Watson: A Useful Tool for Insurance, But Not a Trusted Ally

It isn’t often that my ramblings elicit paragraphs of response from sources as important as IBM, but since my friend (and yes, he is my friend), Jamie Bisker, global insurance industry leader for IBM, has sought to wax poetic on the niceties of Watson in a response to my writings, I thought it fitting to devote at least one more blog posting to this important subject.

Watson, you may recall, is an IBM-built expert knowledge computer that famously beat the pants off two “Jeopardy!” champions in a specially designed “Jeopardy!”-style competition. IBM had hoped, though this event, to demonstrate Watson’s ability to process and respond to natural language. I found Watson to be an impressive storehouse of factual knowledge, but not so impressive on understanding the finer points of our language. I also agreed that a Watson-style expert system would be very useful for insurers, but I questioned whether or not it could be as helpful as, say, the mythical onboard computer from the fictional Starship Enterprise, which was apparently part of the inspiration for the competition.

In his response to me, Jamie says, “The BHAG (big, hairy, audacious goal) of competing against expert “Jeopardy!” players occurred to an IBM executive after seeing almost an entire bar full of people get up en mass to see if Ken Jennings could keep his winning streak alive. He wondered if a computer could ever compete at that level.”

I would say that despite the final score of the event, Watson did not demonstrate that it could compete at the same level, because while the human contestants had to base their “questions” on auditory processing of language, Watson had to be spoon-fed the “answers” in electronic form. (Remember the “Jeopardy!” contestants must provide the question suggested by an answer that is given.) Part and parcel of the natural language picture is that the human, or the device, must be able to understand and act on spoken language. There are plenty of applications out there that are designed to understand, and even reproduce, human speech. The problem with those applications, however, is that some “training” to a certain speaker’s voice is necessary. Even then, such devices are prone to make errors that could dramatically alter the message.

I agree with Jamie that “Jeopardy!” categories and clues are “intentionally vague, misleading, colloquial or rhyming in nature,” but if your intention is to demonstrate the natural language facility of Watson, surely that comes with the territory. The human contestants could complain that Watson has been packed chock-full of facts from human-generated sources such as almanacs and other references—and humans did the packing! But, if you’re going to take on a computer in a round of “Jeopardy!,” that also comes with the territory.

In the end, though, an expert system merely spits out the answers that it knows—or those in its database that are statistically most likely. This is very useful to insurers for questions of fact or reference citations that are unquestionable in their meaning, especially in a highly regulated industry such as ours. What the computer doesn’t get, however, is context. A phrase like “fat chance” can have several meanings, depending on the context in which it is uttered. Humans sort out the meaning by considering all facets of the situation and by inferring the speaker’s intention from tone of voice. In fact, some humor or sarcasm may be involved. Such nuances may some day be readable by expert systems, but that day is not near.

So, while I applaud Watson as a marvelous storehouse of knowledge and a useful suggester of possible answers, I don’t believe we are even close to the point where computers fully understand our spoken language, nor do I think we should make decisions based solely on the basis of Watson-style responses. I think Jamie correctly characterizes Watson as a tool, which is to say that in the hands of a human, it can be of help.

Will we have a computer that is fully responsive to and understanding of language any time soon? Fat chance.

Ara C. Trembly (www.aratremblytechnology.com) is the founder of Ara Trembly, The Tech Consultant, and a longtime observer of technology in insurance and financial services.

Readers are encouraged to respond to Ara using the “Add Your Comments” box below. He can also be reached at ara@aratremblytechnology.com.

This blog was exclusively written for Insurance Networking News. It may not be reposted or reused without permission from Insurance Networking News.

The opinions of bloggers on www.insurancenetworking.com do not necessarily reflect those of Insurance Networking News.

For reprint and licensing requests for this article, click here.
Analytics Data and information management Policy adminstration
MORE FROM DIGITAL INSURANCE