In the Loop

Sara Kiesler is the university's Hillman Professor of Computer Science and Human-Computer Interaction. She came to Carnegie Mellon in 1979 as a professor in the Department of Social and Decision Sciences and conducted some of the first scientific studies of computer-mediated communication and the Internet.

A graduate of Boston's Simmons College, she earned masters' degrees in psychology and communication at stanford and a doctorate in psychology at Ohio State.

Kiesler spoke to Link Managing Editor Jason Togyer about her current research, which includes the project on people and robots, an interdisciplinary initiative funded by the National Science Foundation, Microsoft and Pittsburgh's Hillman Foundation.

One offshoot of the project on people and robots has been Snackbot, which debuted at the dedication ceremonies for the Gates and Hillman centers.

Why a "snackbot"? Why not some other kind of robot?

Jodi Forlizzi and Paul Rybski wanted a robot that would deliver aservice, so we decided to work on the idea that Snackbot would dispensesnacks and encourage healthy snacking. We didn't have the resources tobuild a real humanoid robot--Snackbot doesn't even have "arms"--but itgives the impression of one. The head, the mouth that smiles--we've beenstudying the things that make robots seem human-like. One of mycontributions has been research into people's mental models of whatthey think is going on behind the machine.

Do people assume robots are smarter than they really are?

Actually, you can apply some of the same ways that people think about dogs to their mental models of robots. In my spare time, I train dogs, and I'm always amazed when I show people what dogs can and can't do. They come home and there's garbage on the floor and they think, "Oh, my dog was mad at me, because I was late." They think the dog can feel resentment or revenge. But dogs don't understand causal connections in the same way we do. In the same way people impute characteristics to their dogs, people impute characteristics to robots.

Does that include managing people's expectations?

Yes, their expectations are very high, even with Snackbot. Snackbot isactually very, very simple, but if a robot can speak, people imagine itcan think. If you tell people a robot was built in China, for instance,they assume it knows about the Great Wall of China. Theyovergeneralize.

People make these assumptions based on very small amounts ofinformation. The robot itself is like an inkblot--they imputecharacteristics to the robot. This can create feelings ofdisappointment and confusion when the robot doesn't behave the way theythink it should. Even some robotics students do it, which is amazing tome.

Are people reluctant to interact with a robot?

Not necessarily, and in fact we can learn about people from the way that they interact with robots. One of my former students, Cristen Torrey, studied robots designed to help people. Sometimes people don't want to be helped by a robot because it seems demeaning. It turns out that people are more accepting of a robot's help if it hesitates a little bit and is less directive. It seems strange for a machine to act that way, but it might work when we're designing robots in the future that are designed to act as caretakers.

We're learning that we can do a better job building those kinds of robots, even with very minimal designs.

Your other research has been into online collaboration--is "crowd-sourcing" a model for distributing other kinds of work besides software?

Software collaboration works better than other types of online collaboration because it can be modularized. Otherwise, distributed research doesn't necessarily work very well. We recently did a study for the National Science Foundation to examine the outcomes of a program they had in distributed research, and we found that the more universities that were involved in a project, the less likely it was that the project was productive.
 
One of the stumbling blocks was coordinating meetings. In a distributed project, we might have to make appointments to meet--we can't just depend on running into each other in the hallway. Another problem is that people don't have situational awareness of what their colleagues are working on, or what problems they're having. When people lose touch with colleagues, they start working on local problems and the research is more apt to become fragmented and uncoordinated.

Does that mean the Internet is an impediment to collaboration?

No, the Internet has made it possible to form distributed collaborations. There are online communities of researchers that are very strong, and wouldn't exist if they couldn't be online. But you can't just impose the old organizational structures.

That's why I want to understand the ways technology and changes in the way we do work can support collaboration. What kinds of newer technologies are there to support organizations--and what's going to have to be done differently when we're forming organizations in the future?

There are people at other places who design wonderful technologies and then find uses for them--that's not for me. I still believe in studying people and organizations first, and then designing systems to support them.
For More Information: 

Jason Togyer | 412-268-8721 | jt3y@cs.cmu.edu