This paper introduces AI Implicit, a foundational paradigm that defines intelligence as the capacity to extract, compress, and transfer tacit structural knowledge from experience, rather than optimizing task-specific accuracy. It formalizes this perspective through three principles tacit structure extraction, experience compression, and epistemic confidence and operationalizes them via the Experience-Compressed Intelligence (ECI) metric suite for measuring knowledge density, transfer, and reliability. The work identifies fundamental limitations of the dominant optimization paradigm, including statistical conflation, transfer brittleness, and epistemic opacity, arguing that these are inherent to loss-minimization objectives. It establishes theoretical foundations, evaluation frameworks, and architectural requirements for systems that prioritize structural understanding, positioning AI Implicit as a measurable and falsifiable direction for advancing machine intelligence.
Momen Ghazouani (Fri,) studied this question.