HARTFORD, CONN. -- At the American International Group of insurance companies, computers are making recommendations to underwriters about how earthquake insurance policies should be written.

At Hartford Steam Boiler Inspection and Insurance Co., computers are analyzing customers' machinery and suggesting ways breakdowns can be avoided.

At Aetna Life and Casualty Co., computers soon will be helping new underwriters learn the complex skills and thought processes that experts use in writing workers compensation coverage.

Those programs represent a bold new use of computers in the insurance industry. Computers, long vital in processing vast amounts of data, now are going one step further -- they are helping to make judgments and assist insurance professionals in underwriting, claims processing, loss prevention, financial advice and other tasks.

Called "expert systems," these sophisticated software systems for computers have the ability to ask "what if" and "why" questions. The systems emulate human thought through a combination of computer science, electronics and engineering, which is known in the computer world as "artificial intelligence."

The systems have the potential to increase productivity dramatically, prompting some of the largest insurers to spend several million dollars a year in a tightly guarded research race to produce the most sophisticated systems.

Insurers say that expert systems will increase the consistency, speed and quality of insurance work, while freeing some human experts from the drudgery of repetitive tasks. Insurance companies believe the systems will enhance their competitiveness and eventually will pay for themselves several times over.

"We're starting to view knowledge as an asset," said John Vande Creek, a vice president of systems architecture with Continental Insurance Corp. "We want to retain it, broaden it and share it."

A recent survey by the accounting firm of Coopers & Lybrand showed that two-thirds of the top 100 insurance companies are using or developing expert systems.

"A well-done expert system will allow you to arrive at logical conclusions in some cases, even though you may be missing some of the facts. It can mimic the behavior of an underwriter," Vande Creek said.

Expert systems are not being viewed as replacements for human thinking and expertise, but as powerful tools to supplement and enhance human knowledge. "Unlike data processing, expert systems need to be constantly monitored and changed because of the dynamics of the marketplace," he said.

Companies stress that expert systems' recommendations can be modified and their judgments overruled by users.

An example of how an expert system works can be seen in the Underwriting Advisor, one of the most sophisticated systems, now being used by the American International Group of New York to help write coverage for special risks including earthquakes.

Like a trained underwriter, the system raises a series of questions when analyzing the risk and, after arriving at an assessment, spells out its reasoning.

It can also ask for more information or indicate to an underwriter what other information would be useful in reaching a decision on a risk.

The battle to develop the best and most far-reaching expert systems is being waged by such companies as Travelers Corp., Aetna, the American International Group, Cigna and Fireman's Fund Insurance Co.

Travelers wants to apply expert systems to property and casualty underwriting, claims and almost every area of the company, said Ron Bristol, a vice president of computer science.

"We're trying to train more and more people in the company. ... We think these programs give us a competitive edge," Bristol said, noting that Travelers is spending about $500,000 a year on expert systems and it expects a payoff of three to four times the development cost.

One of the most ambitious ventures under way is Underwriting Advisor, which is being used by two other insurers besides American International Group.

It is being used in writing coverage for workers compensation insurance, fire insurance and a special property-casualty insurance known as "difference in conditions."

Difference in conditions focuses on such potential problems as earthquakes, floods and burglaries.

Developed by a fledgling Sunnyvale, Calif., firm, Syntelligence Inc., it was backed by American International, Fireman's Fund and The St. Paul Cos.

Syntelligence developed a single core system for all three, which it modified for each of the businesses.

Joel B. Mittler, the insurance group manager at Syntelligence, said the system has two parts -- a controlling program that interprets the rules for all three companies and an individual knowledge base for each line of business.

Fireman's Fund has used its expert system to aid underwriters with workers compensation policies in four locations in California and will expand nationwide by the end of next year.

"They will have a significant impact on underwriting losses and they'll improve productivity," Mittler said.

He said Syntelligence has spent about $20 million on the Underwriting Advisor and another expert system it developed for banks called the Lending Advisor, which is being used by Wells Fargo and Bank of America.

Donald Mick, director of corporate technology planning for Aetna, said the company is spending more than $1 million a year on expert systems and they are being studied for use throughout the company.

One expert system will be used at Aetna next spring to supplement the training of underwriters in the commercial insurance division, starting with workers compensation policy-writing, according to Mary Ellen Negri, an Aetna senior software consultant working on a tutorial system there.

Negri said the system would simulate cases similar to those an underwriter usually would handle.

"The interaction between the system and the student is directed by the student," said Negri. "The instruction will be individualized to address a student's strengths and weaknesses."

Aetna showed 20 underwriting trainees the system in March, she said, and the reaction was enthusiastic.

"They felt it would enable them to develop a model for underwriting. They had a very positive response," she said.