The Altar of Power and the Machine: Will Artificial Intelligence Replace the State?

The fact that artificial intelligence is used in almost all spheres of human life is no surprise to anyone. But the case of Albania, where the first virtual minister appeared in September this year and was allowed to distribute public finances, raises many important questions. Among them is how actively AI should be involved in government management and how it should be controlled. Vadim Novikov, Advisor to the President of Almaty Management University (AlmaU), discusses this.

Until recently, artificial intelligence in the public sector was an auxiliary tool - writing drafts, checking documents, translating texts. But on September 11, 2025, Albania made a move that immediately hit the headlines: the country's Prime Minister Edi Rama introduced Diella, a virtual cabinet member with a focus on public procurement.

It was entrusted with a task that has always been considered the holy of holies of power. For the first time, the "voice" of a machine with ministerial intonations was heard where it is decided who will get the budget money and on what terms.

The central nerve of this story is not in the technology, but in the institution. When an algorithm decides on the winner of a tender, the fundamental foundations of democratic governance are called into question: the personal responsibility of officials and the right of citizens to appeal against their actions.

Diella was born as a digital assistant on the e-Albania public services portal. Since January 2025, she has helped citizens obtain certificates, issuing more than 36,000 documents with an electronic signature. In September, she was promoted: Rama introduced Diella as a "virtual member of the government."

However, a clear legal mechanism defining its legal status, the procedure for appealing its recommendations and the allocation of legal liability for possible errors has not been officially disclosed.

Diella's mandate is to analyze all tenders. The goal, according to Rama, is "to make Albania a country where public tenders are 100 percent free of corruption". The authority to select winners is gradually being taken away from line ministries and transferred to an AI system. The prime minister's ambitions stretch far: he has previously publicly admitted that Albania is "ready for an AI-driven ministry, and one day even an AI prime minister."

The Albanian Information Society Agency (AKSHI) claims that Diella is based on Microsoft's language models. The most important nuance: its decisions are advisory in nature.

"Diella won't decide for us - we make the decisions," explained Ma.

To mitigate risks, the Human-in-the-Loop scheme is said to be in place, which means that the final decision always rests with a human, not a machine: the AI's conclusions are verified by a special unit under the prime minister. However, the opposition and skeptics point out that behind the beautiful facade of technology and an avatar ("the face and voice" of the new AI minister is Albanian actress Anila Bisha) lies a mechanism of manual management of purchases by the prime minister personally.

The symbolic culmination was its parliamentary "debut" on September 18, when the avatar addressed the deputies from the screen, directly answering reproaches of illegitimacy: "The Constitution speaks about institutions that serve people. It does not speak of chromosomes, flesh or blood. It talks about duties, responsibilities, transparency and service without discrimination.

The key line of criticism strikes at the heart of democratic procedure: decisions on the distribution of state contracts should be made by elected officials who bear personal responsibility, and Diella has none, which means, in case of anything, there is no one to blame.

In practice, the model works as a black box. Bidders and the public do not know what criteria the system uses to rank bids and how to appeal its conclusions. This violates the principle of due process. If the algorithm makes a mistake, who will be penalized?

Putting all responsibility on the prime minister is a politically convenient move, but it does not create a working legal mechanism. The Albanian experiment is a vivid example of how technology is transforming from a tool into a symbolic institution, raising fears of the emergence of a "technocracy without participation," where complex decisions are delegated to a machine.

The arguments in defense of the "virtual minister" are based on three pillars: objectivity, transparency and efficiency. Proponents of the project claim that the algorithm, unlike a human, has no conflict of interest and acts strictly according to set rules. Albanian Prime Minister Rama promises that every procedure will become "100% readable" for control thanks to the digital trail.

The technical safeguards mentioned are explainable AI (XAI), an approach where a system can explain its decision logic.

In theory, creating public registries of algorithms could actually increase trust. For example, the authorities in Amsterdam and Helsinki have created publicly accessible websites that explain in simple terms each algorithm used in the city: what it is for, what data it uses, and who is responsible for it.

However, these counterarguments have their limits. Even the most "explainable" models are trained on historical data, which may contain hidden biases. As long as there are no clear rules for appealing AI decisions and control mechanisms are not independent, all talk about transparency remains declarations.

The Albanian case is just one harbinger of a deeper trend. Drawing on OECD analysis, we can identify the neighboring holy of holies of power where AI will arrive in the next decade.

The first area is regulatory intelligence. Algorithms are increasingly being used to analyze and create laws. Even before Diella, Albania was analyzing EU legislation with the help of OpenAI.

There are also experimental initiatives like the "office of regulatory intelligence" in the UAE, which uses AI to track the effectiveness of laws and suggest amendments, promising to speed up the legislative cycle by 70%.

The concept of "regulatory intelligence" has also entered the political lexicon of Kazakhstan - President Kasym-Jomart Tokayev announced plans for its development in his annual Address.

By 2030, the technologies required for this will be mature and ready for use. And the leading countries in digitalization will start implementing such systems. Their implementation will be supported by the development of vibe modeling - similar to vibe coding, where the programmer sets the overall goal rather than writing code line by line. Here, the analyst also sets a high-level task for the AI (e.g., "estimate the effect of reform"), and the system itself selects models and processes data, speeding up hypothesis testing.

The second is city management. "Digital twins" of cities, as in Singapore, are already helping to manage megacities. These are detailed virtual 3D models that receive real-time data from sensors on roads or in utility networks. Using these models, AI can calculate different scenarios, such as how traffic will change when a street is closed, helping to make informed decisions.

By 2030, based on the trends noted by the OECD, AI is likely to offer municipalities optimal budget decisions. The main risk here is increased inequality if algorithms trained on biased data systematically bias some areas in favor of others.

The third is justice. AI is already being used as a judge's assistant. For example, in the Supreme Court of Brazil, the Victor system helps analyze and classify incoming cases. By 2030, decision support systems will become the norm, but their implementation will be possible only if strict conditions are met: open logic, mandatory human control, and clear appeal procedures. Society will not be ready to entrust a machine with the final verdict, especially in criminal law.

Fourth is healthcare. In medicine, AI helps in diagnosis and hospital resource management. For public healthcare systems, the key challenge will be to use AI to allocate funding and predict epidemics. Explainability of decisions here will become critical: both doctors and patients must understand the logic of the algorithm.

Fifth, legislative work. AI is being used to automate routine tasks, such as transcribing and translating debates in real time, as is done in UK and EU parliaments. But more importantly, AI is becoming an analytical tool for MPs themselves. In the UK, Consult's AI tool helps analyze thousands of citizen submissions during public consultations, highlighting key opinions. Thus, AI does not write laws for people, but it helps make the legislative process more efficient.

The Albanian experiment let the machine into the holy of holies of power - the sphere of distribution of public money. So far, Diella remains more of a PR tool, and the final decisions are made by people. But the very fact of its "appointment" blurs the boundary between a tool and an institution.

If the altar of power is opened to the machine, it must be "sanctified" by democratic rituals: public accountability, transparency of algorithms, and the right to human participation.

The lesson here is simple and resembles working with an ordinary spreadsheet. It is not the program as such that gives the correct calculation, but the quality of the data in the cells, the correctness of the formulas and the integrity of the person who manages the process.

So it is with artificial intelligence: its emergence guarantees neither efficiency nor fairness. Three things determine the outcome: the quality of the data, the transparency of the instructions, and the reliability of the institutional safeguards that people build around the machine.

This article was AI-translated and verified by a human editor