Manuela Veloso still gets surprised when the robot shows up outside her office.

"I can hear his motor coming down my corridor," says Veloso, the head of the machine learning department at Carnegie Mellon's School of Computer Science, "and when it stops right there in front of my door by itself, I'm always in awe."

The department uses so-called co-bots, short for collaborative robots, for mundane tasks like distributing mail or delivering messages. They can greet visitors and guide them to their host's office or fetch an item and deliver it to a researcher down in the laboratory. 

There are no humans guiding them--the bots use artificial intelligence and operate mostly autonomously. They were developed by a CMU team led by Joydeep Biswas and first deployed in 2013. The department is using them as an experiment of sorts as it studies the best approach to rolling out robots to work alongside humans.

It's a question that's becoming increasingly relevant. As bots become smarter, safer, and more nimble, they're moving from strictly behind-the-scenes manual labor, like car manufacturing, to working right next to their human counterparts. 

At CMU, the co-bots are starting to fit in. This Halloween, when Veloso and her colleagues realized the bot cruising around on their floor wasn't in costume, they dressed it up as a ghost. 

Manufactured personality.

The Carnegie Mellon team still views the co-bots as a research project. The machines are very good at what they do. "They never once have missed my door," Veloso says. But they're still limited as far as interactions go. A robot will understand commands, but if a person says hello or asks how it's doing, it won't respond. And if it needs help pushing an elevator button, it won't direct its request at a person, but instead will shout it mindlessly into the air.

Creating bots that can engage seamlessly with humans is challenging, but it's a challenge that might be worth it if it makes those humans actually want to work with them. That's why teams around the country are working on creating personalities for their robots.

"If you want to get people to work with technology," says Lionel Robert, a robotics professor at the University of Michigan, "you're going to want to make it as similar to them as possible. You need to find ways for people to have an emotional bond with it." For robots, this can include physical features, like a humanoid body shape, as well as the ability to understand spoken language and to speak it out loud. "If people believe a robot has their personality," Robert says, "they're more inclined to trust the robot, and therefore more inclined to work with the robot."

Even the subtlest actions can make people feel like they're working with a collaborator instead of a machine. Elizabeth Croft, director of the University of British Columbia's Collaborative Advanced Robotics and Intelligent Systems Lab, researches the ways that robots' gestures affect how people interact with them. Croft's team experimented by programming robots to hand objects to people while looking at either the object, the ground, or the person's face. In the instances where the bot looked at the object, like a person often would, the handoffs happened quicker and more smoothly.

Separately, Croft's team tried having robots behave in different ways when reaching for an object at the same time as a person. In some cases, the bot would either continue grabbing as if the person wasn't there, and in others, it would freeze so as not to impede its human peer. But in a third condition, Croft's team set the robot to hesitate, pulling the awkward will-I-or-won't-I, back-and-forth arm gesture that many humans would make in that situation. In those trials, humans rated the robots as being the best to work with on post-trial questionnaires. 

"The gesture is saying, 'I recognize you're there. I recognize the value of your task as well as the value of mine,'" says Croft. "We actually had some people afterward say things like, 'The robot was polite,' or 'The robot was shy.' If we can give people a mental model of the behavior of the robot, we can increase the likability and comfort people feel in working with them. And that actually leads to less errors and more efficient work by the humans."  

Doing the dirty work.

Experts generally agree that on a long enough timeline, machines will be capable of doing just about anything humans can do today. But that's not to say they will. "I don't think we want robot teachers, or robot cops," says Ben Pring, director of Cognizant's Center for the Future of Work and co-author of "What to Do When Machines Do Everything."

"What's more likely to happen," he says, "is that most jobs are going to be infused with AI, to take them to the next level." Tasks characterized by the "three Ds,"--dirty, dull, or dangerous--are the ones most likely to be handed over to robots. "The machine is going to take some elements of work away, absolutely," he says. "But it's also going to create new opportunities around it."

At Amazon fulfillment centers where bots help humans pack orders, the company has been able to increase its inventory by 50 percent thanks to more efficient use of floor space. Some workers recently told the The New York Times their jobs have become more interesting, shifting from heavy lifting to more engaging machine supervisory roles.

In addition to the dirty jobs, some customer-facing professions could look different in the not-too-distant future. At the University of California, San Diego, researchers are working on creating machines that can perform low-level but useful tasks in health facilities.

"Clinicians are interrupted constantly in hospital emergency departments, which is shown to lead to patient harm," says Laurel Riek, a professor in the university's department of computer science and engineering. Brief conversations, like asking a colleague to restock supplies, can cause a several-second distraction.

Riek's lab has multiple research projects underway to explore how robots can understand and anticipate the needs of both doctors and patients in high-pressure situations. "This enables clinicians to focus on the work they are best at," she says.

Designing these bots to best perform their tasks alongside people requires a special ingredient: The very people with which they will someday be working. Riek and her team are including clinicians and patients with disabilities in the design process from start to finish. "This helps ensure a close match between the stakeholders' expectations and what the robot is able to deliver," she says. It helps us build robust, safe algorithms for our robots so they'll be able to function in the real world long-term."

And once those bots can perform their tasks as reliably as people can, some believe that human acceptance of them is inevitable.

"Smiles and a friendly appearance--that would be nice," says Veloso, the Carnegie Mellon professor. "But before I worry about their appearance, I need them to be intelligent. I am very based on function. I believe that they will conquer people by doing great things for people."