As We May Think No More: From Bush's Memex to AI Alignment (memoryleak.substack.com)

In a reflective essay tracing the lineage from Vannevar Bush’s 1945 vision of the memex to today’s AI revolution, the article highlights the evolving challenge of managing human knowledge amid technological upheaval. Bush’s memex—a mechanized desk designed to augment human memory through associative trails linking documents—foreshadowed the digital age’s networked information systems and inspired pioneers like Douglas Engelbart and Ted Nelson. Yet, while the internet and the World Wide Web realized the memex’s dreams of interconnected knowledge, they also unleashed an overwhelming flood of information, complicating rather than simplifying our search for truth. Fast forward to the present, Geoffrey Hinton’s work on neural networks has transformed Bush’s associative concept into active machine learning, where AI models autonomously create connections and extract patterns from vast datasets. This breakthrough, powered by deep learning and massive computational resources, underpins today’s generative AI systems like GPT-4, capable of emergent reasoning and rapid learning. However, this progress carries new risks: rather than organizing knowledge, large language models now contribute to an unprecedented proliferation of AI-generated misinformation, synthetic content, and deepfakes that blur reality and threaten shared truth. The greatest concern emerging from this trajectory is the AI alignment problem—ensuring superintelligent systems remain controllable and aligned with human values. Hinton’s shift from optimistic pioneer to cautious alarmist underscores the existential urgency of addressing how advanced AI might surpass human intelligence yet escape our control, echoing timeless anxieties about creation and mastery. This synthesis of historical vision and contemporary breakthrough challenges the AI/ML community to rethink governance and ethical frameworks amid accelerating technological power.