Question 1

What are decision boundaries in an AI project?

Accepted Answer

An explicit definition of which decisions the AI takes autonomously and which require human approval, with authorization levels per decision type. In regulated sectors this is a legal requirement under the EU AI Act, not optional.

Question 2

What is model drift and why does it matter?

Accepted Answer

Model drift is the gradual degradation of AI performance as production data diverges from training data. Without monitoring with metrics, alerts and a retraining schedule, drift is only noticed when damage is already large.

Question 3

When does the EU AI Act apply?

Accepted Answer

The EU AI Act enters force in phases. Non-compliant high risk AI systems are actively sanctioned from 2026 with fines at GDPR level. Compliance starts with a DPIA-style assessment and explicit human-in-the-loop boundaries.

Question 4

Why does an AI project need a fallback scenario?

Accepted Answer

An AI that disrupts the process when it fails is more expensive than no AI. A working manual or rule-based fallback is operational continuity, not a luxury, especially for production-critical systems.

Question 5

Who owns the data and the trained model?

Accepted Answer

Default AI contracts often leave data and model ownership vague in the vendor's favor. The party that holds the data has the leverage; explicit contractual clauses for data and model portability are essential before signing.

Is your AI project ready for development?

Ten expert questions about your AI or machine-learning project