Artificial intelligence gets acquainted with human social standards

Marion “Tank” LaFleur, the receptionist at Newell-Simon Hall, greets visitors, jokes around… Marion “Tank” LaFleur, the receptionist at Newell-Simon Hall, greets visitors, jokes around with them and watches their movements with unnerving concentration. Tank, as he prefers to be called, is not a typical receptionist, but a “Roboceptionist” at Carnegie Mellon University.

Tank sports a digital screen for a head, displaying animated facial expressions on top of a body akin to “Star Wars” android R2-D2. He responds to requests for staff directories and information on local eateries and even jokes threateningly to tell people to use his nickname instead of “Marion.”.

He’s also quick to remind visitors, “The next time your computer doesn’t work, don’t hit it. We have feelings too!” It’s a reminder that Tank has a more anthropomorphic  personality than we’re used to from the artificial intelligence we interact with daily. Pandora predicts what you’d like to hear next, the Roomba navigates your living room to vacuum — but Tank can socialize with you.

Tank represents a new frontier for AI — social artificial intelligence.

Ph.D. student Heather Knight, who works with Reid Simmons, Tank’s progenitor, at CMU, envisions a future where machines are capable of interacting with humans on a social level.

“We want to add something extra, add entertainment. We want to make not just an operating system with utility, but one that makes you feel good, that makes life better,” she said.

And though we don’t all have attentive robots like Tank — nicknamed the “Creeperbot” by those who find his ever-vigilant eyes unsettling — we might still interact with social artificial intelligence that’s sitting in our pockets. Siri is a personal-assistant application launched as an integrated part of iOS 5 for the iPhone 4S in October. Though its voice still rings with a certain technological monotone, Siri uses natural language to answer questions, make recommendations and perform tasks. It adapts to individual preferences over time and personalizes its services.

“It works amazingly well. In the future, I hope we see proactive systems that can support our needs but that can also be sensitive to the social setting,” associate professor of human-computer interaction at CMU Jodi Forlizzi said.

Knight said that successful human encounters involve social etiquette that comes second-nature to us, like knowing which side of a hallway to walk down or the way we use nonverbal signals for tasks like figuring out who goes through a doorway first. This fluid interpersonal socialization makes the replication of human social intelligence in robots difficult.

“In building these systems, you think a lot about what [it means] to be human and how we can make machines more useful. How do we communicate with each other in the first place?” Knight said.

Knight said one of her robots has been “dabbling in comedy.” Data — named for a character from “Star Trek” — can aggregate the sounds of laughter, applause and booing and then adjust his joke-telling routine accordingly. Last December, during Data’s debut at a TED talk — a Technology Entertainment and Design mini-presentation that can be found on www.ted.org — Knight asked audience members to hold up green panels after jokes they liked and red panels after jokes they didn’t. This allowed Data to interpret the visual information and choose his next joke from his internal database. Knight said that, over time, Data can adjust to different audiences’ tastes, similar to the way Netflix adjusts its viewer recommendations.

“Comedy acts have the opportunity to be more dynamic than scripted plays and give the robot the opportunity to change direction based on how people are responding,” she said.

This provides a litmus test of how well the application can gauge social cues. Knight believes that finding ways for people and robots to interact successfully will lead to smoother interfaces. She views artificial intelligence research as “empowering humanity through machines. I think about how [we can] make people enjoy life more or be able to prosper, how we can create caretakers or tools that are more fun to use.”

These humanoid robots are easier for people to identify with and consequently work with more effectively — up to a point. When robots start to resemble people too much, researchers have found a sharp dip — known as the uncanny valley — in how much people like them.

Classic examples of technology creeping toward the uncanny valley are the android robots built by Japanese researcher Hiroshi Ishiguro. Forlizzi described the robots, complete with facial movements and simulated breathing, as “frighteningly real humanoid systems.” Forlizzi said that in one test, where participants were briefly exposed to Ishiguro next to the robot replica he built of himself, participants were unable to tell them apart.

This means that researchers must gauge whether or not a human user will feel comfortable with the robots to ensure that they’re helpful. Knight said that the goal of her research was to enable humans and machines to do things together rather than to displace humanity. She gave the example of robots assisting the elderly to live at home and maintain their independence longer, which would enhance their personal freedom and quality of life. But Forlizzi stresses that this technology won’t replace humans.

“I do not think a robot could ever replace a human caregiver — only augment what they do so human assistance can be available for more people,” Forlizzi said.

Another way artificial-intelligence-enhanced tools can supplement human work is as classroom aids. Janet Schofield researches the effects of computers on classroom social processes.

“As AI becomes more advanced, I expect it will be able to contribute more and more to instruction, especially in certain domains where ‘right’ and ‘wrong’ answers are clear and the common paths to right and wrong answers are understood,” the Pitt psychology professor said in an email.

Schofield said that a geometry-tutoring program she studied in high school classrooms increased students’ motivation and promoted friendly competition between students. It also allowed students to progress at their own pace to a greater extent than traditional instructional methods. The program, which adapts its content to its users, frees up teachers’ time to focus on other educational areas.

It’s these sorts of programs that artificial intelligence enables. We’re already adapting to Siri’s sweetly human-like responses to our requests for help, and it remains to be seen what technology will integrate itself into our lives next.