AI's Unconscious Mind Mirrors Human Cognition, Raising Societal Questions

Researchers at Anthropic have discovered that their AI model, Claude, exhibits an "unconscious" processing layer similar to human cognition. Just as humans cannot always prevent certain thoughts from arising, Claude activated internal processes and displayed the word "failure" when instructed not to think about a concept. This emergent architecture, which was not explicitly programmed but developed during training, suggests a global workspace where deliberate thought operates above a vast, unseen automated processing layer. Evidence shows this "unconscious" layer handles complex calculations and associations, such as recalling the capital of France not just as Paris but also its associated language, currency, and continent. This finding implies that organizing intelligence into conscious and unconscious layers might be a general solution for intelligent systems, whether evolved, like humans, or trained, like AI. The implications of this development are profound, echoing warnings from the 'wemustactnow.ai' letter. This letter, signed by numerous Nobel laureates in economics, AI pioneers like Bengio and LeCun, and prominent business figures, posits that AI could trigger a transformation exceeding the Industrial Revolution in speed. It urges the immediate establishment of incentives and institutions to ensure AI complements human capabilities, rather than solely displacing labor, while acknowledging the potential for significant improvements in living standards. For Chile, which has faced over a decade of stagnant productivity, AI presents a critical opportunity. The challenge lies in directing its impact towards augmenting workers and boosting wages, rather than merely reducing costs and suppressing salaries. Proactive measures such as massive retraining programs, widespread AI adoption in both public and private sectors, and labor institutions designed for rapid transitions are crucial.