Insurance Networking News asked James Bisker, global insurance industry leader, IBM Institute for Business Value, to define knowledge-based, expert and artificial intelligence systems and provide insight into how they can benefit insurance industry operations.INN: There has been some confusion in the marketplace about knowledge-based/expert and artificial intelligence systems. Can you clarify?
JB: It's important to recognize that some of the terms that are used interchangeably when discussing this topic really refer to different things. For example, artificial intelligence (AI) is an academic discipline comprised of a set of sub-disciplines. The most common of these is knowledge-based systems (KBS) that work to make the existing components of knowledge in a particular area (called a "domain" in AI lingo) available in a consistent and reusable form. A term I like to use when referring to the use of AI in business situations is "intelligent systems."
The type of KBS that is often referred to as an "expert system" is technically known as a rule-based system. The expert systems moniker was attached to this technology because it was initially used to provide the same level (i.e., quality) of answers that a human expert would be expected to deliver for a given situation. Rule-based systems capture and retain knowledge for reuse as rules that can take the form of an "if-then-else" type of statement that is familiar to computer programmers. The same structure is familiar to almost anyone else as "rules of thumb;" for example, if it is cloudy and the temperature is cool, take an umbrella in case it rains.
INN: What about their use in problem-solving?
JB: The problem-solving capability of rule-based systems and other KBSs is limited to situations where existing information can be manipulated to produce an answer to a problem. In the insurance industry for example, for almost two decades the problem of consistently underwriting a large volume of straightforward cases has been increasingly managed by "expert underwriting" systems.
INN: Where else in the insurance industry today are these systems put to use...are they embedded in other applications (are insurers aware they are using them)?
JB: Although a majority of insurance companies in this country use some form of automated underwriting, even more use various forms of intelligent systems that are embedded in other applications, such as the grammar-checking function of Microsoft's word processing application, MS Word. Other examples include any system that makes use of business rules.
Business rules provide a way to abstract instructions that control the flow of a program or a system and place it more directly in the control of business users. Usually there is a set of screens or windows in such an application that allow users to enter rules, see how they impact existing rules or logic flows and test outcomes. Internally, any system that makes use of business rules is a rule-based system by definition, but may not necessarily have the full complement of robust tools that come with a true rule-based system. Regardless, exposing these rules from within business applications reduces miscommuni- cation between business units and their IT counterparts and reduces the complexity of routine maintenance for programmers.
INN: Isn't KBS tied to data mining?
JB: Knowledge-based systems are constructed to input, manipulate, edit, store and eventually execute or react to existing information. One of the most important aspects of this process is acquiring the knowledge from individuals or from data sources such as a database, or a data warehouse. A professional knowledge engineer can interview subject matter experts to collect information and knowledge based on their experience in a specific aspect of business or insurance operation.
A knowledge engineer can also direct the tasks needed to acquire knowledge from data sources. This is where data mining comes in. Without getting too technical, data mining is the process of extracting information and knowledge from data sources in a way that produces relevant and accurate conclusions. As an example, you can understand what "rules" are used to underwrite a life insurance policy by examining the underwriting handbook or by asking a selection of experienced underwriters how to process a particular case. This would be considered the "top-down" approach because you are starting with the directions for your future actions.
The bottom-up approach would be to review (via data mining) the existing policies in the company's policy database. By examining what policies were issued to which customers, a truly complete picture of the actual rules that were used can emerge. A data-mining program would read the database and start collecting and categorizing a series of facts to eventually build a picture of what applications the company actually allowed to become policies. A straightforward example of what can be learned is minimum and maximum age: By simply sorting all the policies in existence, a knowledge engineer can spot the youngest and the oldest new clients. This amounts to "mining" a "rule" from the data, bottom-up.
INN: What are some organizational implications of using KBSs?
JB: One of the important implications of using KBSs will center around their impact on individual employees. This is especially true as more insurance employees leave the workforce as they retire or seek other employment. In this case, using knowledge management systems to capture the knowledge of internal experts will be crucial. Being able to extract corporate knowledge and distribute it consistently will ensure steady performance and efficiency in times of transition. KBSs also allow less experienced staff to operate at higher levels with less oversight, which frees up more senior personnel for more complex activities. Finally, and perhaps most importantly, the use of these systems increases consistency and compliance to internal and external policies and regulations.
