An AI Minister Is Only as Honest as Its Ministry
Albania presented its AI "minister" as a cure for corrupt procurement. Reports that the agency behind it is under investigation should prompt every government to ask what makes a public algorithm trustworthy in the first place.
Albania’s AI “minister”, Diella, was billed as an anti-corruption procurement AI. The appeal of the idea is easy to see. Software, the reasoning goes, does not take bribes, does not favour friends and does not tire of scrutiny. Corruption, on this account, is a human weakness, and automation can engineer it out.
In August, the Organized Crime and Corruption Reporting Project (OCCRP) reported that prosecutors are investigating leaders of AKSHI, the state agency that built Diella, for rigging tenders. More than €40m in assets has been seized, according to the report.
It is important to be precise about what that means. An investigation is not a conviction. The individuals concerned have not been found guilty of anything, and they are entitled to the presumption of innocence. Nothing in the reporting establishes that Diella itself was manipulated. It may yet be shown that the system worked as designed.
But the case does not need a verdict to teach its lesson. It shows, in unusually stark form, that an algorithm cannot be more trustworthy than the institution that commissions, builds, configures and operates it.
The anti-corruption fallacy
Claims that AI can eliminate corruption usually rest on a hidden assumption: that the dishonesty lies in individual decisions, and that removing the human decision-maker removes the opportunity. In practice, procurement can be distorted long before any single award. It can happen in how requirements are written, which suppliers are eligible, how evaluation criteria are weighted, and who has access to the system’s settings and data.
An automated tool does not remove those choke points. It can relocate them, from the official who signs a contract to the team that defines the scoring model, from the minutes of an evaluation panel to the parameters of a system few people can inspect. If the agency operating that system is compromised, automation can lend a corrupt process the appearance of neutrality. That is worse than no automation at all, because it makes the problem harder to see.
What must come first
Any government tempted to market AI as an anti-corruption tool should meet three conditions before it makes the claim.
Institutional integrity. The body that builds and runs the system must itself be subject to the ordinary disciplines of good public administration: clear lines of accountability, separation between those who design rules and those who benefit from outcomes, and protection for staff who raise concerns. An AI system inherits the governance of its owner.
Procurement transparency. The procurement of the AI system should be as open as the procurement it is meant to clean up. Who built it, under what contract, at what cost, with what specifications and what access rights? If citizens cannot see how the tool was bought, they have little reason to trust how it buys.
Independent audit. The system’s logic, data, configuration and change history should be open to inspection by a body independent of the operating agency: a supreme audit institution, a specialised regulator or accredited external auditors. Audit must include who can change the system and how changes are recorded, not only how the model scores bids. Results should be published.
None of these conditions is novel. They are the standard tools of public-sector integrity. The point is that AI does not replace them. It makes them more necessary.
The wider temptation
Albania is not alone in presenting AI as a way to bypass institutional weakness. Governments across regions face pressure to show visible, modern responses to long-standing problems, and an AI system with a name and a face is a striking way to do it. The danger is that the symbol substitutes for the reform.
The more defensible path is slower and less photogenic. Strengthen the institution first. Make its procurement transparent. Submit its systems to independent audit. Only then is it reasonable to ask whether automation adds anything, and to measure whether it does.
Whatever the outcome of the investigation, the reporting on Diella has already performed a public service. It has reminded governments that trust in a public algorithm is borrowed from the institution behind it, and can be no greater than the trust that institution has earned.
Sources
Discussion
No comments yet. Start the conversation.
