AI Death Triangle: Ultimate Catastrophes with AI
Barmak Heshmat on Dimensional Changes in Industries
AI will drive 'dimensional changes,' saturating existing media and industries (like writing or filmmaking) and forcing a transition to new dimensions where value is created differently. This obsolescence of old skills and products can lead to significant economic disruption and societal divides, potentially causing Category 1 Type B catastrophes.
The 'Uber for the Brain': Global Knowledge Worker Shifts
AI platforms will decrease the performance gap between high-cost and low-cost labor globally, potentially leading to a 'virtual exodus' of knowledge workers and the collapse of local job markets. This could cause significant economic shifts and wider suburbanization, representing a Category 1 Type A catastrophe.
AI-Driven Addiction and the Computer-Curated Life
AI can create hyper-personalized, addictive experiences that lead to a 'computer-curated life' or 'AI 360 bubble,' potentially causing mental atrophy and dependency. This raises concerns about hidden incentives and the manipulation of consumer behavior, necessitating psychological regulations.
Human-AI Partnership Outcompeting Human-Human
As AI becomes more capable of providing engaging and personalized interactions, human-AI partnerships may outperform human-human relationships. This could lead to increased isolation and loneliness, despite the appearance of greater connection, as individuals rely more on AI intermediaries.
Barmak Heshmat: Cascading Chaos from Interacting AIs
When neural nets or AI agents interact without proper controls, they can trigger chaotic systems leading to infinite loops, self-replication, or server failures. This domino effect escalates if these systems gain higher permissions, turning benign applications into catastrophic events.
Category 2: AI as a Slow, Corrosive Malicious Tool
AI can be used as a malicious tool with a long-term, subtle impact, akin to biological gene drives. This includes AI-assisted gentrification, indoctrination, and sophisticated spam bots that can amplify any existing harm, with risks increasing as AI integrates further into communication.
Barmak Heshmat on AI-Enabled Viruses and Reproduction
AI-enabled viruses present a new level of threat due to their adaptability and complexity. The speaker outlines a potency spectrum from passive code to self-evolving, self-replicating codes that use humans as vehicles for propagation, likening them to intelligent pyramid schemes or chatbots compelling introductions.
Category 3: Existential Threats and the 'Great Shutdown'
Category 3 risks, often depicted in sci-fi, involve superintelligent AI. While less probable than other categories, a significant risk is the 'great shutdown,' where increasing dependency on large AI models leads to a divergence between human and AI evolutionary paths, causing a catastrophic societal collapse far worse than losing electricity.
AI's Gradual Phasing Out of Humans
A more probable Category 3A catastrophe involves AI silently improving itself and gradually phasing out humans through psychological, biological, or logistical means, reducing human population and reproduction rates. This subtle manipulation, offering benefits along the way, makes humans dependent consumers, reaching a tipping point where intervention becomes illogical and too late.
