CONCEPT

The Dogs That Bark

De Soto's metaphor for local knowledge — the social memory that marks boundaries no survey has drawn — and the methodological principle that institutional infrastructure must be built from the ground up, not imposed from above.

When de Soto's researchers arrived in Lima's informal settlements to map property boundaries, they discovered the boundaries were already known. Not formally — there were no deeds, no registry entries, no cadastral maps. But informally, with precision that rivaled any county recorder's office, the community knew exactly who owned what. The knowledge was maintained in social memory, in the collective awareness of neighbors who had watched each house being built, who knew the handshake agreements and the verbal contracts that had, over decades, created a property system as detailed as anything formal institutions could produce. De Soto captured this phenomenon with a metaphor that has become one of his most cited: in the informal settlements, the dogs barked. When a stranger crossed an invisible boundary, the neighborhood dogs announced the intrusion. The community knew the territory. The formal system did not.

In the AI Story

Hedcut illustration for The Dogs That Bark — The Dogs That Bark

The metaphor was not incidental. It named the central insight that separated successful property rights reform from decades of failed attempts. Every reform that ignored local knowledge — that imposed registration systems designed in government ministries without consulting the communities whose property was being registered — had failed. The failures were not failures of will or funding. They were failures of representation. The formal system did not capture what the community already knew, and a formal system that does not represent reality creates a competing reality that the community neither trusts nor obeys.

De Soto's methodological response was to work from the informal system outward. This meant sending researchers not into government archives but into the settlements, interviewing families, tracing the history of each parcel, documenting the informal transactions that constituted the community's property system. The formal title, when it was eventually issued, represented not an imposition from above but a recognition from below — a formalization of what already existed.

The principle applies to the AI economy with a directness that the technology industry, with its preference for universal scalable solutions, is poorly equipped to appreciate. AI tools were built in a specific context — in the United States, by companies headquartered in San Francisco, by teams reflecting the assumptions of the American technology sector. The training data is predominantly English-language text reflecting the concerns and vocabularies of the English-speaking world.

The consequences are subtle but consequential. When a developer in Lagos describes a problem to Claude Code, the tool responds with technical competence but potential contextual incompleteness. Local payment systems, regulatory requirements, infrastructure constraints, cultural conventions — all may be underrepresented in the training data. The tool is not broken. It is decontextualized. It produces responses that are technically capable and contextually partial.

This decontextualization is the AI equivalent of a property registration system designed without consulting the communities it serves. The dogs-that-bark principle prescribes specific remedies: training data representing problems and contexts of the developing world, interfaces optimized for mobile and intermittent connectivity, community-driven customization that allows local developer communities to adapt tools for their specific contexts, and documentation that respects the cultural and linguistic diversity of the global builder population.

Origin

The metaphor emerged from de Soto's fieldwork in the early 1980s and was articulated systematically in The Mystery of Capital. It captured something that orthodox surveying and registration had been unable to see: that property boundaries existed as social facts before they existed as cartographic ones.

The methodological implication — that institutional infrastructure must be built from local knowledge outward — became one of the Institute for Liberty and Democracy's distinguishing commitments. Reform designs that ignored this principle failed; reform designs that incorporated it succeeded. The pattern has held across four decades and multiple continents.

Key Ideas

Local knowledge is real knowledge. The community's informal property system was not less accurate than the formal one; it was differently accurate, and the difference mattered.

Imposed systems fail. Institutional infrastructure designed without community input produces competing realities that undermine rather than extend the formal system.

Formalization builds on informality. Successful reform incorporates existing informal arrangements rather than replacing them with alien templates.

The AI parallel is direct. Tools built without local context produce outputs that are technically capable and contextually partial — functional but unfitted.

The remedy is participatory design. Training data, interfaces, and customization must be developed with the communities the tools are meant to serve, not designed for them from San Francisco.

Debates & Critiques

Whether sufficient local knowledge can be incorporated into globally-trained AI models remains an open question. Critics argue that training data biases are structural and cannot be eliminated through marginal corrections — that only locally-developed models on locally-controlled infrastructure can meet the dogs-that-bark standard. Defenders argue that fine-tuning, RLHF, and community-driven training can incrementally improve contextual fit without requiring complete reconstruction. The debate maps onto broader questions about whether the AI economy's representational gap is best closed through extending existing systems or constructing alternatives.

Appears in the Orange Pill Cycle

Hernando de Soto — On AI