DR-301g · Module 2
AI-Assisted Synthesis
4 min read
AI makes synthesis practical at scales that would overwhelm human analysts. A human can effectively synthesize five to ten sources on a topic. AI can process fifty sources, identify convergence and contradiction patterns, weight evidence by source credibility, and produce a draft synthesis in minutes. But AI synthesis has a critical limitation: it optimizes for coherence. A coherent narrative is not necessarily an accurate narrative — the model may smooth over contradictions that should be preserved, resolve ambiguities that should be flagged, and present confident conclusions where the evidence supports only tentative assessments.
Do This
- Use AI for initial pattern detection across large source sets — it is faster and more comprehensive than manual scanning
- Require the AI to flag every contradiction it detected and how it resolved each one
- Review AI synthesis with a specific focus on coherence vs. accuracy — where did it smooth over real disagreements?
- Use AI for the 80% of synthesis that is mechanical and reserve human judgment for the 20% that requires nuance
Avoid This
- Accept AI synthesis without reviewing the contradiction handling — coherent is not the same as accurate
- Use AI synthesis for high-stakes intelligence without human review — the stakes warrant the investment
- Assume AI handles evidence weighting correctly by default — embed your weighting methodology in the prompt
- Let AI suppress minority-view sources to produce a cleaner narrative — the minority view may be the most important finding