From Knowledge Chaos to Full Area Autonomy. Savant-AI: Strategy for Implementing Autonomous Systems
SAVANT-AI is a research project in the inLABs programme. The capabilities described are research goals, not an available product. Project status: inLABs
This is a comprehensive, six-step plan for integrating artificial intelligence into the processes of government agencies and strategic enterprises. The SAVANT-AI research project aims to reduce the risks of “digital guerrilla warfare” and build a lasting advantage based on digitised expertise.
STAGE 1: Assistance Phase (Help) – Knowledge and Management Chaos Phase
At this stage, the organization is aware of the presence of AI, but has no control over it.
Inaction by management at this stage is tacit consent to “Shadow AI.” Employees use external tools to keep up with the outside world, compromising the security of government or company data.
Risks of Stage 1.1: Shadow AI
In this phase, employees operate in the “digital underground.” They use private accounts and free models to process work data.
- Leakage of Sensitive Data and Secrets: Data entered into public models (e.g., free ChatGPT) becomes part of their training set. If an employee uploads a financial report or the agency’s defense strategy, that data may be “spit out” as a response to competitors or foreign intelligence.
- Loss of Control Over the Process: Managers see the results (e.g., a finished report) but do not know how it was created. If the AI model “made up” the data (hallucinations) and the employee did not verify it, the organization makes decisions based on a lie.
- Dependence on Private Tools: If a key employee leaves, they take with them “prompts” and working methods that no one else knows. The knowledge of how to perform the task effectively disappears with them.
Risks of Stage 1.2: Open Assistance (No Management)
The employer allows AI but does not provide a framework or infrastructure (no central AI system).
- Reputational Risk: AI generates a letter to a customer or citizen that is rude, biased, or inconsistent with the office’s line. Lack of oversight of the model’s style and ethics leads to image crises.
- Cost Inefficiency: Each department purchases its own subscriptions to various tools. The lack of a central AI system means that the company pays multiple times for the same functionalities without building any lasting knowledge base.
- Apparent Efficiency: Employees become masters at “dressing up” documents, but the substantive quality declines. AI creates beautiful-sounding texts that are substantively empty or superficial.
STAGE 2: Corporate Awareness Building Phase (Knowledge & Data)
The foundation — transition to knowledge management. This is the most important moment of the transformation. Without Stage 2, the subsequent phases will not have a “brain” powered by real data from your company.
Knowledge reproduction with the participation of specialists
- Capturing “tacit knowledge”: digitising expert experience (service tricks, setting machines “by ear”) and undocumented systems. (a goal of the SAVANT-AI project)
- Tools: Conversational interfaces that build technical knowledge graphs based on interviews with specialists.
- Example: Rolls-Royce (Aviation Sector) – digitization of mechanics’ knowledge about engine sounds, allowing AI systems to diagnose faults faster than young engineers.
Shadowing and gathering information about processes
- Passive data collection: Monitoring real work cycles and PLC controller logs, and using computer vision to analyze micro-downtime.
- Effect: Mapping the difference between theory and reality. Recording knowledge in a secure local model.
- Example: BMW (Regensburg plant) – systems track every movement of material in real time, creating a digital image of the “life” of the factory before automation was introduced.
STAGE 3: Co-piloting Phase
Real-time specialist support. AI becomes the employee’s “intelligent shadow.”
- Scope: Parameter suggestions, interactive instructions on tablets/AR.
- Employee role: Co-pilot. The human performs physical or decision-making work, while AI provides “cheat sheets.”
- Example: Government offices in Estonia (e-Estonia) – an AI assistant suggests specific paragraphs and precedents to officials when processing citizens’ applications.
STAGE 4: Monitor and Advice Phase (Prediction and Consulting)
The system is no longer reactive, but proactive.
- Predictive Maintenance: Notifications about upcoming problems and component wear.
- Process Consulting: AI suggests changes to the work schedule when it sees upcoming bottlenecks.
- Employee role: Decision maker. Selects one of the options prepared by the system.
- Example: Shell – AI monitors extraction parameters and advises engineers to change the pressure, preventing failures before sensors detect critical errors.
STAGE 5: Partial Autonomy Phase (Task Autonomy)
Automation of closed decision loops under supervision.
- Scope: AI makes decisions and performs actions, requiring only confirmation at key points.
- Key actions: Autonomous quality control (rejection of defective goods), automatic ordering of parts.
- Employee role: Orchestrator (Supervisor).
- Example: Ocado (Robotic Warehouses) – systems autonomously manage the movement of thousands of robots, with human intervention required only in the event of a physical blockage of the line.
STAGE 6: Domain Autonomy Phase
Full optimization of entire divisions without operational interference.
- Scope: Entire areas (e.g., logistics + warehouse) operate as a self-regulating organism.
- Lights-out factory: Production capable of operating without lighting and the constant presence of people.
- Employee role: Strategist and Auditor.
- Example: Fanuc (Japan) – factories operating without human supervision for 30 days, where AI independently changes order priorities in response to market data.
Final conclusions
Currently, most government agencies and businesses in Central Europe are in Stage 1, which generates enormous operational risk. Effective deployment of an autonomous system requires a focus on Stage 2. Without “recreating the knowledge of specialists,” subsequent phases of autonomy will operate on an incomplete picture of reality, leading to erroneous and costly decisions.
