Need a digital assistant?
We’ve been talking to several startup and spinout firms whose intention is to thread cognitive software into a new generation of digital assistant applications tightly focused on specific fields of expertise. Think smart apps for health insurance providers seeking to optimize their decisions on coverage across millions of different kinds of conditions and treatment options. Or financial portfolio analysis assistants incorporating comprehensive recommendations from professional research, breaking news, social media commentary and trading data.
In those cases, data is multivariate, multiformat, highly unstructured and rapidly changing, with a high requirement for currency and precision. In each of those examples, the application must be a skilled, knowledgeable advisor that operates in an information environment uniquely suited to that particular domain. Requirements include extensive content and an easy-to-use environment that enables professionals to interact intuitively with the application.
This kind of problem has been a tantalizing challenge for information systems for decades. An early generation of expert systems received a lot of attention in the first wave of artificial intelligence research in the 1980s. Those projects centered on knowledge engineering—deconstructing the logical and systematic processes used by human experts in a particular field and representing them in a computer program. The goal then was to replace the human experts with a machine that could make good, human-like decisions when faced with the kinds of cases it was programmed to solve. Interest came from firms that couldn’t hire enough human experts in specific fields, that were facing “brain drain” problems with generations of engineers and other experts retiring at the same time, and that wanted to automate human support personnel out of the payroll.
The human process
The first-generation expert systems failed for many reasons, but the most important one was that they did not take into account the way humans actually make complex decisions.
When the human brain makes decisions, it first gathers evidence—facts, similar experiences, recommendations, reviews. Then it processes the evidence by categorizing it, weighing it for strength of evidence and examining each possibility for its advantages and disadvantages. But the straightforward logic we believe we rely on only gets us so far. If there are too many choices, too many parameters or the decision must be made instantaneously, we default to an entirely different decision-making pathway. We call it relying on our “gut feelings.”
Snap decisions or occasional flashes of brilliance may actually dredge up connections that we didn’t know we’d made, triggered by a need to make sense of something new. The need to decide complex issues quickly causes us to round up ideas and sort them anew. It bypasses the slower pathway of logical weighing of pros and cons. Jonah Lehrer wrote about it in his book, How We Decide.
The successful system for enhancing human expertise has to take both kinds of decision pathways into account—lock in the highly structured approaches, but enable the unexpected, the intuitive, the creative, the thinking “outside the box.”
Here’s an example of why finding the unexpected, although unsettling, is far more valuable than sticking to the highly structured. A straight-A college student’s grades began to plummet. The counseling team recommended tutoring, and a visit to a therapist to evaluate her for depression. But one of her counselors was skeptical. He knew her well, and he also had a background in biology. His gut feeling was that it didn’t feel like depression. He felt that this abrupt change in her behavior might have a medical cause. He sent her for a physical. She had cancer. They caught it in time. Her grades went back up, and today she is a doctor. That counselor happened to have the right facts and could combine them with what he knew about that student to come up with an alternative hypothesis. His intuition saved her life, but it was a happy accident.
How can we make sure that happy accidents happen more frequently, even if a practitioner doesn’t have the requisite facts at his or her fingertips? We need help amassing information so that we can tap into it, sort it, re-sort it and examine multiple hypotheses to see what makes the best sense within the context of what is happening right now to that person or organization. Because we can’t anticipate every situation, we need to create a kind of information soup with facts, causes, effects and relationships, all waiting to be assembled when needed.
This is where cognitive software applications can shine. By amassing more information than any of us can individually, and then presenting it in an analytic environment with our problem as the lens, a cognitive app can encourage the intuition and creative imagination of the human expert. Cognitive computing augments the human capacity for taking in random information and combining it in novel ways within the context of a current problem.
That’s the promise of the new cognitive digital assistant applications. From planning a vacation to detecting epidemics or healing children and pets, this new generation of digital assistants is primed to answer complex questions as they arise, by selecting the appropriate elements in the information soup and combining them to detect powerful and unexpected patterns.
We’re no longer building machines to replace the experts. It’s more accurate to say that experts are now demanding machines that can help them navigate and organize the raw materials for their complex work environments—so that they can find those accidental insights.