University Of Washington Developing Artificial Intelligence Caretakers For Alzheimer's Sufferers

"As my father lost the ability to do things for himself, my mother would give him gentle prompts to keep him on track," recalled Kautz, associate professor in the University of Washington's Department of Computer Science & Engineering. "So at a stage of the disease where, according to the clinical scales, it would seem he couldn't do anything for himself, he could still perform many of the functions of life. He could shower, get dressed, and so forth because my mother would monitor him and give a prompt when needed."

It's a recollection that has guided Kautz in initiating a research effort at the UW to explore ways in which computer science can compensate for diminished mental capacity. The Assisted Cognition Project is a collaborative effort by the UW, Intel Computers and Elite Care, a private company developing a state-of-the-art retirement community in the Portland area that utilizes so-called ubiquitous computing to keep tabs on residents' needs. The project partners envision a network of digital devices and wireless sensors – in the walls of homes, in appliances and furniture, even in the clothing we wear. The devices would work together to do essentially what Kautz's mother did for his father: monitor an individual, offer prompts when appropriate and summon help when needed.

"The difference is it lessens the enormous toll this takes on a human caregiver, usually a spouse or other loved one," Kautz said.

As a first step, the UW group, which also includes computer scientists Oren Etzioni, Dieter Fox and Gaetano Borriello, has developed a prototype for a wireless handheld device, dubbed the Activity Compass, which will memorize a patient's daily routine, then offer advice and directions if the user becomes lost or confused. The university is also working with Intel and Elite Care on developing an Adaptive Prompter, an in-home system to keep patients on task in performing daily activities.

Kautz will give a presentation on the effort next week at the annual conference of the American Association for Artificial Intelligence in Edmonton, Alberta. He's also scheduled to speak in the fall at an international meeting on ubiquitous computing in Sweden.

In the meantime, he and his team are chipping away at the complex issues involved in creating an AI caretaker. The stakes, they say, are high.

In the last 50 years, Alzheimer's disease has grown from relative obscurity to become a defining characteristic of industrialized society. In 1950, at the most 200,000 people in the United States suffered from the ailment. The total stands at 4 million today, according to the National Institute on Aging. By 2050, barring a cure, the number of U.S. sufferers is expected to reach 15 million, out of a total of 80 million worldwide.

As the need for assistance increases, AI caretakers could give patients more independence, save family members from emotional burnout and free adult caregivers to return to the workforce.

While the level of computing sophistication needed to build a comprehensive AI caretaker is probably 20 to 30 years away, the time to begin is now, says Kautz, with tools like the Activity Compass.

The compass is a meshing of several technologies – a Palm Pilot, a GPS receiver and a wireless modem. Over time, the device will "learn" a user's typical schedule, monitor for variations then decide whether a prompt is necessary.

For example, Kautz said, imagine you're walking down the street on a weekday afternoon. The Activity Compass senses your location, direction of travel and can tell by your velocity and pattern of movement that you're walking. It also knows that you usually catch a bus that time of day at a nearby bus stop. In addition, the device is patched into real-time information on the bus's exact location (data that is available in Seattle).

"The Activity Compass sees that the bus will be there in five minutes, but at the speed you're walking you won't be there for 10 minutes," Kautz said. "It then decides 'maybe I should annoy you' and says, 'You better walk faster to catch that bus.' Or maybe 'I think you're going to miss that bus – should I suggest an alternative route?'

"By the same token, if it senses that you're lost, it might say, 'Hey, I think you need help.' That gives the person the option of saying, 'No, I know what I'm doing.'"

The device will be designed to learn from interactions with the user and adapt to changing behavior as the disease progresses.

Currently, graduate student Don Patterson carries the Activity Compass with him as he goes about his business, traveling to classes, coffee shops and activities, and downloads the information about once a week for use in future sessions to "train" the device. Kautz hopes to have it programmed by the end of the summer.

Navigating around town is one thing, but being able to function at home is another.

As mental ability slips, Alzheimer's patients often have trouble with household tasks – turning off the water faucet, taking meals off the stovetop so they won't burn, remembering that socks go on before shoes. Patients recall the individual steps in such multiple-step tasks, but have trouble putting them together.

Another project, the Adaptive Prompter, would help keep patients on task.

The UW is working closely with Intel and Elite Care to develop the system, which uses a sensor network embedded in the home of the user. Elite Care's Oatfield Estates is the development site. The facility is a prototype home for 60 residents that senses and records nearly all the activity in and around its campus – movement of residents and staff, turning on and off of lights and appliances, opening and closing of doors. The intent is to feed the data to an AI brain, which would monitor, weigh uncertainty and make good decisions about when to intervene and, for example, tell a resident stalled in the bathroom during her morning ritual to pick up the toothbrush, or another to put his shoe on and tie it, or someone cooking lunch to take the frying pan off the stove.

"Our focus is on the software," Kautz said. "The sensor technology is pretty good, and we have all this great data. Now we need to put it to use. By the end of summer or fall, we'll be taking full-time sensor data from the residence home."

Individual applications, like the Activity Compass and portions of the Adaptive Prompter, could be developed relatively quickly, Kautz believes. But don't look for a perfected, comprehensive cyber nurse tomorrow.

"The question is whether we can we build systems that are smart enough, flexible enough and reliable enough to replace human caregivers," Kautz said. "I'm confident that we can, but it will be an incremental process."