Ellie (right) is a computer simulation designed to engage real people, like the woman on the left, in meaningful conversation and take their measure. The computer system looks for subtle patterns in body language and vocal inflections that might be clues to underlying depression or other emotional distress.
Ellie (right) is a computer simulation designed to engage real people, like the woman on the left, in meaningful conversation and take their measure. The computer system looks for subtle patterns in body language and vocal inflections that might be clues to underlying depression or other emotional distress. YouTube
Her hair is brown and tied back into a professional-looking ponytail. She wears a blue shirt, tan sweater and delicate gold chain. It's the first time she has met the man sitting across from her, and she looks out at him, her eyes curious.
"So how are you doing today?" she asks cautiously, trying to build rapport.
"I'm doing well," he answers. His eyes blink.
"That's good," she continues. "Where are you from originally?"
"I'm from L.A.," he tells her, and this makes her smile slightly.
"Oh!" she says with surprise in her voice. "I'm from L.A. myself!"
She is from L.A. She was created in Los Angeles and "lives out her life" there on a computer screen in a lab at the University of Southern California. She's not a real woman but a virtual one, created to talk to people who are struggling emotionally, and to take their measure in a way no human can. Her makers believe that her ability to do this will ultimately revolutionize the way mental health care is practiced in this country. Her name is Ellie.
There's Power In A Well-Timed 'Uh-Huh'
The project that resulted in Ellie began almost two years ago at USC's Institute for Creative Technologies. Two scientists in particular are responsible for her existence: psychologist Albert "Skip" Rizzo and computer scientist Louis-Philippe Morency.
Rizzo and Morency spent months laboring over every element of Ellie's presentation and interaction with patients, experimenting with a range of different personalities, outfits and vocal mannerisms.
"Everything has been thought of," says Morency. For example, when patients talk, Ellie encourages them to continue talking with a well timed "uh-huh," just as real people do.
"We have recorded more than 200 of these uh-huhs," Morency says, "and these are so powerful. Because a simple 'uh-huh' and a silence — if they are done the right way — can be extremely powerful. So we spent a lot of time on these little details."
But the most important thing about Ellie is not her skill at gently probing all of the people her scientist brings into the lab to talk to her. Her real value, the reason she was built at all, is her skill at taking and analyzing thousands of measurements of those people.
Under the wide screen where Ellie's image sits, there are three devices. A video camera tracks facial expressions of the person sitting opposite. A movement sensor — Microsoft Kinect — tracks the person's gestures, fidgeting and other movements. A microphone records every inflection and tone in his or her voice. The point, Rizzo explains, is to analyze in almost microscopic detail the way people talk and move — to read their body language.
"We can look at the position of the head, the eye gaze," Rizzo says. Does the head tilt? Does it lean forward? Is it static and fixed?" In fact, Ellie tracks and analyzes around 60 different features — various body and facial movements, and different aspects of the voice.
The theory of all this is that a detailed analysis of those movements and vocal features can give us new insights into people who are struggling with emotional issues. The body, face and voice express things that words sometimes obscure.
"You know, people are in a constant state of impression management," Rizzo says. "They've got their true self and the self that they want to project to the world. And we know that the body displays things that sometimes people try to keep contained."
So, as Ellie gets the person in front of her to ruminate about when they were happy and when they were sad, the machines below her screen take measurements, cataloging how much the person smiles and for how long, how often they touch their head.
Morency says the machines record 30 measurements per second, or "about 1,800 measurements per minute." Literally every wince, pause and verbal stumble is captured and later analyzed.
Ellie was originally commissioned by the U.S. Department of Defense. After all of the deployment in Iraq and Afghanistan, the military was seeing a lot of suicides and wanted to find a way to help military therapists stop them. Soldiers don't always like to confess that they're having problems, but maybe their bodies would say what their words wouldn't.
This is why Ellie is being programmed to produce a report after each of her sessions — it's a kind of visual representation of the 60 different movements she tracks.
"For each indicator," Morency explains, "we will display three things." First, the report will show the physical behavior of the person Ellie just interviewed, tallying how many times he or she smiled, for instance, and for how long. Then the report will show how much depressed people typically smile, and finally how much healthy people typically smile. Essentially it's a visualization of the person's behavior compared with a population of depressed and nondepressed people.
If the person's physical behaviors are similar to someone who's depressed, then the person will be flagged.
The idea here is not for Ellie to actually diagnose people and replace trained therapists. She's just there to offer insight to therapists, Morency says, by providing some objective measurements.
"Think about it as a blood sample," he says. "You send a blood sample to the lab and you get the result. The [people] doing the diagnosis [are] still the clinicians, but they use these objective measures to make the diagnosis."
Real People Are Complicated
Now, obviously this work raises all kinds of issues, and even on a practical level, real obstacles remain. Jeff Cohn, a psychologist at the University of Pittsburgh, studies the relationship between physical movements and emotion and says signals from the face, voice and body are incredibly complicated to interpret.
"Individuals vary a lot in how expressive they are," Cohn explains. "You know, if I'm someone who is very expressive and I smile frequently, [even] when I'm depressed and smiling less, I may still smile more than you do if you're a tight-lipped, not very emotive individual."
This means, Cohn says, that using Ellie in the way blood tests are used — as proof positive of one diagnosis or another — will be really difficult.
"It strikes me as unlikely that face or voice will provide that information with such certainty," he says.
But Skip Rizzo, the psychologist working on Ellie, genuinely believes these technologies will eventually change the field of mental health. One of the central problems with humans, he says, is that they bring their own biases to whatever they encounter, and those biases often make it hard for them to see what's directly in front of them.
"You can get training to be a health care provider or psychologist," he says, "and try to put those things on hold and be very objective. But it's still a challenge. It's always going to be biased by experience. What computers [like Ellie] offer is the ability to look at massive amounts of data and begin to look at patterns, and that, I think, far outstrips the mere mortal brain."
This summer, Ellie is being tested. She's scheduled to sit down with dozens of veterans from Iraq and Afghanistan.
She'll ask them about their lives, encourage them to open up.
Then, silently, Ellie will measure their answers.