AI’s big mistake: trying to imitate humans

Many discussions about future artificial intelligence products invoke comparisons to human intelligence. That’s a fundamental mistake, according to All Turtles CEO Phil Libin.

In the video above, Libin explains that the Turing Test went wrong by establishing the ability to imitate a human as a benchmark. “It’s ridiculously hard to do,“ he says, “and it’s completely unnecessary.”

Libin urges the AI community to think less about sentient robots, and more about service animals. “Think of [AI] as a really smart sheepdog, or a guide dog, or a bomb-sniffing dog. A really good sheepdog is ridiculously better than a human can be at its job, which is herding sheep,” he explains, offering it as an example of how AI can be developed to solve specific tasks.

The sheepdog “has a supernatural understanding of sheep, what motivates them, which way they’re going to go. It’s amazing and it’s got a real personality,” Libin says. “They can be cute, they can be approachable, you can form an actual bond to it, it has a nice user experience, and it’s brilliant at its job — much better than a person is. But it’s not smart like a human.”

Like the best AIs, sheepdogs are simply really good at their jobs.

 

No one is asking when will we have fully-sentient artificially intelligent robot sheep dogs. Credit: SheltieBoy/Flickr

The above video is part of a panel discussion hosted by IDEO, Silicon Valley Bank, and All Turtles entitled “Artificial Intelligence & Trust” held on September 19, 2017. Vivek Wadhwa, Distinguished Fellow at Carnegie Mellon University College of Engineering, moderated the panel. Phil Libin, Cofounder/CEO of All Turtles, Jana Eggers, CEO of Nara Logics, and Muddu Sudhakar, an entrepreneur and executive, served as panelists.