(Image credit: Yuichiro Chino/Getty Images)
The rapid development of artificial intelligence (AI) technologies is raising questions about the limits of the field. Topics that were once considered science fiction – such as the concept of superintelligent AI – are now being seriously discussed by scientists and professionals.
The concept that machines could one day match or even surpass human intelligence has a long history. However, the pace of AI advances in recent decades has made the topic relevant again, especially with the advent of powerful large language models (LLMs) from companies like OpenAI, Google, and Anthropic, among others.
Experts differ widely on the feasibility of the idea of “artificial superintelligence” (ASI) and when it will appear, but some believe that such highly advanced machines are already close. One thing is clear: if ASI does emerge, it will have a huge impact on the future of humanity.
“I believe we will enter a new era of automated scientific discovery, greatly accelerated economic growth, extended lifespans, and new forms of entertainment,” Tim Rocktesel, a professor of AI at University College London and a lead researcher at Google DeepMind, told Live Science, emphasizing that this is his personal view and not Google DeepMind’s official position. But he cautioned, “As with any significant technology in history, there are potential risks.”
What is Artificial Superintelligence (ASI)?
Traditionally, AI research has focused on replicating specific skills that intelligent creatures demonstrate. These include the ability to visually analyze their surroundings, understand language, or navigate spatially. In some of these specialized areas, AI has already demonstrated abilities beyond those of humans, Rokteshel noted, particularly in games like Go and chess.
However, the ultimate goal of the field has always been to replicate a more general form of intelligence seen in animals and humans, combining many of these skills. This idea has gone by different names over time, including “strong AI” or “general AI,” but the most common term is currently artificial general intelligence (AGI).
“For a long time, AGI was a distant north star for AI research,” Rokteshel said. “However, with the advent of base models [another term for LLMs], we now have AI that can pass a wide range of university entrance exams and compete in international math and programming competitions.”
This makes people take the possibility of AGI more seriously, Rokteshel added. And most importantly, once we create AI that is comparable to humans in a variety of tasks, it may not be long before it achieves superhuman abilities in all areas. That’s the idea, at least. “Once AI reaches human levels of ability, we can use it to improve itself,” Rokteshel explained. “I personally believe that if we can achieve AGI, we will soon achieve ASI, maybe in a few years.”
Sourse: www.livescience.com
