Humans are "close to being outsmarted" by superintelligence, AI expert says
Updated on: September 9, 2026 / 5:47 PM EDT / CBS News
TL;DR
- Former AI researchers, including Jacob Coxon and Evan Hubinger, have warned that superintelligence could emerge soon and pose an existential threat.
- Superintelligence is defined as AI that can significantly outperform humans on essentially all cognitive tasks.
- Concerns include the concentration of power in a few companies, misuse of AI, and the potential for AI to evade human control.
- Recursive self-improvement (RSI), where AI builds smarter versions of itself, is a major factor accelerating development and increasing risks.
- Estimates for the emergence of superintelligence vary, with some suggesting within 1 to 10 years, while others believe it is many years away.
- Despite potential upsides like scientific breakthroughs, the current focus on multi-agent systems may not align with beneficial applications for humanity.
- Lawmakers are also expressing concern, with plans to introduce legislation like the Ban Artificial Superintelligence Act.