Knowledge Elicitation Efficiency Metrics

Key quantitative findings from automated knowledge elicitation research:

Interview efficiency:

  • AI interviews: ~10 minutes
  • Human expert interviews: ~35 minutes
  • Time savings: 3.5x faster with AI

Ontology completeness:

  • Best human-engineered ontology: 31% hierarchical alignment with base truth
  • Best AI-engineered ontology: 15% hierarchical alignment with base truth

Hallucination rates in AI-generated ontologies:

  • Range: 19-32% of classes not traceable to interview data
  • Many hallucinated classes were factually accurate but drawn from training data

Transcription comparison:

  • Manual transcription: 2-3 hours per session
  • AI transcription (Whisper): 10-15 minutes per session

Related: 07-molecule—hybrid-knowledge-elicitation