What We Don't Understand Well
Exact Thresholds and Distributions: There is no well-established minimum IQ (or equivalent g-factor measure) for "effective" collaboration at different levels — basic vs. expert. Conversions from AFQT/military data or general IQ don't directly map to AI-specific skills. How low can someone go while still gaining major benefits through good interfaces/training? How high for frontier work?
How Much AI Flattens vs. Amplifies Differences: Does AI act as a great equalizer (helping lower-ability users close gaps) or a multiplier (widening gaps as high-ability users leverage it better for complex prompting/iteration)? Early evidence is mixed; cognitive diversity and unequal access sometimes boost teams.
Role of Non-g Factors: IQ/g is a strong predictor of many outcomes, but metacognition, conscientiousness, creativity, emotional intelligence, and "AI literacy" (prompting + reflection) may matter more in dynamic collaboration. We don't fully know the relative weights or how trainable they are.
Long-Term Dynamics and Atrophy: Does heavy AI use erode human skills over time (cognitive offloading)? Or build new hybrid capabilities? Effects on learning, intuition, and skill development are understudied, especially across ability levels.
Context-Dependence and Complementarity: Optimal human-AI pairing varies hugely by task (content creation gains vs. decision-making losses). We lack comprehensive models for when/why humans add unique value (intuition, ethics, novel problem framing) vs. becoming bottlenecks.
Measurement Challenges: Traditional IQ tests don't capture "human-AI synergy" well. New frameworks (e.g., collaborative ability metrics) are nascent.