When Anything Can Talk to Anything, What Could Possibly Go Wrong?

An audience of 60 students, faculty and business leaders in the Chicago area joined Marconi Chairman Vint Cerf and 2011 Young Scholar Joseph Kakande at Illinois Institute of Technology to discuss the Internet – past, present and future.  The topics included whether broadband capacity can keep up with demand, problems and challenges inherent in Internet of Things (IoT), machine learning/AI, and the key challenges for the future.

A particular focus of the discussion was the problems posed by the IoT. Cerf warned that a typical consumer could soon have hundreds of Internet-connected devices running autonomously on sloppy or untested software, none of which was subject to uniform standards.  “The problems will only get worse,” he says.  “If the company that made a particular device goes bankrupt or out of business, who will support the device? And with the prevailing lax or non-existent security, how can we prevent devices from being hijacked and harnessed—like the web cams recently—for denial of service attacks?  Developers need to think hard about these problems and approach this with a strong sense of moral responsibility.”

Cerf suggested that TLA+ (described as exhaustively-testable pseudocode and blueprints for software systems) can potentially improve and eliminate some of the problems as software is being designed for devices. While it couldn’t write software, it could provide feedback to developers about their assertions.

That prompted an audience member to ask whether there is a role for machine learning in testing code—an idea Cerf found intriguing.  However, he pointed out the complexity of machine learning algorithms. “If you open up a neural network and adjust just one of hundreds of switches, you literally have no idea what the result might be,” he said.  “The plus side is that neural network learning can be transferred among many devices.”

Audience member and 2008 Marconi Fellow Sir David Payne provided an anecdote to show just how difficult creating software to run devices autonomously can be. He told the story of a room service robot encountered during a hotel stay.  The robot was supposed to deliver a hamburger to the friend, but when it arrived, it began to shuffle around outside, without ringing the bell.  His friend heard the noise and opened the door, whereupon the robot rolled inside, seemed to stop almost in panic, and exited backwards without delivering the food.  It then returned to the elevator where it exhibited more confused behavior by pinning Sir David against the wall before exiting again.  The hotel had to put out an all-points bulletin on the rogue robot, because no one knew where it had gone. David and his friend eventually surmised that his friend had accidentally hit the “Do Not Disturb” button for his room, so that when the robot arrived it had countermanding orders—which sent the poor machine into a tailspin.

Cerf agreed that this is a significant problem, because “software does what you told it to do, not necessarily what you want it to do.”  Thus, problems arise when cases come up that the software isn’t trained to address. “People who write software for autonomous devices should feel a huge responsibility to think things through,” he admonished.

Regarding the IoT, he predicted that in the not so distant future “anything will be able to talk to anything else”—which is not necessarily a good idea.  “What we need is to empower both parties to say, “I don’t want to talk until you’ve convinced me there’s a good reason,” he said.

The lunch and chat with Cerf and Kakande was part of an outreach program to raise awareness of the Paul Baran Young Scholar Awards, for which the Society is currently accepting nominations.