Templates & forms

Clear boundaries for automated systems

This chapter describes, how Selmo classifies the use of AI and automated code generation and why formal models are the decisive prerequisite for accountable AI use are.

It's not about evaluating AI, but about clearly limiting its sphere of action.

---

The starting point

The use of AI in automation is increasing:

• Code generation

• Suggestions for procedures

• Optimization of parameters

• Assistance systems in engineering

This raises a central question:

Who bears responsibility for the behavior of the machine, when the code is (partially) generated automatically?

This question cannot be answered at the code level.

---

Code is not an object of responsibility

Code describes:

• an implementation

• a technical implementation

• a concrete manifestation

Code describes not:

• the intention

• the permissibility

• the responsibility for behavior

Regardless of whether code:

• was created by a human

• by a generator

• or by an AI

was created, the following applies:

Responsibility cannot be anchored in the code.

---

The model as the responsible instance

In Selmo responsibility lies not in the code, but in the formal behavior model.

The model defines:

• which states exist

• which behavior is allowed

• which conditions are mandatory

• which reaction occurs in case of deviation

Thus it applies:

• Code may implement the model

• Code must not extend the model

• Code must not violate the model

The model limits the code – regardless of who generates it.

---

AI as a tool, not a decision-maker

In a Selmo-based system AI can:

• generate code

• make suggestions

• compare variants

• support optimizations

However, AI may not:

• decide which states are allowed

• change safety assumptions

• bypass monitoring logic

• take on responsibility

AI may implement and optimize – but not define what is permissible.

---

Determinism as a protective mechanism

The deterministic character of the Selmo model is crucial for the use of AI:

• same states → same behavior

• same deviations → same reactions

• no implicit assumptions

This prevents:

• AI from introducing implicit logic

• behavior from becoming inexplicable

• decisions from no longer being justifiable

Determinism is the limit for automation.

---

Traceability and liability

In the event of damage or an audit, key questions are:

• What behavior was allowed?

• Which condition applied?

• Why did the system react?

With a formal model these questions can be answered, regardless of whether the code was produced manually or automatically.

Liability follows the model – not the origin of the code.

---

Delineation

To clarify:

• Selmo is no AI

• Selmo is no code generator

• Selmo is no automation AI

Selmo is:

the formal framework, in which AI can be used responsibly.

---

Summary

The use of AI in automation is inevitable – but responsibility remains human.

Selmo ensures that:

• behavior is explicitly defined

• code (including AI-generated) remains constrained

• traceability is ensured at all times

AI can generate code. The model carries the responsibility.