The office has a wooden desk, the clerk is courteous and plain. The refusal came from somewhere else, dissolving down the vendor chain.
Consider a constructed scene from community grievance practice, typical of infrastructure projects financed under international performance standards.
A smallholder farmer arrives at a project liaison office. Her application for resettlement compensation was rejected, resetting her timeline and blocking access to alternative land before the planting season. The community liaison officer looks up the file. The screen displays an automated eligibility score below the qualification threshold. The officer cannot explain how the score was calculated. The software vendor who built the scoring pipeline cites proprietary risk models. The engineering contractor points to the project company's specifications. The project sponsor refers back to the procurement contract.
Every office was polite. Every institution pointed to a valid contractual boundary. Yet the complainant left without an explanation, because the decision dissolved across a delegation chain that terminates in no physical room.
The assumption of the reachable party
International grievance mechanisms, such as those aligned with the International Finance Corporation's Performance Standards, are built on a simple institutional premise: an affected party must have access to an identifiable decision-maker.
The standard grievance architecture specifies time-bound acknowledgments and clear escalation pathways. It presumes that somewhere in the institutional hierarchy sits an individual or committee with the authority to review the facts and overturn an erroneous finding.
Autonomous screening systems shatter this architecture by removing human discretion at the operational boundary. When an algorithm classifies a claim or filters an intake queue, the human staff at the front desk function as mere messengers. They inherit the status of The Liability Sponge, absorbing community frustration while possessing zero administrative power to alter the automated determination.
As examined in The Accountability Gap, diffuse responsibility distributes fault so finely that nobody owns the outcome. In human-facing administrative systems, this diffusion operates as an active barrier to justice: closing a grievance requires assigning responsibility, and automation renders assignment technically contestable.
What enterprise metrics miss
Enterprise adoption statistics celebrate efficiency and throughput across millions of transactions. These metrics capture the operator's operational convenience; they measure nothing about an affected person's capacity to obtain redress.
At the community level, the relevant metric is whether a person can secure an intelligible explanation in their own language and on a workable timeline from someone empowered to fix an error. When an enterprise replaces human intake with automated triage, it transfers the cost of explanation directly onto the most vulnerable participant.
This extends the core dynamic of Who Gets an Advocate. Well-resourced corporate clients hire specialized consultants to navigate complex algorithmic barriers. Displaced community members face an impenetrable wall of automated rejection notices, without advocacy or recourse.
The channel that has to stay open
Grievance mechanisms cannot function alongside autonomous systems unless accountability is designated before deployment.
An operational remedy requires four things that can be written into a contract. Every automated workflow touching land access or compensation designates a specific named person carrying accountability for its errors, recorded before the first determination is issued. Frontline liaison staff hold documented authority to suspend an automated determination pending manual review, which is the difference between a messenger and an officer. A meaningful non-automated channel for participation and grievance stays open, and using it must never cost the complainant access, consideration, or remedy. And when an automated determination is found defective after the fact, the operator preserves the evidence, notifies the affected party, corrects or withdraws the finding, and records what changed; a quiet fix is its own finding.
The Interim Protocol carries the third and fourth of those as conditions of use, and its never-machine-settled rule names eligibility, compensation, and consent among the judgments a machine may inform and may never conclude. That rule was written for consultants producing deliverables. The scene at the liaison office is what it looks like when nobody applied it upstream: a determination already settled in software, arriving at a desk where the only human present has no standing to reopen it.
Companions
- The original formulation of the buffer role: The Liability Sponge.
- The structural diffusion of blame: The Accountability Gap.
- Advocacy disparities under automation: Who Gets an Advocate.
- The institutional boundary at the counter: The Door at the Desk.
- The operational remedy: the Interim Protocol, whose conditions of use keep a non-automated grievance channel open and require that using it costs the complainant nothing.
These notes come out of Sociable Systems, a practice that reads AI-shaped documents the way a hostile reviewer will, before a lender or a court finds the gap. The argument has an operational form: the Interim Protocol sets out four rules for AI use in environmental and social deliverables, covering disclosure at touch-point grain, evidence custody, the phrases no automated screening may settle, and a hostile read before anything ships. Free, and written to be cited or retired once institutional guidance arrives.