Towards Diffusion of Virtual Diplomacy

This document assumes familiarity with ideas from Introduction to Virtual Diplomacy, such as virtual embassies, capability consumption, and supported policies.

Previously, we described virtual embassies as monitoring systems which allow third-parties to measure AI capability consumption. These tools help address questions such as: "How much autonomous research has this organization consumed internally?" or "How much autonomous hacking has this organization sold to customers in this jurisdiction?" By resolving such queries, virtual embassies may provide the technical basis on which to develop effective AI policy.

However, as an approach for tackling emerging challenges, virtual diplomacy is as much a political puzzle as it is a technical one. Even assuming the technical maturity of virtual embassies as verifiable monitoring instruments, many questions remain. How should policies be shaped? How can the technology be diffused? How can its governance be future-proofed? In effect, the political maturity of the proposal — accounting for political realities and stakeholder interests — is as important as its technical maturity.

In a previous resource, we dissected virtual embassies into their core components and assessed their technical readiness individually. By taking stock of where we are and where we need to be for each component, we effectively laid out a roadmap for improving the technical maturity of the proposal. Similarly, this resource lays out a roadmap for increasing its political maturity, moving it closer to active use.

Self-Interested Coordination

Our strategy is built on three key premises:

• Credible commitments are torches, and self-interest is their fuel. Our first assumption is that consequentialist self-interest is the decisive force in shaping the behavior of an entity. Voluntary commitments to particular principles or practices are only credible, in the game-theoretic sense, insofar as their implied behavior is compatible with the entity's self-interest. If organizations issue commitments that appear to hinder their standing, these commitments may be dismissed as not credible. Credible commitments are powerful instruments precisely because of being fueled by self-interest. As a corollary, attempts to influence the behavior of an entity which are not mindful of its self-interest, such as merely imploring it to prioritize social welfare, are poised to prove ineffective on their own.

• Surroundings dictate optimal behavior, yet other players are shaping them. Even if entities are assumed to pursue self-interest, the behavior which affords them the largest gains is a function of their environment. Different courses of action may help further their agenda in some settings but prove counterproductive in others. However, an entity's environment consists in large part of others like it, such as multiple companies competing in a market niche. Therefore, an entity's optimal behavior under self-interest is directly shaped and informed by its peers. It is precisely entities themselves that effect the largest influence on each other's behavior, rather than some other feature of their environment.

• Hard law as sufficient, but not necessary. An entity's surroundings may include rule-making entities that are in a position to force it into rule-taking by means of legal deterrence, such as a company found in the jurisdiction of a state. However, hard law, as the totality of binding rules advanced by states and certain intergovernmental organizations, is not the only way in which an entity's self-interested behavior can be shaped. Indeed, a broad variety of soft law instruments has emerged to equip non-state actors with ways of shaping each other's behavior. Even if not backed by deterrence from subjects of domestic or international law, soft law can enable them to influence each other in many ways, from conditioning investment on the adoption of sustainability practices to non-assertion covenants for intellectual property.

Coordination Amid Anarchy

An alternative way of describing this headspace is in terms of international relations and international law:

• The world is anarchic. That is, in the technical sense of the word. There is no international force which is policing states, as international law since Westphalia recognizes sovereign equality as a core pillar. Despite that, international coordination has proven extensive: international trade is a sizeable portion of world trade, states routinely come together militarily in the face of aggression, and international scientific efforts range in focus from the smallest particles to the largest celestial objects. Granted, seemingly stable world orders have often proven fragile, yet such achievements act as a proof of existence for coordination under anarchy.

• International law as lesson in coordination amid anarchy. The significance of international law in AI governance should not be limited to hard law imposed on states. Perhaps even more significant than that is its body of knowledge centered on coordination in anarchic settings. This is not because AI companies are somehow above domestic law, but because they are largely operating in a pre-normative setting. Hard laws have simply not yet coalesced to shape their behavior in meaningful ways. Besides, the speed of the field, catalyzed by remarkable pressures and feedback loops, may continue to outpace the traditional model of global governance, which involves formal decision-making protocols spanning years.

• Soft laws as coordination practice. The recent proliferation of soft law and the growing variety of international organizations are steadily increasing the influence of non-state actors at the global level. If wielded effectively, these may enable international coordination across non-state actors even in a pre-normative setting, by aggregating the interests of key stakeholders through coherent policies backed by each other's levers. If designed thoughtfully, such soft law instruments may also facilitate subsequent hardening through state actor engagement. In the following section, we provide a vignette for how one such instrument may look like.

Aggregate, Demonstrate, Mobilize

Following the above headspace as a starting point, these are the three stages of our strategy for facilitating the diffusion of virtual diplomacy:

• Aggregate. This stage is concerned with interest aggregation, which refers to the creation of coherent policies which strive to accommodate the interests articulated by disparate stakeholders. Many actors are influencing or being influenced by AI development. Consider the impacts of labor market disruption on workers, the impacts of serving pretrained models on owners of intellectual property, or the impacts of dual-use capabilities on national security institutions. Effective interest aggregation refers to finding ways of reconciling these concerns as well as possible, despite many of them being conflicting, overlapping, or orthogonal. Consider the following vignette, which speculates on a multilateral soft law instrument for aggregating such interests:

• Demonstrate. This stage focuses on demonstrating previously designed instruments in limited, controlled settings. By deploying virtual embassies in genuine products and using these to execute proof-of-concepts in partnership with representative entities, we attempt to demonstrate technical readiness. For instance, a product involving autonomous hacking may help showcase a soft law instrument concerned with managing dual-use capabilities. These products can also help make the development of virtual diplomacy more sustainable.

• Mobilize. Building on the previous stages, mobilization involves engaging stakeholders to put the designed arrangements into practice. This ranges from direct engagement with specific entities to broader advocacy across entire categories of stakeholders, depending on the instrument. While the previous stages establish game-theoretic properties and technical feasibility, success in this stage is measured by actual uptake of instruments and the reach of virtual embassies in real-world applications.

Towards Diffusion

Taken together, these stages represent the core components of our strategy for achieving the political maturity of virtual diplomacy, independent of progress towards technical maturity of its constituent virtual embassies. These stages may overlap and branch out in organic ways, especially when concurrently exploring different instruments. The strategy focuses on channeling the aggregated interests of key stakeholders to effect change on incentive landscapes in targeted ways. Virtual diplomacy gets positioned as an instrumental goal in furthering many diverse interests, allowing stakeholders to join forces in a mutually beneficial way. In this, soft law instruments may facilitate coordination in pre-normative settings, and potentially even in tandem with hard law.