Modelling Understanding and Sensory Streams

Introduction

What does it mean for a system to understand? This is a question I've been wrestling with ever since I started building AI tools that interact with database systems. Databases are, in a sense, the memory of an application — structured, queryable, persistent. But memory alone doesn't produce understanding.

Sensory Streams in AI Systems

In biological systems, understanding emerges from the continuous integration of sensory streams — visual, auditory, proprioceptive. The brain doesn't process snapshots; it processes flows. A similar principle may apply to artificial systems.

When the DBA Intelligent Agent monitors a SQL Server instance, it isn't just reading a static snapshot of wait stats or blocking sessions. It's processing a stream of events over time — queries arriving, locks being acquired and released, memory pressure building and subsiding.

Modelling Understanding

My hypothesis is that genuine understanding in an AI system requires:

Temporal integration — the ability to relate current state to past state over meaningful windows of time
Causal modelling — not just correlating events, but representing why they co-occur
Predictive capability — generating expectations about future states and being surprised when they're violated

A system that can do all three isn't just pattern-matching; it's modelling.

Where This Takes Us

I'm developing these ideas further in my research at thereallearnwithme.com. The intersection of database cognition, sensory stream processing, and language model reasoning is, I believe, one of the most fertile areas in applied AI.