Want to Get Along With Your Robot Co-Worker? You'll Have to Take Orders From It

If a robot doesn't take your job sometime in the future, there's a decent chance you'll at least have one as a co-worker—be it a tireless delivery bot traversing a massive warehouse or a piece of artificial intelligence writing a news story. How will we all get along? That's a question that researchers are busy trying to answer.

Earlier this year, a team of Canadian researchers found that humans may not only be willing to obey commands given by a robot, but that they may in fact be more likely to respond to the robot than a human, at least for certain tasks. Now, a team at MIT's Computer Science and Artificial Intelligence Lab (or CSAIL) has published a study offering some more insight on how humans and robots may best work together in a manufacturing environment.

The researchers went into the study with the hypothesis that a middle ground—one where humans and robots each had a hand in the decision-making process—would lead to the best working relationship. But their experiment, which involved a human-robot team assembling Lego models together, didn't exactly bear that out.

Rather than preferring to tell the robot what to do, the researchers found that the human subjects instead put a premium on "team fluency," and that they were willing to let the robot allocate all of the tasks if it made the team more efficient.

They came to that conclusion in part through a post-experiment questionnaire that delved into issues like whether the human subjects felt the robot was "trustworthy" and "committed to the task," and even whether they "believe the robot likes me" or not. While one subject responded that there was "something soul-sucking about taking the thinking away from the workers," others said that they never felt like they were "wasting time or waiting" when the robot scheduled tasks, and that it removed the possibility of scheduling being "influence[d] by the ego of the team leader."

As with the earlier study, though, there are still plenty of unanswered questions, and some limitations to the experiment. The researchers note that their sample subjects were all "college students and young professionals whose livelihoods are not threatened by the possibility of robots replacing them," and that each participant worked for a relatively short amount of time (90 minutes).

In the paper, they go on to explain that in a real-world scenario, "manufacturing workers would be working with robots every workday, possibly for years," and that "human workers may have strong preferences for some jobs over others, and may make different choices or have different preferences in task allocation when working with robots every day for the long term."

Speaking with Motherboard, lead researcher Matthew Gombolay explains that "our performance deteriorates if we are under or over-tasked," and "one could imagine that people may want to rely more or less on a robotic or human teammate based on how labor-intensive their job is."

The type of robot involved could be another factor. Gombolay notes that "research has shown that people view robots that are more human-like as more trustworthy, predictable, credible, and intelligent." That, he says, could ultimately play out in two different ways.

"People could inappropriately rely on a robot that is not capable of adequately performing a task," he says. Or, "people could also dismiss a system that could be incredibly beneficial if it does not appear intelligent."

That makes it all the more important that "roboticists develop robots with the right level of anthropomorphism to reflect the robots' true capabilities," he says.

"If a robot is quite human-like but is not very capable, this could cause people to inappropriately place too much trust in the system."

As Gombolay told MIT News, however, he doesn't see the research as a question of simply putting robots in charge, but "about developing tools to help create plans automatically." He also sees those tools as having applications well beyond human-robot manufacturing teams, and points to everything from search-and-rescue drones to tasks like scheduling hospital resources for all-human teams.