Scholarly abstract The information retrieval industry is spending USD 4. 7 billion to rediscover a solution that was distributed for free in 1985. While modern vector databases struggle to scale beyond 100 million documents, hitting a wall of GPU inefficiency and exorbitant costs, we identify a fundamental pattern: every successful modern system—from NVIDIA's TiDAR to Microsoft's SPANN—has independently converged on the architectural principles of CDS/ISIS, a UNESCO system designed for developing nations. This convergence is not a coincidence; we argue it is an architecturally inevitable response to fundamental limits of information organization. We propose the Two-Level Theory, suggesting that efficient retrieval requires separating semantic understanding (Level 1) from structural organization (Level 2). Modern systems fail when they force neural tools to perform symbolic work. We present the Neural-to-Symbolic Bridge, a proposed framework that uses modern AI to populate classical B-tree indices, predicting 10× cost reduction (2. 4M → 240k). Plain-language summary Modern AI companies are spending billions trying to search through massive document collections quickly. They're hitting a wall: neural networks are terrible at organizing data for fast retrieval. We noticed something remarkable: every successful system has unknowingly recreated the same solution that UNESCO gave away for free in 1985. A system called CDS/ISIS, designed for computers with less memory than your smartwatch, solved this problem decades ago. The key insight: understanding what something means and finding where it's stored are fundamentally different problems requiring different tools. Significance statement This paper identifies architectural convergence as evidence for fundamental information-theoretic constraints on retrieval systems. The Two-Level Theory provides a theoretical framework explaining why billion-dollar systems inevitably rediscover forty-year-old architectures, with implications for AI infrastructure investment and system design.
Kafkas M. Caprazli (Wed,) studied this question.