Abstract
Recent breakthroughs in AI capability have been attributed to increasinglysophisticated architectures and alignment techniques, but a simpler principlemay explain these advances: memory makes computation universal. Memory enablesuniversal computation through two fundamental capabilities: recursive statemaintenance and reliable history access. We formally prove these requirementsare both necessary and sufficient for universal computation. This principlemanifests across scales, from cellular computation to neural networks tolanguage models. Complex behavior emerges not from sophisticated processingunits but from maintaining and accessing state across time. We demonstrate howparallel systems like neural networks achieve universal computation despitelimitations in their basic units by maintaining state across iterations. Thistheoretical framework reveals a universal pattern: computational advancesconsistently emerge from enhanced abilities to maintain and access state ratherthan from more complex basic operations. Our analysis unifies understanding ofcomputation across biological systems, artificial intelligence, and humancognition, reminding us that humanity's own computational capabilities haveevolved in step with our technical ability to remember through oral traditions,writing, and now computing.