Space Regulatory Glossary AI & Data Science

confidence scoring

noun

1. A method of attaching a numerical reliability estimate to an AI system's output, indicating how certain the system is about a given extraction, classification, or answer. Low-confidence results can be flagged for human review rather than accepted at face value.
2. In compliance-critical applications, confidence scoring determines the boundary between automated processing and human oversight. A filing extraction system might auto-accept results above 0.95 confidence but queue anything below that threshold for manual verification.

Read: Benchmarking LLMs for Domain-Specific Extraction

Related terms

hallucination An AI model output that is factually incorrect, fabricated, or unsupported by the source data, presented with the same… large language model A neural network trained on large text corpora that generates, summarizes, and reasons about natural language. Models… structured extraction The process of converting free-text or semi-structured documents into machine-readable data with defined fields, types,…

Mentioned in The Downlink

• AI Verification Engineering: Architecture for Trustworthy Regulatory AI

  RAG hallucinates at 17 to 33 percent in production legal tools. The architecture of trustworthy regulatory AI: source grounding, extraction validation, confidence scoring, and standards-aligned verification.

Browse all terms in the Space Regulatory Glossary →