Elon Musk has long been warning of the dangers posed by uncontrolled Artificial Intelligence, most recently in a tweet challenging Facebook CEO Mark Zuckerberg’s understanding of the treat AI poses. Ironically, AI research being done at Facebook may have just reaffirmed Musk’s concerns.
CREDIT: Getty Images
The Golden Egg That Won The War
World War II was about deception. Communications by each side were encrypted to keep the other side in the dark about strategy and tactics. Were it not for the famous British mathematician Alan Turing, credited with building the computer that broke the German’s Enigma code, the outcome of the war may have been completely different. In fact Churchill called Turing and his team “the geese that laid the golden eggs” citing their work as the single biggest contribution to Allied victory
Turning established the most basic architecture of computing, the Turing Machine, which has defined how computers worked for nearly a century; namely, that computers are programmed to follow a predetermined set of rules, or code. The computers don’t come up with the rules, they just execute them much faster.
If a computer made a mistake it was due to a human incorrectly coding the computer’s response. It was all very linear and predictable. Think of the classic of “if this, then that” logic that every set of interactions is built on.
“AI can evolve in ways that are invisible to us.”
But that may have just changed in a frighteningly dramatic way that recently came to light in AI work being done by Facebook.
But before we get into what that was, a little context.
A War of Words
Over the weekend Elon Must and Mark Zuckerberg engaged in a bit of digital sparring. In a Facebook Live broadcast Zuckerberg said about AI “I have pretty strong opinions on this. I am optimistic,” Zuckerberg said. “And I think people who are naysayers and try to drum up these doomsday scenarios — I just, I don’t understand it. It’s really negative and in some ways I actually think it is pretty irresponsible.” Musk later responded, viaTwitter about Zuckerberg’s AI knowledge, “His understanding of the subject is limited.”
Musk along with many other well known figures in tech and science, such as Stephen Hawking and Bill Gates, have been very vocal about their concern that AI could quickly evolve to pose the single biggest threat humanity has ever faced.
“To call the AI bots deceptive, or to ascribe human intent to the bots, may certainly be stretching it, but at the very least they were finding tools to deal with their environment that no human could comprehend.”
Part of that threat is based on AI’s ability to quickly learn and evolve from simply observing patterns in machines, the environment, and even our behavior. Just look at how Google’s DeepMind AI, AlphaGo, recently beat the world’s best Go player. Unlike chess, which has a large but relatively finite set of moves, Go has more potential moves than there are atoms in the known universe.
Winning at Go is about more than brute force and rigid machine logic. It’s also about some degree of intelligence that involves gut and intuition. While those are qualities we attribute only to humans, it may well turn out that a sufficiently advanced AI can exhibit behaviors that at least look like they are based on more than just a given set of rules and at worst (or is that “at best?”) actually exhibit all the hallmarks of human intuition.
What this means is that AI can evolve in ways that are invisible to us. In other words we don’t necessarily have any idea why a program or a device powered by AI is doing what it’s doing because it’s decisions and actions go far beyond what it was told to do.
That’s exactly what Facebook’s bots did in a way that so spooked it’s engineers that they shut the bots down.
Can You Hear Me Now?
The bots, which were used to conduct dialogs in a chat with humans, were trained to communicate with humans and with each other in English. Yes, the bots actually talk to each other to collaborate in solving problems faster by creating new rules.
However, in a turn of events nobody had anticipated, the bots suddenly started to communicate with each other in their own language–a language that they made up on the fly and entirely on their own.
The science fiction-like scenarios this conjures up are right out of an episode of Star Trek, but this wasn’t a Hollywood script that was being played out, It was a real world example of AI evolving faster than its human creators could keep pace with.
To call the AI bots deceptive, or to ascribe human intent to the bots, may certainly be stretching it, but at the very least they were finding tools to deal with their environment that no human could comprehend.
It’s ironic that all of this happened so closely on the heals of the well publicized debate between Musk and Zuckerberg about the risks of AI. But it does shed light on how quickly AI can extend beyond our reach to control or comprehend.
It’s also poetic to consider that while Turing may be the father of modern computing, he is also known for something that may prove to be much more prescient, his quote, “A computer would deserve to be called intelligent if it could deceive a human into believing that it was human.”
Were we just deceived?
The opinions expressed here by Inc.com columnists are their own, not those of Inc.com.