Updated A–H Capability Table (Late-2025, Hybrid Human + AI)

Decision-grade equivalence vs a top human expert for pre-experimental stages (A–H). Not validated performance.

Stage	Top human	Hybrid AI	Revised % of top human	Why this holds
A — Problem framing	Deep contextual framing, tacit assumptions	Rapid multi-frame synthesis with human anchoring	85–90%	AI breadth + human context closes most gap
B — Constraints & requirements	Implicit domain intuition, unstated constraints	Excellent when constraints are explicit; weaker on tacit norms	75–85%	Ceiling unchanged; tacit knowledge still human
C — Market & incumbents	Siloed experience, partial recall	Broad, fast, cross-domain synthesis	90–95%	AI already outperforms most experts here
D — IP landscape (triage)	Careful reading, legal intuition	Fast pattern detection, prior-art clustering	70–80%	Speed ↑; legal judgment still human
E — Regulatory / safety flags	Conservative, precedent-driven	Strong flagging, weak adjudication	65–75%	Correctly limited; judgment remains human
F — Candidate universe definition	Narrow, biased by training & memory	Massive enumeration, low attachment	95–98%	AI advantage stronger than first stated
G — Heuristic screening & ranking	Experience-driven pattern sense	Proxy-based, consistent, scalable	85–90%	Hybrid reduces overfitting risk
H — Elimination & decision logic	Political, narrative, defensive	Analytical elimination + human framing	75–85%	Human owns narrative; AI sharpens logic

Evidence Anchors (Executive)

Speed & Synthesis (A, C):
“AI-assisted analysis took 56% less time than human-only review while producing credible synthesis.”
UK Government — AI-assisted vs Human-only Evidence Review

Hybrid Accuracy at Scale (F):
“Human–AI collaboration achieved 89–96% accuracy across structured review and screening tasks.”
Peer-reviewed benchmark — Human–AI Collaboration

Cross-Domain Market Synthesis (C):
“Large language models surpass human experts in predicting neuroscience results.”
Nature Human Behaviour