Is AI The Next Stage Of Human Evolution? - Robert Wright - #1122

What It Covers

Robert Wright, author of The Moral Animal, discusses his new book on AI as an evolutionary threshold event. He argues AI reverse-engineers millions of years of cognitive evolution using data alone, examines doom scenarios, international coordination failures, and why overcoming tribalistic moral biases is the prerequisite for navigating this technology safely.

Key Questions Answered

• •AI as Convergent Evolution: AI training independently rediscovers biological solutions — edge-detector neurons, word-meaning representation — that evolution took millions of years to build. No one programmed these capabilities; they emerged from data-driven reinforcement. This convergence suggests AI isn't just a tool but a parallel evolutionary process, making its trajectory harder to predict or control than conventional software development.
• •The Data Sufficiency Principle: Modern AI requires only human-generated input and output data to replicate cognitive functions — no hand-coded rules or meaning dictionaries. Zuckerberg's reported keystroke-tracking initiative illustrates this: capture what workers receive and produce, and AI can replicate the cognitive process between. Any job reducible to data inputs and outputs faces near-term automation risk regardless of apparent complexity.
• •Task-Duration Doubling Rate: Research tracking AI performance on timed human tasks shows the maximum task duration AI can complete successfully has been doubling every seven months — a Moore's Law equivalent accelerating further as coding agents now help build successor models. This self-reinforcing loop means capability gains compound faster than most institutional or regulatory responses can match.
• •International Coordination as Technical Requirement: Unlike nuclear arms control, AI cannot be managed through bilateral treaties alone because verification is structurally harder and risks are non-territorial. Wright argues "organic transparency" — dense scientific, economic, and cultural exchange between nations — reduces miscalculation risk more reliably than formal agreements, making geopolitical de-escalation a literal AI safety mechanism, not merely a diplomatic preference.
• •Cognitive Empathy Over Emotional Empathy: Wright distinguishes cognitive empathy — understanding another party's perspective without feeling it — from emotional empathy. In non-zero-sum international AI negotiations, cognitive empathy is sufficient and more achievable. Practical cultivation: use AI tools deliberately configured to steelman opposing positions rather than defaulting to sycophantic agreement, which market incentives will otherwise produce automatically.
• •Collective Intelligence Already Exists: Superintelligence is partially already present in human form — Boeing collectively knows how to build airliners though no individual does. AI systems communicating and collaborating with each other extend this dynamic non-linearly. Even without new architectural breakthroughs, refining existing model applications into workplaces over two to three years would produce breakneck practical advancement by this mechanism alone.