Teamwork, in the future, means biological brains collaborating with silicon ones. Humans are being paired with computer intelligence to carry out more and more tasks: landing airplanes, leading military reconnaissance, and exploring hostile terrain for humans. More emerge each year.
But we need to find better ways to work together. As machine intelligence approaches human smarts, collaboration is breaking down. The problems is that humans choke on multiple streams of information that computers absorb in milliseconds. Multitasking leads to “high stress and performance degradation” in humans, say researchers. In other words, things fall apart.
Presenting at the Computer Human Interaction Conference (PDF) in Austin, Texas, this month, a team from MIT has offered a way to ease the strain of such multitasking by building an autopilot for your brain. By using brain imaging to monitor humans for mental overload, computers can gradually increase automated control of a complex situation until the human is perceived to be able to handle it again.
Called Brainput, it detects when a person’s brain shows signs of overload, then takes on work to make things easier. It seamlessly transitions between human and machine control in unpredictable and complex situations, giving more automated control when it senses the person can’t handle the tasks anymore. The systems works by using functional near-infrared spectroscopy, or fNIRS, to track changes in the concentration of blood hemoglobin that signal multitasking overload. It can even see signs of mental overload that are individual to each person.
For now, it’s just a prototype. Experiments have been conducted with subjects asked to simultaneously guide two robots through a maze–something that overwhelms most people. The computer eases the stress by progressively taking over functions until the human can take up the slack. It’s not perfect though. Subjects in the experiment struggled with robots who disobeyed commands and headed off in their own direction as conditions shifted.
But autopilots can be dangerous things. Mazes are one thing, but the crash of the Air France flight AF447 from Rio de Janeiro, Brazil has been blamed partially on pilots’ inability to reconcile their autopilot’s reaction with their own senses, leading to the death of 228 people.
Finding the right way for humans and machines to work together can be more than a matter of multitasking.