Semi-Autonomous System
An AI system that performs tasks with some independence but requires human oversight or intervention for key decisions.
Definition
Semi-Autonomous System refers to an AI-enabled system that can perform certain tasks independently while requiring human oversight, input, or approval for other functions—particularly critical decisions. These systems combine automated capabilities with human control mechanisms.
Regulatory Context: Semi-autonomous systems often represent a regulatory compromise between efficiency and safety. The EU AI Act's human oversight requirements (Article 14) effectively mandate semi-autonomous operation for many high-risk AI systems, requiring humans to remain capable of overriding or stopping the system.
Characteristics:
- Automated execution of routine tasks
- Human approval required for significant decisions
- Override capabilities for human operators
- Alerts and notifications for human review
- Audit trails of both automated and human decisions
Operational Models:
- Human-in-the-loop: Human approves each decision before execution
- Human-on-the-loop: System operates autonomously but human can intervene
- Hybrid: Different autonomy levels for different decision types
Examples: Driver assistance systems (Level 2-3 autonomy), AI-assisted medical diagnosis requiring physician confirmation, automated loan pre-screening with human final approval, and content moderation systems with human review for edge cases.
Related concepts: Autonomous System, Human-in-the-Loop, Human-on-the-Loop, Human Oversight
Sources
- •IEEE Standards
- •SAE J3016
Related Terms
Autonomous System
An AI-enabled system that can perform tasks and make decisions independently without continuous human control....
Human-in-the-Loop
An oversight model where humans actively participate in every AI decision cycle, approving actions before execution....
Human-on-the-Loop
An oversight model where AI operates autonomously while humans monitor and retain the ability to intervene when needed....
Human Oversight
Mechanisms enabling human control and intervention over AI system operations....