Cognitive Computing Helps Vision Impaired

Data analytics, a force for change in nearly every business, is now being leveraged to make lives easier for the vision impaired by increasing the graduation rate for guide dogs.

According to Guiding Eyes for the Blind, a nonprofit that breeds and trains service guide dogs, a mix of genetic and temperament data combined with natural language processing will bump the graduation rate to 59 percent from fewer than half. The initiative is also expected to lower the $50,000 cost to train each dog.

Why Guide Dogs Are So Expensive

To see why a few percentage points matter, it helps to know how long it takes to make a guide dog. Guiding Eyes, based in New York State, says it takes about two years. For roughly the first 16 months, puppies live with volunteer raisers who socialize them and teach basic manners. Dogs that show strong foundational skills then spend at least five months in formal harness training before they’re matched with a person.

Every dog that drops out late in that process represents two years of care, food, veterinary bills and volunteer time that doesn’t end in a working guide. The organization says it can cost up to $50,000 annually to train and care for a guide dog through its working life, and it provides its dogs to people who are blind or have vision loss completely free of charge. So the graduation rate is the single biggest lever on cost. Raising it from under half to 59 percent means many more working dogs from the same number of puppies.

Guiding Eyes is using a combination of IBM Bluemix and Watson to analyze the organization’s structured and unstructured data. By moving half a million medical and genetic records and more than 65,000 temperament records to the cloud, Guiding Eyes uncovered insights into genetic, health, behavioral, and environmental factors that correlate to successful guide dog behavior and performance.

The organization is also applying natural language processing to trainer and foster family questionnaire answers to uncover insights into personalities and temperaments. This information is expected to improve the process of breeding, raising, and matching service dogs to owner.

Structured and Unstructured Data

The mix of data types is what made this project interesting. Medical and genetic records are structured: they fit neatly into tables, with fixed fields like hip scores or test results. Questionnaire answers are unstructured. A puppy raiser might write that a dog “gets nervous around buses but settles quickly,” and that kind of detail doesn’t fit in a spreadsheet.

Natural language processing is the technology that turns free text like that into something a computer can count and compare. Put the two kinds of data together and an organization can start asking better questions, such as which early behaviours, in which family lines, tend to predict a dog that finishes training.

Cognitive Potential

Cognitive computing simulates human thought processes in a computerized model. According to author Bernard Marr, who has written several books about big data in business, the intent of cognitive technology is not to replace humans, but to expand on our capabilities and allow us to better process and understand the world around us.

IBM executives recently told Fortune that cognitive computing or machine learning is expected to become a $2 trillion USD market in the next 10 years, in addition to the $3 trillion opportunity in more traditional IT gear like servers, software, storage boxes. For startups, it presents a lucrative opportunity that can help deliver more innovative solutions and enhanced customer experiences across every industry.

Guiding Eyes Today

The data work has become part of how Guiding Eyes describes itself. It now calls itself “a global leader in breeding and genetics”, with a breeding program “built upon best-in-class analyses of genetic and behavioral data from every step of a dog’s life.” It says it chronicles “millions of data points” from the reproduction lab where puppies are conceived through to the final years of a working, retired or career-change dog’s life, “helping us improve the next generation of guide dogs with every new litter.”

The organization was founded in 1954 and says it has graduated more than 10,000 guide dog teams. It places approximately 100 guide dogs a year, supported by more than 2,000 volunteers. Its breeding colony, at its Canine Development Center in Patterson, New York, is made up mostly of Labrador Retrievers, with some German Shepherds.

What Happened to Watson

The “cognitive computing” label and the Watson brand that powered this project have changed since 2016. IBM later narrowed where it used Watson, most visibly in health care. In January 2022, IBM agreed to sell its Watson Health data and analytics assets to investment firm Francisco Partners, saying it was focusing on its hybrid cloud and AI platform strategy.

Machine learning itself kept growing, mostly under other names. Today the same ideas, pattern-finding in large datasets and reading natural language, sit underneath the generative AI tools that took off from late 2022.

What It Means for Nonprofits and Startups

The Guiding Eyes story is a good model because the goal was so concrete. It didn’t set out to “use AI.” It set out to raise a graduation rate and cut a cost, and it had years of records to work with.

That’s the lesson for any organization thinking about analytics: start with a measurable problem and the data you already collect. The fanciest model won’t help if the records behind it are patchy, and a modest one can go a long way when the data is good.