AI as Social Infrastructure

In the [YOU] on AI Field Guide

[YOU] on AI observes that among the most consequential things the orange pill reveals is that access to AI is not merely a productivity advantage but a structural one: the builder with AI has access to more non-redundant information, more creative synthesis, more capability to act on good ideas. Zuckerberg’s social infrastructure claim extends this observation to a broader population—not just builders and developers but anyone navigating health decisions, legal situations, financial choices, or professional transitions without the social network that would normally provide expert guidance.

The cycle’s concern with who benefits from AI finds its most hopeful expression in this framing and its most critical examination in the same framing. Hopeful, because the redistribution of informal expertise is a genuine democratizing force if it works as described. Critical, because the delivery mechanism is Meta’s platform infrastructure—the same infrastructure that runs on a social graph whose data practices have been the most contested in the history of commercial internet. Framing the AI assistant as social infrastructure does not resolve the question of who controls the infrastructure; it intensifies it.

Granovetter’s framework adds precision: the informal social capital Zuckerberg describes is, in network terms, the product of strong-tie relationships built through sustained, reciprocal engagement. The friend who is a doctor gives different advice than a diagnostic AI precisely because the friend understands the context of the patient’s life in ways that no query-response system can match—not because the AI lacks information, but because the relationship provides a dimension of knowledge that information alone cannot supply. AI as social infrastructure may raise the floor of accessible guidance while leaving the ceiling of genuine social support exactly where it was.

Origin

Zuckerberg articulated the social infrastructure argument most explicitly in his discussions of Meta AI’s design goals in 2023 and 2024: the vision of people having AI that functions like a close advisor—something between a knowledgeable friend and a professional consultant. The argument was not new to his thinking. It is a direct extension of the connectivity thesis he developed through Free Basics and Internet.org: the internet (or AI) as a precondition for participating in the opportunities available to those who are already connected.

The argument intersects with a broader debate in the welfare economics literature about the role of social capital in determining life outcomes. Robert Putnam’s work on bonding and bridging capital documents the extent to which access to trusted expert guidance through personal networks predicts health, educational, and economic outcomes independently of formal market access. Zuckerberg’s social infrastructure claim is, implicitly, a proposal to use AI to address the Putnam gap—to give those with weak informal networks some of the guidance advantages that strong networks have always provided.

Key Ideas

The informal expertise gap. The most consequential asymmetry in access to expertise is not formal—the legal system, the healthcare system, the financial system are all nominally accessible—but informal. The person who knows a doctor personally gets different health guidance than the person who books a fifteen-minute appointment with a stranger. The person whose network includes senior professionals in her field gets career guidance that no formal mentorship program fully replicates. AI as social infrastructure proposes to partially close this gap by providing a form of expert engagement that is available on demand, at low cost, to anyone with a connection.

Personalization as the critical variable. The social infrastructure claim depends on personalization: an AI assistant that knows your relationships, your history, your context can give contextually relevant advice in ways that a generic chatbot cannot. This is where the social graph becomes the delivery mechanism: Meta’s AI assistants, running through the social graph, have access to a relational context no other AI deployment possesses. The same data structures that make this contextual intelligence possible make them the most powerful behavioral profile ever assembled.

The relational quality question. Human advice comes embedded in relationships, mutual accountability, and shared history that AI systems cannot replicate. The friend who is a doctor cares about you in a way that shapes her advice: she understands that you are anxious about procedures, that you have a low pain threshold, that you are more likely to follow through on simple interventions than complex ones. Nussbaum’s capabilities framework would ask whether AI social infrastructure expands the capability for genuine human affiliation or produces a functioning that simulates connection while leaving the capability for authentic relationship undeveloped. The difference between an AI assistant that helps you navigate a health decision and a friend who does the same is not merely a difference in information quality; it is a difference in the relational goods that the interaction provides.