‘IoT for Puppies’ and AI Help Identify Guide Dog Trainees with the Right Stuff
The dog has been called our "first friend," the bond between dogs and people extends to our pre-history, and one of the most revered forms of that bond is the relationship between the visually impaired and the service dog. However, because the skills required of these dogs is so high, training them is not only expensive, it also has a high failure rate. Now AI and IoT-generated machine data are being used to improve service dog training, breeding and the selection of puppies to enter training school.
Guiding Eyes for the Blind of Yorktown Heights, NY, is a non-profit organization that spends $50,000 to train dogs over the course of a two-year curriculum. Since its founding in 1954, the charitable organization – which provides service dogs at no cost – has graduated more than 7,300 guide dogs. While it breeds dogs (95 percent are Labrador Retrievers; 5 percent are German Shepherds) for service work on a selective, data-driven basis, only 37 percent of puppy candidates become full-fledged guide dogs for the blind (another 13 percent go to other organizations for less demanding service work).
The failure rate isn’t surprising given the extraordinarily complex and life-threatening tasks that service dogs must handle, such as guiding the blind across busy urban street intersections. Among the discernment skills these dogs possess is “intelligent disobedience,” the ability of the dog to delay execution of the owner’s command until it’s safe to do so – or to refuse the command altogether. Some guide dogs have gone into legend, including two courageous dogs (one of them, Salty, a Guiding Eyes graduate) that brought their owners to safety from the upper floors of the World Trade Center on 9/11 (advice to those who watch the video: bring your handkerchiefs!).
In the face of these requirements, there is strong demand for service dogs. Someone in the U.S. becomes blind every seven minutes, and as baby boomers face age-related vision loss, and with the increased incidence of diseases such as diabetes, demand is going up.
Guiding Eyes has teamed with researchers at San Jose State University and IBM in what must be one of the most unusual workloads assigned to IBM's Watson supercomputer: using data to improve selection of puppies for service training, with the goal of increasing service dog graduation rates.
“Our goal is to bring down the cost of training and produce more trained dogs,” Jane Russenberger, Guiding Eyes’ director of genetics and breeding, told EnterpriseTech, noting that the organization received 20 percent more requests for service dogs last year than in 2015. She said Guiding Eyes dogs are provided at no charge because many visually impaired individuals are often passed over for employment – even those qualified for particular jobs – and because service dogs are not covered by health insurance policies in the U.S. (unlike some European countries). “We need to use our donor dollars the best we can.”
This year, Guiding Eyes is on course to produce 165 fully trained service dogs; next year, the goal is 170.
Russenberger said Guiding Eyes and its Canine Development Center (CDC) works to identify preferred service dog qualities, such as health, confidence, love and temperament. This has meant the accumulation of dog data, lots of it. Through 16 generations of selective breeding, the CDC has captured both structured data – such as medical records and complex genetic mapping – and unstructured data from questionnaires provided by trainers and host families – called “raisers” – about their impressions and experiences with the dogs.
Guiding Eyes began working with IBM and data scientists at San Jose State as its data volume became unwieldy and difficult to extract meaningful insights. Guiding Eyes has migrated more than half a million health records and more than 65,000 temperament records on thousands of dogs to Watson in the IBM Cloud.
“It’s a ton of data, it takes a supercomputer to handle it, and then Watson can do the AI to help us interpret it to help us make better decisions,” she said. “What we’re looking for are the correlations and the predictive analytics. Watson can help us get there.”
Using Watson’s Natural Language Classifier, Personality Insights and other services, Guiding Eyes analyzed data regarding genetic, health and behavior data spanning from birth through training for 1,200 dogs. The organizations reported that the Watson analysis predicted - with algorithms developed by data scientists at San Jose State – with 100 percent accuracy which volunteer "puppy raiser+dog" teams would be successful in developing training program graduates based on socialization, raiser environments, the genetic backgrounds of the puppies and the skills of the puppy raiser.
Russenberger explained that the San Jose State researchers have developed algorithms that track and analyze puppies’ responses – measured via heart rate – with timing of various stimuli, such as encountering a stranger, the distraction of a another dog or a wild animal, or a loud noise.
Beyond the traits you’d expect in a pup that sails through service training, Russenberger said the successful canine candidate must above be happy in its work, with a demonstrated joy and willingness. Closely related to this is poise: the dog must be able to continue to function at a high level during stressful situations.
Russenberger said Guiding Eyes and IBM also are working with researchers at North Carolina State University to improve its ability to measure a dog’s aptitude for handling stress via the development of wearable devices equipped with sensors. The goal is to amass data for objective measurement of dogs, a Watson “Internet of Things for puppies” for more effective breeding, raising and training.
Guiding Eyes will make its canine data, stored in the IBM Cloud, available to external partners for additional research work.
“People don’t typically think about an organization like ours as a Big Data company, but we cannot succeed or grow without it,” said Thomas Panek, president/CEO, Guiding Eyes. “By collecting information about our dogs over the years, we can dig into the data to pull out meaningful insights about health, behavior, temperament…. However, all of that was becoming increasingly difficult on our legacy systems. By putting it on the IBM Cloud, we will simplify our IT and make our data more accessible for more analysis.”