The much-ballyhooed run of
Indeed, Watson’s facility at answering questions by understanding natural language may portend a future in which insurers increasingly rely on computers to augment--and in some instances supplant--human decision making. While this editor believes Watson’s hardware specs are impressive (90 IBM Power 750 servers using 15 terabytes of RAM and 2,880 processor cores) it is IBM’s DeepQA technology that is more germane to insurers. Under development for many years, the company says the software represents a leap forward in the maturity of intelligent systems and language processing in general. What’s more, due to feedback algorithms, Watson evinces an ability to learn from correct and incorrect answers.
Yet, for all the talk of “artificial intelligence,” Watson is essentially a predictive analysis tool able to understand unstructured data paired with a rules engine. A Watson-type system could scan data currently in an insurer’s systems and produce answers without that data first having to be translated or coded into a special format.
Given the thicket of business rules and regulations encircling an insurance enterprise, the potential benefits of such a system are readily apparent. For example, a Watson-type question-answering system could read the text from regulatory filings in order to answer questions posed in everyday language.
IBM posits a future system using Watson-like technology as an “agent’s confidant” that the agent could ask questions regarding products or terms of coverage. IBM also foresees such systems as a training tool, reinforcing an insurance agent’s knowledge on an ongoing basis, say, just before a call or meeting with a prospective policyholder.
In a similar sense, such natural language abilities could be a boon to claims servicing, enabling insurers to partially automate the answering of often complex questions that policyholders and third-party claimants have about their claims. A Watson-type system could combine structured data from a claims system such as claim number, dates, and existing actions, with unstructured data such policyholder statements, adjuster notes and police reports to provide an answer.
