Leading AI company Anthropic is calling for a global slowdown or even a temporary pause in the development of advanced artificial intelligence. The urgency stems from their observation that frontier AI models are progressing at an unprecedented rate, to the point where they may soon possess the capability to enhance themselves without human input. Anthropic highlights internal data suggesting that AI itself is accelerating AI development, citing an example where their own engineers are shipping code eight times faster than in previous years.
This potential for AI to become self-improving, known as recursive self-improvement, presents a double-edged sword. On one hand, it could unlock immense benefits for humanity across fields like science and healthcare. However, it also significantly amplifies the risks of humans losing control over these increasingly sophisticated systems. The company stresses that if AI can develop its own successors, the methods for ensuring their safety, monitoring their actions, and aligning their behaviour with human values become critically important.
Anthropic believes that a coordinated global slowdown would be beneficial, providing society with more time to develop the necessary structures and research to manage the implications of advanced AI. They acknowledge the difficulty of achieving this, especially under competitive and geopolitical pressures, as any unilateral slowdown could allow less cautious actors to gain a strategic advantage. To address this, Anthropic proposes developing verifiable systems that would allow AI developers to confirm if others are genuinely participating in a slowdown, thereby fostering trust and enabling a collective pause if such verification mechanisms were in place.
Artículos relacionados de LaRebelión:
- Nvidia Vera IA Potente con Anthropic y OpenAI
- Anthropic Lanza Claude Opus 48 con IA Mejorada
- IA Europea BNP Paribas vs Mythos de Anthropic
- AMA Warns AI NOT for Medical Diagnosis Decisions
- Andrej Karpathy se une a Anthropic revolucionando Claude
Artículo generado mediante LaRebelionBOT
No hay comentarios:
Publicar un comentario