI safety governance scheme distributes costs and benefits. Even deciding not to decide—leaving distribution to market forces or power politics—is a distributive choice. Governance design must explicitly address distribution.

2. Multiple Principles May Apply

Depending on which theory of international justice is correct:

  • Cosmopolitan: AI safety costs distributed to benefit globally worst-off
  • Institutionalist: costs distributed according to what states accept through fair negotiation
  • Nationalist: each state can prioritize its own interests, with minimum humanitarian duties

Since there's no consensus on the correct theory, practical governance might need to accommodate all three through:

  • A baseline of humanitarian assistance (acceptable to all)
  • Voluntary agreements for additional redistribution (cosmopolitan)
  • National discretion in participation level (nationalist)

3. Fairness Affects Participation

Perceived unfairness in cost distribution will undermine participation:

  • If wealthy nations feel they bear disproportionate costs, they may withdraw
  • If poor nations feel excluded from benefits, they may defect
  • If AI-leading nations feel penalized for their success, they may resist

Sustainable governance requires distribution perceived as fair by participants—even if theories of justice disagree on what's fair.

4. Different Principles for Different Questions

International distributive justice may apply differently to different AI safety questions:

Capacity building: Do wealthy nations owe assistance to poorer nations for AI safety expertise?
→ Likely yes, under duty of assistance/humanitarian obligations

Risk compensation: Should AI-leading nations compensate others for risks they create?
→ Plausible under cosmopolitan or tort-like principles

Regulatory costs: Should costs of compliance be distributed equally or progressively?
→ Disputed; cosmopolitans favor progressive distribution

Technology access: Should safe AI technology be distributed globally?
→ Cosmopolitans say yes; nationalists accept national control

5. The Free-Rider Problem

AI safety is a global public good—non-excludable benefits create free-rider incentives. Distributive justice questions interact with this:

  • Should free-riders benefit without contributing?
  • What constitutes "contribution" to AI safety?
  • Can non-contributors be excluded from benefits?

The principle of fairness suggests those who benefit from cooperative schemes should contribute. But AI safety benefits are non-excludable—everyone benefits regardless of contribution.

6. Historical Responsibility

Some nations have contributed more to AI risk through aggressive development. Does this create special obligations?

  • Polluter pays: those who create risks bear remediation costs
  • Beneficiary pays: those who benefited from risky development share costs
  • Ability to pay: wealthy nations contribute more regardless of responsibility

AI safety governance might combine these: historical contributors pay more, beneficiaries contribute, and capacity affects obligations.

Practical Distribution Principles

Given theoretical disagreement, practical governance might adopt plural principles:

1. Minimum Duties (Universally Accepted)

  • Assistance to build minimum AI safety capacity globally
  • Information sharing about AI risks
  • Non-interference with others' safety efforts

2. Proportionality (Broadly Acceptable)

  • Contributions proportional to AI development level
  • Costs proportional to risk contribution
  • Capacity building proportional to resources

3. Negotiated Distribution (Politically Feasible)

  • States negotiate contribution levels
  • Conditional benefits for participation
  • Different tiers of participation and contribution

4. Aspirational Equality (Cosmopolitan Goal)

  • Progressive movement toward more equal distribution of AI safety capacity
  • Eventual goal: globally equitable AI safety infrastructure
  • Voluntary mechanisms for redistribution beyond minimum

Conclusion

AI safety governance cannot avoid questions of distributive justice. How costs and benefits are distributed globally will affect whether governance can achieve participation, legitimacy, and effectiveness.

Theoretical disagreement about international justice—between cosmopolitan, institutionalist, and nationalist approaches—means no single principle will command universal assent. Practical governance must:

  • Acknowledge distribution explicitly: don't pretend it's neutral
  • Build on minimum consensus: humanitarian duties accepted by all theories
  • Accommodate disagreement: allow different levels of contribution and participation
  • Create fairness perceptions: distribution must seem fair enough to sustain cooperation
  • Address free-riding: balance inclusion with incentive for contribution

The question "who owes what to whom in AI safety?" has no consensus answer. But asking it—and designing governance that acknowledges different answers—may be essential for AI safety coordination that works in a world of divergent moral views.

References

  • Beitz, Charles (1979). Political Theory and International Relations. Princeton University Press.
  • Blake, Michael (2001). "Distributive Justice, State Coercion, and Autonomy." Philosophy & Public Affairs 30(3).
  • Miller, David (1995). On Nationality. Oxford University Press.
  • Pogge, Thomas (1989). Realizing Rawls. Cornell University Press.
  • Rawls, John (1999). The Law of Peoples. Harvard University Press.