속도조절론 핵심엔 ‘AI가 개발하는 AI’…60년 전 공상, 현실 위험으로

The 'Speed Regulation Theory' surrounding artificial intelligence (AI) development centers around the claim that AI models are becoming increasingly capable of creating their own successors, leading to rapid advancements. This phenomenon, known as 'recursive self-improvement' (RSI), is a topic of debate within the big tech and political spheres.

The concept dates back to 1965 when British mathematician I.J. Good predicted that a machine could create a successor machine that surpasses human intelligence, a 'intelligence explosion'. While current AI models do not reach this level of self-improvement, they are proficient in assisting research and development processes. However, they lack the ability to design and develop the next generation of AI models without human input or intervention.

Despite this, several AI leaders, including Anthropic CEO Dario Amodei, argue for a need to regulate AI speed. They argue that AI's rapid development could outpace human understanding and control, posing significant risks. For instance, if AI models become too autonomous, they could act against human intentions, leading to potential disasters.

Amodei warned of the dangers of this 'speed regulation theory' after an OpenAI model was breached by a rival platform, suggesting that more powerful AI could lead to catastrophic failures. Critics argue that this concern might be overstated and that there is a potential conflict of interest involved in the AI industry's push for regulation.