EXPERT/R

Extension specialists and county agents are often called on to do one-on-one counseling. The more success one has in solving problems, the more one's called on to do variations of this kind of work. Apart from using valuable time, these calls can reduce already scarce funds for travel and other program necessities.

One-on-one calls also present another problem. While all specialists and agents have some areas in which they feel competent to provide advice, there are many more areas in which someone else's advice would be better. In other words, it's hard to put the best of each individual's expertise into everyone's head.

In the winter of 1985, we wondered if computer software could be designed for agents doing one-on-one work. Such programs could reduce the time required to do routine work and free the time of state specialists for work on the more unusual and difficult cases. Further, the programs could be designed by those with the most expertise in a subject, but used equally well by everyone to provide advice in the subject area.

The problem we chose to work on was that of diagnosing reproductive problems in dairy cattle. Reproductive problems are of significant economic importance, but their solution usually requires one-on-one work. In a typical problem-solving session with a dairy producer, two or three farm visits are required. The first determines herd status, identifies potential problem causes, and sets priorities for further actions. These actions can include recording specific data for refinement of diagnosis and/or adoption of management practices. The next visit is necessary for about half the herds, to improve the diagnosis. The last visit is required to adapt recommended management practices to the peculiar set of conditions (animal housing, labor availability, etc.) for that herd. This process often takes a year or more to complete.

With no formal training in advanced computer techniques, no money for programmers, and only microcomputers with which to work, we set out to design a computer program that would expand our capabilities to make routine diagnoses. In this article, we briefly describe our program (called EXPERT/R) and how we're using it. Readers interested in learning more about the program's details are encouraged to contact us directly.

EXPERT/R is designed to mimic the thought process of a human expert in solving a problem. It asks questions, uses the answers to make deductions, and produces a statement of the problem's solution. A typical diagnostic session with the program requires about 15 minutes and produces a report of one-half to one-and-one -half pages long.

In EXPERT/R, a collection of 36 questions is the means the program uses to find out information. Typical questions ask information on herd statistics, use of drugs, condition of dry cows, and whether certain tests are routinely performed. The answer to each question asked is used to determine which question is to be asked next, so the user will never be asked questions that aren't relevant to his/her particular situation. A herd with excellent reproductive management will be asked as few as four questions, but a herd with several and/or severe problems will be asked many more questions.

Each session with EXPERT/R produces a diagnostic report. If problems are diagnosed, they're brought to the user's attention in the report, as is the rationale for the diagnosis. This helps users more fully understand the diagnosis and thereby increases their confidence in the report.

Finally, recommendations are made on how to solve the diagnosed problems. At times, an EXPERT/R report might recommend that better records be kept, regular veterinary check-ups be scheduled, or the use of certain drugs be discontinued.

The choice of a programming language for EXPERT/R was based on three factors:

1. The program was to be used in typical situations encountered by county agents, veterinarians, and dairy farmers.
2. The program had to run on the Radio Shack 16B microcomputers that had been installed in all county Extension offices in Maryland.
3. The program had to be relatively compatible with many computers and within the abilities of other specialists to modify.

These factors led us to program EXPERT/R as a BASIC program using extensive text files as data. It will run on a machine with 64K RAM and requires about 120K of disk storage space. The text files, which contain the diagnostic reports, were prepared with an ordinary word processing program and therefore required no programming expertise to assemble. The questions and rules for selecting reports were included in the BASIC program.

Our initial use of EXPERT/R was as a diagnostic aid for 26 dairy herds participating in a demonstration project focusing on the efficiency of dairy production. For these 26 herds, a diagnosis made by the program was compared to one made by the herd's veterinarian. Fifteen different veterinarians worked with the 26 herds. Two were dairy cattle reproductive specialists and the others had practices that treated many types of livestock and companion animals.

The results of the comparison confirmed the accuracy of the program. Only three of the diagnoses made by the veterinarians were judged to be potentially more accurate than those provided by the program, and those were all made by the veterinarians who were reproductive specialists. The program did as well or better than the veterinarians in the other 23 cases. This type of testing is important not only for validation of the existing system, but to indicate directions for future improvement.

We've been pleased with the reaction of county agents who are now using the program. They were provided with example herds to run with EXPERT/R, but quickly went beyond this level of testing and ran other diagnoses. In one county, the agent used the program with four herds. In each case, the producer had contacted the agent about other matters, but dairy reproductive problems came up in conversation, and the agent ran EXPERT/R. These herds all had relatively simple diagnoses, and the agent was able to make recommendations that helped these producers without any specialist visits.

In a second county, the agent ran EXPERT/R using the producer's Dairy Herd Improvement Association records and produced a preliminary report. The agent then made a farm visit with the local veterinarian and obtained answers to the remaining EXPERT/R questions. EXPERT/R indicated that further data were needed before making a full diagnosis. Thus, two of three possible specialist visits were unnecessary for this herd because EXPERT/R was used.

An unexpected use of EXPERT/R will be made in a third county. The agent plans to run EXPERT/R on 15 county herds from the county. He'll then use the pattern of problems found to set priorities for the educational needs of dairy producers in his county. The problem areas will be targeted with mass media releases and annual Extension meetings.

Acceptance of the program was slow at first, but now EXPERT/R has become a useful tool in our overall educational effort. The hardware and software requirements are within easy reach of any Extension effort, the results of the program are accurate, and the county faculty are adapting to the new way of doing business. Our current work involves refining the program, investigating farmer-level response to using the program, and exploring ways to deliver EXPERT/R on a statewide basis.

Based on our experience with EXPERT/R, we feel that computer software of this type holds the promise of helping specialists and agents live better with fewer resources. We can easily see this approach to one-on-one work as an important part of tomorrow's Extension.