AI Now Building Better AI: Researchers Leap Ahead

Two prominent AI researchers have departed Google to establish a London-based startup, Inherent, focused on developing artificial intelligence systems capable of improving themselves. Their prototype, Faraday, analyzes vast amounts of data from researchers' activities to iteratively enhance its own capabilities. This pursuit of recursive self-improvement (RSI) is also being chased by major players like OpenAI and Anthropic, alongside other startups.

These efforts aim to drastically accelerate AI development, potentially leading to breakthroughs in science and medicine, and even surpassing human intelligence. However, the prospect of AI improving itself without human intervention raises significant concerns about loss of control and potential existential risks. Some experts warn that a rapidly self-improving AI could outpace human understanding and control.

The concept of RSI is not new, dating back to the early days of AI research in the 1950s. While early attempts were hampered by computational limitations, advancements in neural networks and AI's ability to generate code have revived interest. Companies like Inherent and Recursive Superintelligence are pushing beyond current methods, seeking AI that can design entirely new AI architectures.

Despite the rapid progress, human oversight remains critical. Current AI systems, while capable of generating code and exploring research avenues, still lack the common sense, creativity, and wisdom of human researchers. The true challenge lies in developing AI that can generate novel ideas for its own improvement, a step many believe is still some decades away.