AI: Yes, Its Here

According to the Dictionary of Computing & Digital Media, artificial intelligence is defined as software that makes decisions based on accumulated experience and information utilizing functions such as learning, adapting, reasoning, and self-correction.

Although the definition is simple, the implications are not. Decision-making, learning, adapting, reasoning. These issues get to the heart of what makes us human.

Following the Rules
While data mining has been referred to by some as AI, it is not actually an AI technology. Data mining is actually an application that utilizes AI technologies to uncover trends in large quantities of data.

The most commercially successful of AI technologies are expert systems, also known as rules-based or knowledge-based systems. Much like having a database in a traditional business application, expert systems have a knowledge base that is frequently represented as an in-depth series of if-then-else statements. When presented a problem, the expert system searches the knowledge base for appropriate statements/logic in order to produce the decision. The knowledge base can be considered the rule book for how to handle an issue and make a decision.

The key is that expert systems apply a set of pre-programmed rules to the data. The system does not learn or establish new rules on its ownsoftware engineers must establish and change those rules as issues are better understood. Because of this limitation, these types of systems are used to help users make better decisions, as opposed to eliminating decision makers.

Neural Networks and The AI Myth
Another flavor of AI used in business that actually has the ability to learn through experience is the neural network. Designed to mimic the human brain, these systems use a multi-layer perception network rather than rules. Neural networks differ from rules-based systems in that they respond to new situations by analyzing previous responses and situations, extrapolating from them to learn a new response. Applications that may use a neural network are credit card companies to detect fraud, or customer relationship management (CRM) software that attempts to anticipate a customers request or need.

Unfortunately, rules-based systems and neural networks are not as glamorous as the images conjured up when we hear the term AI. They still have no common sense, intuition, or ability to put a decision in context.

While the HAL-9000 vision of AI might not have been abandoned, it is far from ready for the business world. The experts may have had 45 years of research and development to work on conquering the Turing Test, they have yet to develop a computer system that can mimic a human.

AI in Insurance
Once you get past the science-fiction vision of artificial intelligence, what kind of value can AI software create for the insurance industry, particularly on the loss side? Its simple: increased revenue and decreased cost.

Enterprise software for services businesses, such as insurers, is designed mostly to accumulate and interpret data from the various stakeholders (vendors, employees, customers). The last 20 years of effort in developing business technology has created the infrastructurenamely, the Internetfor linking these geographically dispersed stakeholders and has delivered the computing power necessary to capture and classify business information.

By using rules-based and other AI technologies, insurers can convert systems from ways of collecting data to a means of leveraging data to improve the quality of financial results.
Want some examples of insurance applications that can use AI technologies?

LITIGATION. Rules-based logic can help find improper billings, overbillings, and duplicate billings. Patterns can be identified to determine what a case should realistically cost by county or line of business, for example.

CLAIMS. When a claims adjuster is considering whether to pursue litigation on a claim, a rules-based system may be able to inform the adjuster that, based on outcomes collected for similar cases in similar jurisdictions, this particular claim may best be settled by the adjusteror stands a good chance of winning in court.

While the adjuster makes the final call, the system provides a second opinion. This makes the adjuster more efficient, saves hard dollars, and raises the quality of the decision-making process.

FRAUD DETECTION. Neural networks, artificial intelligence, some rules-based logic, or other statistical techniques can be used to uncover relationships. Relationship discovery can be as simple as uncovering different claimants seeing the same doctors on a case with the same lawyers, or finding claimants with similar names and addresses who have filed claims for multiple accidents.

TARGETED MARKETING. Using data mining tools, some companies have identified target markets by blending data from such disparate sources as demographics, psychographics, and buying behaviors.

UNDERWRITING. Rules-based underwriting has been around in the personal automobile marketplace for decades. Automating this function is fairly straightforward, freeing the underwriter to handle those customers requiring closer inspection.

I Say You Need an Evolution
All these applications have implemented a flavor of AI; they all make life easier for the claims adjuster, underwriter, or marketing person. But they also still involve setting up the rules and dependent variables, or require multiple manipulations of data using either standard statistical techniques or data mining application tools. The insurance industry has yet to see AI technologies that learn from previous interactions.

You might say that the insurance industry has gotten AI technology to the point where its well educated but lacks street smarts. And oftentimes, street smarts make the difference in a successful business decision.

Although not implemented as widespread as we would have hoped by now, AI technologies are gaining a growing acceptance within the insurance industry as we learn more about the technologies, their advances, and their application to gain true business benefits. From too much hype to too little deliveryand a technology in search of an applicationartificial intelligence has suffered the pitfalls of a good concept: poorly positioned, marketed, and implemented.

But that was then, this is now. The time has come to brush up our definition of AI, and start taking advantage of AIs value to the insurance process. The applications being offered to the insurance industry today and touted as artificial intelligence are most likely AI in its rules-based form.

It may be time to bring the term AI back to life. But this time, lets do it with the right definition, achievable expectations, applications to solve a business problem, and the tools and skills in place to effectively implement AI in order to achieve the business benefits it can.

Derek Koch (derek.koch@visibillity.com) is director of product development with Visibillity.

Glossary of AI Terms

ARTIFICIAL INTELLIGENCE. Software that makes decisions based on accumulated experience and information with human like functions such as learning, adapting, reasoning, and self-correcting. (Source: Dictionary of Computing & Digital Media)

DATA MINING. The application of artificial intelligence techniques to large quantities of data, to discover hidden trends, patterns, and relationships. (Source: Meta Group)
Expert systems. Systems that represent knowledge in a way that matches human problem-solving capabilities. Requires human experts to provide the knowledge that is encoded in the system. Also known as rules-based systems, knowledge-based systems. (Source: Conning & Company).

FUZZY LOGIC. Originally introduced by Lotfi Zadeh in the 1960s, fuzzy logic recognizes more than simple true and false values. It uses more than strict binary (True or False, 0 or 1) decisions by using soft linguistic variables (e.g., large, small, hot, cold, warm).

INTELLIGENT AGENTS. A technology that essentially acts on behalf of a human by performing its actions autonomously and somewhat proactively and/or reactively. Intelligent agents typically exhibit some learning, co-operation, and mobility attributes. Also known as software agents, software robots, or bots.

NEURAL NETWORKS. Based on the design of the human brain, neural networks are an AI technology designed to mimic the network of connected neurons in the brain. Using a multi-layer perception network, neural networks are able to respond to new situations by learning. Also known as connectionist architectures, parallel distributed processing, or neuromorphic systems.