The topic of digitalization and modernization of public administration has been at the forefront of public debate in recent years. Despite the expressed enthusiasm and numerous legislative changes, the actual results often remain unsatisfactory. The main reason for this is not a lack of ideas, but rather an absence of will, practical knowledge, and applicable solutions. This is precisely where artificial intelligence (AI) comes in – not as an abstract vision of the future, but as a fully applicable tool available here and now.
Artificial Intelligence – Applicable Without Legislative Changes
Implementing AI in administrative activities does not necessarily require changes to the regulatory framework. It can be introduced as an auxiliary tool in the process of drafting documents, processing and analyzing data, making preparatory decisions, and even in communication with citizens. Thus, any administrative structure, regardless of the political situation or institutional inertia, can independently begin to apply AI in its daily work.
ResearchGateAI acts as a highly efficient assistant, where human intervention is reduced to final control, signing, and taking responsibility – just as in current procedures performed by humans.
How Does a Language Model Work in Administration?
Modern language models can be trained within a few months to handle complex administrative procedures. The initial "hallucinations" in the model's output are reduced by providing access to high-quality, structured data, with accuracy stabilizing over time with feedback.
AI is capable of:
- analyzing, classifying, and structuring information of any type and volume;
- drafting documents, analyses, opinions, and even court rulings, without grammatical and logical errors;
- recording every change and correction, which allows for remote and complete control of the process.
Necessary Regulatory Steps
Legal Possibility Without Changing the Law
A common argument against the application of AI is that it requires significant legislative changes. This is inaccurate. Many technological solutions can be implemented within the existing legal framework through internal acts and rules.
Internal Rules as a Legitimate Basis
According to the law, administrations can independently regulate their internal organization, including the use of software solutions. The requirement for a signature from an authorized official ensures compliance with the principle of legality and administrative responsibility, regardless of whether the text was drafted by a human or by AI.
An Auxiliary Tool, Not a Subject of Power
Bulgarian law does not prohibit the use of automated means for preparing documents. Provided that:
- the system works with strictly defined data,
- it does not make independent legal decisions,
- it is under constant human supervision, its use is permissible and completely legal.
The Principle of Technological Neutrality
Neither GDPR nor the new AI Act requires specific technologies – what is important is compliance with the principles of personal data protection, non-discrimination, transparency, and accountability. Thus, internal regulation is sufficient, as long as the AI does not contradict the law.
Legal Analogy with Other Countries
Many EU countries (e.g., Estonia, Austria, Finland) already apply AI in their administration through internal acts. The Bulgarian system allows the same through legal analogy, which opens up the possibility of implementation without a new legislative framework.
Advantages of Artificial Intelligence
- Elimination of human errors – grammatical, logical, technical.
- Exclusion of subjectivity – decisions are justified and consistent.
- Objective analysis of facts – based on verifiable data and models.
- Consistency of assessments – especially in repetitive cases.
- Traceability and transparency – every action is recorded.
AI Training: Timelines
Sample Timelines for AI System Training
| Task Type | Training Time |
|---|---|
| Simple Administrative Actions | 2–6 months |
| Specific Functions with Context | 12–24 months |
| Autonomous Decisions with Legal Effect | 12–30 months |
Security and Legal Compliance
Using AI as an auxiliary tool requires:
- data minimization, anonymization, or pseudonymization;
- compliance with the usual legal basis for processing;
- encryption and a full record of all actions;
- the possibility of an audit and an ethics/supervisory committee.
The difference between an auxiliary AI and full automation is significant. The former requires only internal regulation, while the latter presupposes legislative and infrastructural changes.
What Does This Mean in Practice?
AI can participate in drafting documents, notifications, reports, and decisions, adhering to set criteria and algorithms. This ensures:
- Elimination of the subjective factor;
- Full traceability of every step;
- Control and responsibility on the part of the employees.
In case of an unjustified human correction, the AI can generate a signal – to the signatory and their superior. This creates a new standard of accountability and deters corrupt practices.
Public Procurement: A Field for Reform
The sphere of public procurement is vulnerable to corruption and subjectivity. AI offers a solution:
- Formulation of conditions – based on objective and legal criteria;
- Evaluation of offers – automatic, without bias;
- Traceability – every deviation is documented and verifiable.
AI introduces a new benchmark – one that is logical, legally sound, and procedurally robust. This makes opacity and corruption virtually impossible.
How Much Does This Cost? (Indicative Budgets)
Budget Framework for a Simple AI System
| Activity | Costs (in BGN) |
|---|---|
| Analysis and Specifications | 30,000 – 70,000 |
| Data Preparation | 40,000 – 100,000 |
| Development and Training | 100,000 – 250,000 |
| Infrastructure | 50,000 – 150,000 |
| Staff Training | 30,000 – 80,000 |
| Maintenance (1 year) | 50,000 – 120,000 |
| Total (in BGN) | 300,000 – 770,000 |
Budget Framework for a Complex AI System
| Activity | Costs (in BGN) |
|---|---|
| Legal Expertise | 70,000 – 150,000 |
| Data Annotation | 150,000 – 300,000 |
| Hybrid System | 250,000 – 500,000 |
| Infrastructure (Complex) | 100,000 – 300,000 |
| Simulations and Pilot Phase | 100,000 – 200,000 |
| Maintenance (Complex, 1 year) | 100,000 – 200,000 |
| Total (in BGN) | 770,000 – 1,650,000 |
Cultural Transformation: The Real Challenge
The biggest barrier is not technological or legal – it is cultural. It requires will from the leadership, readiness from the employees, and understanding from the public.
But the good news is: it can all start right away, without waiting for a change in the law or a new decree. AI is here – we just need to allow it to be our partner for a transparent, efficient, and honest administration.
While the nationwide implementation of AI is a thing of the future, current legal battles require expertise today. Issues like unjustified civil forfeiture claims or the implementation of complex regulations like the WPA are already a reality for citizens and businesses. As a lawyer working at the edge of innovation, my goal is to offer the best modern solutions for today's legal challenges.
Разширена научна обосновка на този модел е достъпна в ResearchGate STUME Journals.