I. Concept of Artificial Intelligence
1. Technical Definition of Artificial Intelligence
Artificial intelligence is not a judgment subject similar to a human. It is a computational system composed of a model, parameters, data, and computational procedures. Artificial intelligence is not an entity that sets its own purposes or forms judgment standards, but a technical system that performs computation according to externally given conditions and rules.
2. Components of Artificial Intelligence
- Model: a function structure that converts input into output
- Parameters: numerical values that determine the operation of the model
- Data: the set of inputs used by the model to perform computation
- Computational Procedures: algorithms and processing order for calculating outputs
3. Meaning of Objective Function and Optimization
Artificial intelligence always operates on the premise of an objective function. The objective function is always set outside the system, and the output of artificial intelligence is entirely dependent on how the objective function is designed and applied.
4. Technical Meaning and Scope of Autonomy
Technical autonomy is limited to the range of selecting computational paths and does not mean autonomy of judgment or autonomy of responsibility. Judgment acts such as setting standards, changing standards, and confirming conclusions are not included.
II. Structural Limits of AI Autonomy and Errors
1. Inseparability of Autonomy and Error Possibility
As a computational system, artificial intelligence inherently contains the possibility of errors depending on the limitations of input data, simplification of the model, and similar factors.
2. Inability to Attribute Responsibility and Procedure When Errors Occur
Artificial intelligence cannot be a subject capable of bearing legal or ethical responsibility for computational results, nor can it be a subject of responsibility determination for procedural violations or changes in standards.
3. Structural Risks Caused by Inability of Attribution
- The responsible subject becomes unclear.
- Attribution and correction of procedural violations become impossible.
- There arises a risk that changes in standards or omissions of procedure may be justified after the fact.
4. Conclusion
The autonomy of artificial intelligence must necessarily be limited on the premise of structural control.
III. Related Regulations and Their Premise
- UNESCO Recommendation on the Ethics of Artificial Intelligence: human dignity, accountability, transparency, and human control
- OECD Recommendation on Artificial Intelligence: human-centeredness, trustworthiness, responsible innovation, and risk-based approach
- Framework Act on Artificial Intelligence of the Republic of Korea: definition of high-risk AI, protection of citizens’ rights, and securing safety
These regulations are not designed to affirm the autonomy of artificial intelligence, but are normative devices for controlling the problem of inability to attribute responsibility and procedure.
IV. Necessity of Structural Control
For autonomy to be permitted on a limited basis, the following requirements must be secured structurally.
- The judgment subject must be clearly fixed as a human or institution.
- The maximization of efficiency or performance must be restricted so that it cannot infringe legal requirements and rights protection.
- Omission of procedures or change of standards must not be permitted.
- When control fails, the attribution of responsibility must be clearly specified.
Autonomy in the absence of these requirements amounts to leaving risks unattended and directly conflicts with the legislative purposes of international ethical norms and domestic laws.