We are now at a crossroads of artificial intelligence: We can design computers that sharpen our wits or we can let our machines turn us into ignoramuses.
That observation capped a provocative panel on Tuesday at Austin’s South By Southwest conference.
Increasingly intelligent machines — search engines that yield knowledge on demand, smartphones that understand plain English, computers that proffer medical diagnoses, ad tech that offers to sell you just what you’re looking for — represent a “tipping point,” said Doug Lenat, a former Stanford and Carnegie Mellon computer science professor who is CEO of Cycorp, a maker of machine reasoning software. “We could become smarter or dumber – much smarter or much dumber.”
Electronic calculators, Lenat argued during the panel entitled “AI State of the Union,” have created generations of students who can perform mathematical tasks very quickly but don’t understand the underlying concepts. Similarly, Google “swaddles” users in a blanket of instant information, relieving them of the burden of independent thought and inquiry. The next wave of artificial intelligence — loosely defined as a computer’s ability to distinguish between useful and useless information at any given moment — could propel us irrevocably down that path.
“This could lead to Idiocracy,” Lenat said, referring to the 2006 Hollywood satire about a future in which human intellect has taken a steep dive. The result would be a society “where no one has to understand anything about the world, where everything just seems like magic.”
Alternatively, he said, computer scientists could design artificial intelligence “to challenge us the way Aristotle challenged Alexander the Great, to make us smarter, more rational, more human, to understand the world more deeply.”
AI lately has taken a turn from digital smarts based on rules (furry faces with whiskers and triangular ears are cats) toward those based on recognizing patterns (some online pictures have certain similarities, and the word “cat” in the caption may mean they’re cats), the panelists agreed. The latest advance, known as deep learning, is based on decades-old technology that became practical only in recent years, as greater processing power has enabled computers to achieve higher orders of learning and vast accumulations of data on the Internet have given them enough training to draw relevant conclusions.
The shift could be viewed as a transition from left-brain thinking based on logic to right-brain thinking based on intuition. Both modes will be necessary to further advances in AI, yet the brain devotes far more time to intuitive than rational thinking, pointed out Monica Anderson, CTO of Sensai and a specialist in artificial intuition.
“Reasoning is the sugar on top of intuitive understanding,” Anderson said.
The key challenge will be getting computers to recognize context, the panelists said. Humans have little problem recognizing biases introduced by politics or limitations due to outdated information, but machines are blind to contexts they haven’t been specifically programmed to recognize.
One answer is to design AI that asks clarifying questions, said Fred Brown, CEO of Next IT Corp, which builds virtual assistant technology. “Geeks want machines that do things without asking,” he said. “But what makes good AI is good communication with the user.”