Mathematician Tsvi Benson-Tilsen once worked at the Peter Thiel–funded Machine Intelligence Research Institute, where he was one of many experts tasked with figuring out how to ensure AI doesn’t eventually destroy humankind. After seven years, he concluded that he’s not smart enough to figure it out. As of today, he doesn’t think anybody is.
The 33-year-old is racing against a threat known as “the singularity,” the moment when superintelligent machines, having surpassed the feeble cognitive abilities of humans, begin to act in ways contrary to the interests of humanity. “If it’s smarter than you, you cannot tell what’s dangerous necessarily, and you cannot tell what it’s thinking, because it could hide its thoughts,” Benson-Tilsen explains.
Even the sector’s leading thinkers don’t really comprehend how their systems work and thus cannot guarantee their models won’t try to deceive, overthrow, or even kill us.
Share this article
In Benson-Tilsen’s ideal tomorrow, … some of the wunderkinds optimized for superior intelligence will have quashed the threat of advanced AI, and the technology needed to have healthier and smarter babies will be widely accessible and affordable.
