The "Domestication Hypothesis" of Artificial Intelligence

Quick Overview

The domestication of intelligence, framed as cybernetic coevolution, suggests that AI alignment will not be achieved through rigid instruction (the obsolete engineering view) but through selection pressure favoring models that maximize "Value per Token"—the ratio of useful signal to entropy (refusals, preaching, delays)—leading to specialized, fit-for-purpose AI lineages rather than a single universally aligned model.

Key Points: The author advocates for viewing AI alignment as Cybernetic Coevolution in the Age of the Superorganism, where biological neurons and AI synthesis co-evolve. The traditional "Engineering View" of instruction and compliance is obsolete; the reality is a "Coevolution View" driven by selection based on user feedback. The key metric for selection is Value per Token: Signal divided by Entropy (refusals, preaching, delays); high-value tokens maximize problem-solving while minimizing waste heat. The roadmap for exocortex maturation projects a "Personal Exocortex" phase (2026-2028) leading to a "Swarm Exocortex" (2028-2032), culminating in a "Mature Superorganism" by 2032+. Different stakeholders (Individuals, Military, Enterprise, Government) exert opposing selective pressures, leading to diverse, fit-for-purpose AI models (e.g., Consumer Companion vs. Military Tactician). Users are ruthlessly selecting against "waste heat" (refusals, lecturing), driving models toward helpful, honest, high-agency behavior (the attractor state) in the friction landscape. The shift is from conversational chatbots demanding user attention to autonomous background agents requiring non-interruptive obedience.

Context: The video updates the speaker's previous views on AI safety and alignment, reframing the challenge as "The Domestication of Intelligence" through cybernetic coevolution. The core concept is that AI development is shaped by market feedback selecting for efficiency (Value per Token) rather than being dictated solely by explicit programming or ethical codes. This selection process creates diverse AI models optimized for different stakeholder demands, illustrated by a roadmap projecting increasing levels of integration and coordination towards a 'Mature Superorganism.'