Advanced Computer Technology - Princeton University - Home

Chapter 4

Advanced Computer Technology

Photo credit: U.S. Department of Agriculture, Agricultural Research Service

Contents

Page

INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99SPECIFIC COMPUTER TECHNOLOGIES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

Knowledge-Based Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99Interfacing Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109Other Computer Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

SUMMARY/PROGNOSIS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123The Current State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123Mid-1990s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1232000 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

CHAPTER PREFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

BoxesBox Page4-A. An Example Rule for an Expert System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1014-B. An Example of an Explanation provided by an Expert System . . . . . . . . . . . . . . . . . . 1014-C. An Example Recommendation from FLAYER . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1044-D. An Example Query and Answer to a Natural-Language Interface . . . . . . . . . . . . . . . . 112

FiguresFigure Page

4-1. Trends in Semiconductor RAM Density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1004-2. Trends in Microprocessor and Mainframe CPU Performance Growth . . . . . . . . . . . 1004-3. Effect of Quality of Management on Milk Response of Dairy Cows

Receiving bat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10044. Structure of an Expert System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1004-5. System Structure for an Object-Oriented Simulation System . . . . . . . . . . . . . . . . . . . 1064-6. Functional Components of the Crop Production Expert Advisor System . . . . . . . . 1154-7. Topology of BITNET Commections in the United States . . . . . . . . . . . . . . . . . . . . . . . 1164-8. States with Participants in DAIRY-L . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1174-9. Volume of DAIRY-L Messages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

4-10. Volume of DAIRY-L Requests for Remote Retrieval of Text Filesand Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

.—

TablesTable Page4-1. A Partial Catalog of Research Applications of Robots in Agriculture.. . . . . . . . . . . 1194-2. A Partial Catalog of Research Applications of Sensors in Agriculture . . . . . . . . . . . . 121

Chapter 4

Advanced Computer Technologies

INTRODUCTIONSince the industrial revolution, agricultural systems

have intensified, and agricultural productivity has sig-nificantly increased along with farm size. Labor-savingdevices on farms have increased output per workerseveral-fold. and advances in understanding and appli-cation of biological principles have significantly boostedagricultural yields. With greater production per acre andanimal, however, farm management becomes corre-spondingly more challenging and complex. In general,methods for making management decisions have failedto meet this challenge. As a result, many decisions are‘‘ uninformed” and many agricultural systems poorlymanaged.

The application of advanced computer technologies toagricultural management can help remedy this situation.improved access to information will allow farmers tomore effectively monitor progress toward optimal per-formance. Computer technologies of potential use to ag-ricultural managers are advancing at a tremendous rate.The performance of computers has increased several-foldwith each new generation of computer chip (figures 4-1and 4-2). In the last decade. microcomputers have evolvedfrom 64-kilobyte machines with a 320-kilobyte floppydrive to machines with several megabytes of memoryand several hundred megabytes of permanent storage;such machines approach the performance of mainframecomputers (25 , 54) and can store massive amounts ofinformation.

Advances are also occurring in software technologies,allowing improved utilization of stored information. De-cision support systems, for example, provide enterprise-specific, expert recommendations to decisionmakers.Several other types of information technologies allow forrapid access to the latest information.

These advances will provide the tools to improve farmmanagement. For example. close monitoring of” animalperformance will allow early detection of diseases andcan help reduce stress in animals. Overall, advancedcomputer technologies can provide managers with theability to systematically determine the best decision ratherthan arrive at decisions in an ad hoc fashion. Optimaldecisionmaking requires a holistic view of a farm enter-prise, factors that affect it, and the probable conse-quences of management decisions. Thus, a farmer decidingwhether to plant a specific crop on a specific field shouldweigh the profitability of the crop as well as overall farm

needs (i. e., nutrition requirements if it is an animal en-terprise). The decision will impact land sustainability andthe need to use certain pesticides and herbicides or otherpest-control methodologies. Computer technologies, byproviding the capability of taking these multiple factorsinto account, can help producers arrive at the best pos-sible decisions and management strategies.

The quality of management, in turn, will influenceproductivity as well as the future impact of some bio-technologies. For example, the response of milk cows tobST is directly related to management. Poorly manageddairy herds have a lower response to bST than well-managed herds (figure 4-3).

SPECIFIC COMPUTERTECHNOLOGIES

Computer technology is changing at an unprecedentedrate on three different fronts, causing a "three-dimen-sional" information revolution. Rapid advancements intraditional database and computational programs: in sym-bolic computing and artificial intelligence: and in systemsthat improve access to information constitute the threedimensions of the information revolution.

Knowledge-Based Systems

Traditional database and computational programs, whichare largely numeric and follow established algorithms.are invaluable resources, but they cannot easily deal withsymbolic data or mimic an expert’s reasoning process.The so-called knowledge-based systems in the categoryof symbolic computing and artificial intelligence havethese capabilities. American agriculture is just now be-ginning to capitalize on these resources.

Essentially, knowledge-based systems present expertknowledge in a form that can be used to solve problems.In addition to expert knowledge, such systems requiresituation-specific databases. For systems that operate inreal-time, sensors may play an important part in col-lecting data for knowledge-based systems (40). Generaluses

1.

2.

-99-

of knowledge-based systems include:

recommending solutions for problems (e. g., di-agnosis),monitoring the status of a system to determine sig-nificant deviations (i. e.. management-by-excep-tion). and

.

100 ● A New Technological Era for American Agriculture

Figure 4-l—Trends in Semiconductor RAM Density

lo+@ I

1985 1975 1985 1995 2065Year

SOURCE: J L Hennessy and N P Jouppi, Computer Technology andArchitecture: An Evolving Interaction, ’ IEEE Computer Septem-ber:18. 1991

Figure 4-3—Effect of Quality of Management on MilkResponse of Dairy Cows Receiving bST

Excellent management

Good management

Average management

I

Somatotropin dose

SOURCE: D.E. Bauman, “Bovine Somatotropin: The Cornell Experience.”Proceedings of the National Invitational Workshop on BovineSomatotropin, USDA Extension Service, Washington, DC, pp.46–56.

Figure 4-2—Trends in Microprocessor and‘Mainframe CPU Performance Growth Figure 4-4—Structure of an Expert System

1985 1970 1975 1980 1985 1990

YearSOURCE: J L. Hennessy and N.P. Jouppi, “Computer Technology and

Architecture: An Evolving Interaction, ” IEEE Computer Septem-ber:18, 1991

3. forecasting the behavior of a system (i. e.. simu-lation).

Expert Systems

Expert systems are the most popular knowledge-basedtechnology in agriculture. The main benefit of expertsystems is that they emphasize knowledge acquisition.not programming.

User lnterface

Interface engine Knowledge basecontrol strategy domain knowledge

rulefacts

“How to do it” “what to do”

SOURCE: Office of Technology Assessment, 1992

Expert systems are distinguished by a unique structurethat separates ‘‘What to do’ from ‘‘How to do it’ (figure4-4). The knowledge base tells the program what to do.It contains the expertise for solving the problem withoutthe control structure found in traditional programs. Thesecond component of an expert system is an ‘‘inferenceengine’ that, in effect, shows the program how to dothe task at hand. The inference engine contains the con-trol strategy that determines how to combine domainknowledge to solve the problem.

Domain knowledge can be represented in the knowl-edge base in several different forms, the most common

Chapter 4—Advanced Computer Technologies .101

of which is rules (e. g., “If the leaves are brown, thenapply insecticide X’ see box 4-A). Rules correspondclosely to the natural reasoning of experts, are modular,and are easy to maintain. As a result. expert systems areeasy to develop and to support. The knowledge in anexpert system tends to be symbolic instead of numeric.This feature allows rules to be “heuristic” in nature,akin to "rules-of-thumb. ” When exact algorithms do notexist, the rules represent the expert’s best guess (94).

Another interesting feature of expert systems is theircapability of incorporating uncertainty into rules. Forexample, the rule "If the leaves are brown, then applyinsecticide X; 0.3’ means that there is a 30-percent cer-tainty or confidence in the conclusion. Strategies havebeen developed for combining the uncertainty of rules togive a confidence value for each recommendation (7,76). Therefore, the expert system is able to make rec-ommendations even when the circumstances of the prob-lem are uncertain. This ability mimics the reasoning ofan expert. Expert systems have the added capability ofexplaining the reasoning used to derive a solution (see

Box 4-A—An Example Rule for an ExpertSystem

IF● you are willing to speculate for higher prices AND● the price trend is up A N D● the basis trend is weakening AND● the basis trend is not expected to reverse soon

AND● the timing is harvest AND● the type of available storage is farm OR● the type of available storage is commercial AND● you need downside price insurance AND● Storage revenues are greater than storage costs,

THEN● forward contract your grain and buy call options.

SOURCE: R.H. Thieme et al., “Expert System Tech-niques Applied to Grain Marketing Analysis,” Computersand Electronics in Agriculture 1:299, 1987.

box 4-B), much as an expert might. The explanation isa map of the rules chained together by the inferenceengine ( 102).

Because expert systems separate the inference engineand knowledge base, it is easy to remove the knowledgefrom the expert system. leaving a shell that can be reusedin other applications. The shell contains the inferenceengine, user interface. and other domain-independentmodules. The first expert system shell was EMYCIN,which resulted when the knowledge base was removedfrom MYCIN, an expert system that diagnosed humanblood diseases (89). Expert system shells have becomesaleable products, and several are commercially availablefor use in agriculture ( 14).

There are numerous examples of expert system appli-cations in agriculture. These systems have tended to bediagnostic systems for addressing relatively narrow prob-lems. Large-scale, broad-based expert systems have notbeen developed in agriculture. The following overviewof agricultural expert systems includes systems devel-oped for business decisions. animal production, and cropproduction.

Farm and Area-Wide Management—effective de-cisions regarding the planning, organization, and controlof a farm enterprise are essential to agriculture. The leg-islative, economic, and environmental demands placedon farmers and government agencies that implement ag-ricultural policy create a need for tools that help makesound farm-level and regional policy decisions in agri-culture. Tools that help with agricultural problems at awatershed or farm will become increasingly important inthe future. This will involve integration of expert systemswith geographic information systems, area-wide moni-toring systems (78 ), and remote sensing.

The financial difficulties of the 1980s vividly docu-ment the cost of poor decisionmaking in the businesssector. However, a major obstacle impedes the adoptionof expert systems in making business decisions. Businessdecisions. unlike production decisions, are generally in-

Box 4-B—An Example of an Explanation Provided by an Expert SystemI conclude that the patient has dreaded lurgy caused by bug-eyed germs with a certainty factor of .76.WHY:I concluded that the patient’s dreaded lurgy was caused by bug-eyed germs because tight abdomen (E3) and acidsaliva (E4) indicate swelling of the spleen (E), which taken together with yellowish skin cast (D) provide evidence(CF = .76) that bug-eyed germs are the cause of the dreaded lurgy.SOURCE: M. Van Horn, Understanding Expert Systems, Bantam Books. New York, NY, 1986.


fluenced by values, goals, and risk attitudes. Thus, twoexperts with the same knowledge and expertise may se-lect different courses of actions (91).

Only a small number of expert system applications isavailable for farm decisionmaking. Most existing expertsystems in this area relate to design, planning, and con-trol. Unfortunately, such functions are considered rela-tively unimportant by farm managers. Expert systemsdealing with data acquisition and interpretation, predic-tion, and monitoring have not been developed. This mayindicate that expert system development efforts are fo-cusing on applications not in the area of greatest needfor farm managers (91),

Farm-level planning and financial analysis are activeareas of expert system development. Several prototypesystems are under development. One effort at farm-levelplanning directed at farmers’ needs is the Crop RotationPlanning System (CROPS) developed at Virginia Tech(6). This system uses a map-based interface to let farmersenter data about their land (soil type, topography, land-use, and field sizes) and their farming enterprise. Basedon these data, CROPS provides farm-level or field-levelenvironmental risk evaluations for soil erosion, and nu-trient and pesticide leaching and runoff. It then uses Alplanning and scheduling techniques to generate a whole-farm production plan so that the overall farming operationcan meet user-defined yield and/or acreage targets, eco-nomic return goals, while also reducing potential envi-ronmental risks to acceptable levels. The system runs onApple Macintosh 11 systems and is adapted for use bythe Soil Conservation Service and the Virginia Depart-ment of Conservation and Recreation in their farm plan-ning activities.

The best known farm financial system is the Agricul-tural Financial Analysis Expert System (AFAES) fromTexas A&M University (63). AFAES consists of aspreadsheet to prepare operating-year and multiyear fi-nancial statements; a program that calculates financialratios and trends from the spreadsheet; and two expertsystems that develop a performance operating-year anal-ysis and multiyear analysis, respectively. This expertsystem operates on an IBM-compatible microcomputerand is marketed through the Texas Agricultural Experi-ment Station at a variety of prices based on the type ofuser making the purchase.

Other agricultural expert systems have been developedfor specific business decisions. One example is the GrainMarket program developed at Purdue University (98).This system provides advice for marketing storable com-modities (e. g., crops). An example rule from this expert

system is shown in Box 4-A. The machinery selectionprocess is aided by the Farm-level intelligent DecisionSupport system (FINDS) (49). This system integrates alinear program (REPFARM), a database managementsystem, and an expert system. The expert system is usedto form the link between the linear program and the userand to interpret the output of the linear model. The linearprogram component operates on a minicomputer, but theother components operate on a microcomputer. A deci-sion support system for planning of land use and foragesupply for a dairy farm has been developed in Denmark(34). The main components of the system are a knowl-edge base, a linear programming model, and a PASCALprogram connecting the knowledge base, model, and in-terface. The model integrates the varied business activ-ities of a dairy farm, such as crop production, storingfeeds, milk production, and utilization of manure. In-teractions between feeding and production of milk andmeat are established by use of knowledge sets. The userinterface allows for consequent analysis and can functionas a tool for calculation and optimization planning.

In addition to agriculture-specific expert systems forbusiness decisions, nonfarm business systems will impactagriculture (91 ). For example, Dologite (24) developedthe Strategic Planning Advisor to provide strategic plan-ning advice. This system provides recommendations suchas:

●

●

●

●

●

Get out of a business.Hold current position.Focus on one market niche.Invest selectively.Invest aggressively.

Animal Production—Expert systems for animal pro-duction deal with the management of farm animals andgenerally focus on disease diagnosis and suboptimumperformance identification based on technical expertise.

Most expert system activity in the area of animal pro-duction focuses on the dairy industry. There are at leasttwo reasons for this. First, the dairy industry has a na-tional data recording system (i. e., Dairy Herd Improve-ment, DHI), that provides centralized databases fromwhich expert systems can be built (99). A second reasonis that dairy animals are generally housed in confinement,and they produce a product (i.e., milk) that can be rou-tinely monitored on an individual animal basis. This isconducive to intensive management. Spahr et al., (92)outlined several potential applications of expert systemsfor dairy herd management.

Chapter 4—Advanced Computer Technologies ● 103

Some of the earliest dairy expert systems were devel-oped by Extension Specialists at the University of Min-nesota. Their first system (DMGTSCOR) ranks dairy-herd management strengths and best opportunities forimprovement using DHI management measures ( 16).Management action is suggested for the three best op-portunities for improvement. A second system,SCCXPERT, was developed to diagnose herd mastitisproblems using DHI somatic cell data and to recommendcorrective actions. Another system, BLKTNKCL, pro-vides interpretation and information about bulk tank cul-ture data for primary mastitis causing organisms. A fourthsystem, MLKSYS, provides expertise to troubleshootoperational and design problems with a milking system( 15). Two other systems have recently been developedto assist in manure management and to provide an overallanalysis of the production and financial status of a dairyfarm. All of these systems were developed in the Level5 expert system shell; as a result an effort is underwayto integrate them into a single system to allow data shar-ing among the programs. These expert systems are dis-tributed by the Dairy Extension office at the Universityof Minnesota freely to extension personnel and com-mercially for $75 ( 17).

Tomaszewski and others at Texas A&M University havedeveloped a Dairy Herd Lactation Expert System (DHLES)to analyzes DHI milk production data and to provide rec-ommendations for improving milk production ( 106). DHLEScontains a separate module (LacCurv) to graphically displaylactation curves. This system was developed in PROLOGand operates on an IBM–compatible computer. It is mar-keted through Texas Dairy Herd Improvement Associationfor $99 ( 100).

Several expert system projects are under developmentfor the dairy industry. Kalter and coworkers (45) aredeveloping a comprehensive expert system (Dairy Pert)to evaluate dairy-herd management. The impetus behindthis effort is the possible future adoption of bovine so-matotropin (bST), but the system has general applica-bility. This system currently contains over 320 rules inthe Nexpert expert system shell, a spreadsheet-based nu-trition model, and entry and advice routines based onFox’s database management software. DairyPert does notutilize DHI data because of inconsistencies among thenine national Dairy Record Processing Centers. DairyPertis funded by and will be distributed to the private sectorthrough a large pharmaceutical company. Cornell Uni-versity will distribute the system to public agencies andinstitutions. Oltenacu et al. (73) are developing a repro-duction expert system that will analyze DHI reproductiverecords and determine weaknesses in the reproductive

program. This system utilizes LISP on an IBM worksta-tion. Allore and Jones (42) are developing an expertsystem to evaluate DHI somatic cell counts that willidentify areas of management that predispose cows tomastitis. This system is being developed in CLIPS andwill operate on an IBM-compatible microcomputer.

Oltjen et al. (74) have developed a prototype expertsystem that recommends whether to keep or cull com-mercial beef cows. The rules contain knowledge relatingto the cow’s age, body condition score, calving diffi-culty, structural correctness, health, and previous repro-ductive performance. The expert system was integratedwith a simulation model to calculate net present valuefor each animal. This expert system was developed inthe CALEX expert system shell.

An expert system to assist in the management of asheep enterprise has been developed in Scotland ( 104).This system was developed without the aid of an expertsystem shell. Once a working prototype that could bedelivered to an agricultural unit was developed, this pro-ject was halted as a research project. Expert systems forthe management of sheep flocks are also under devel-opment in Australia.

CHESS is a Dutch decision-support system designedto analyze individual swine breeding herds within aneconomic framework (22). It determines strengths andweaknesses in the management of a pig enterprise. CHESSconsists of a decision-support system and three expertsystems. The decision-support system identifies and as-sesses the importance of relevant deviations between per-formance and standards. The expert systems combineand evaluate deviations to identify management strengthsand weaknesses.

XLAYER (84) is a management expert system for thepoultry industry and is one of the most comprehensiveexpert systems in animal production. XLAYER is de-signed to diagnose and estimate economic and associatedlosses as well as recommend remedial management ac-tions for over 80 individual production managementproblems significantly affecting a flock’s profitability.An example output is shown in box 4-C. This systemcontains over 400 production rules and was developedin the M1 expert system shell.

Crop Production—All commercial crop productionsystems are potential candidates for expert system ap-plications. In particular, expert systems should be con-sidered for integrated crop management decisions thatwould encompass irrigation, nutrition, fertilization, weed


Box 4-C—An Example Recommendation From XLAYERYou are experiencing an economic loss of about $725 per week because of a sudden change in the grain portion

of your layer ration. Reformulate the ration and phase in new grains gradually, even if the cost per pound is higher.

Production losses amounting to some $500 per week are being experienced because temperature in yourlayer house is exceeding 29.4 degrees Celsius. Use artificial cooling systems in regions where hot weather isexpected to continue. If layer barn has no cooling system, construct a partial budget to evaluate alternative poolingsystems such as evaporative cooling pads, roof sprinklers, high pressure misting and other forms of cooling,

Water intake is very low. Check watering systems to make sure that birds are getting adequate fresh, cleanwater.

Equipment repair costs are running $100 per week higher than normal. Check management practices relatedto the routine servicing of mechanical equipment. If repair and maintenance costs are consistently high, constructa partial budget to evaluate the replacement of old or poor functioning equipment.SOURCE: E. Schmisseur and J. Pankratz, “XLAYER: An Expert System for Layer Management,” Poultry Science 88:1047, 1989.

control-cultivation, herbicide application, insect control,and insecticide and/or nematicide application (64).

The first expert systems developed in agriculture werePLANT/ds (65), a program developed at the Universityof Illinois that identified diseases of soybeans in Illinois.and POMME(81 ), developed at Virginia Tech to identifydiseases of apple orchards. Both were written by com-puter scientists who were using agriculture as a noveldomain. Michalski, for example, was primarily inter-ested in machine learning.

Of the major crops, cotton has received the most at-tention to date, with at least three expert systems andone simulation-based management model now availableto the public (94). COMAX (COtton MAnagementeXpert), the expert system component of GOSSYM/COMAX was developed by the U.S. Department of Ag-riculture, Agricultural Research Service (USDA/ARS) inMississippi (56).1 Users of this system purchase a weatherstation linked to a personal computer running the pro-gram. The GOSSYM component is a simulation modelof cotton production that uses weather data collected fromthe weather station. The COMAX expert system uses themodel to project when to irrigate and fertilize to achieveoptimal agronomic goals. The entire GOSSYM/COMAXsystem including the weather station and computer costsseveral thousand dollars. Despite the high price tag, itis used by as many as 500 cotton farms in 15 States.

COTFLEX is an integrated expert system and databasepackage developed at Texas A&M and released to thepublic through the Cooperative Extension Service (93).

Photo credit: U.S. Department of Agriculture,Agricultural Research Service,

Farmer and consultant examine data from COMAX(Cotton MAnagement eXpert) computer program.

The overall system will eventually include a whole-farmeconomic analysis module that lets farmers evaluatewhether or not to participate in Federal farm programsor to purchase Federal crop insurance. The componentreleased to the public, however, is devoted to insect pestmanagement of the three major insect pests of cotton inTexas.

CALEX/Cotton is another integrated cotton expert sys-tem and database management tool (79). CALEX wasdeveloped as an expert system shell, and cotton was the

1 GOSSYM is a hybrid term formed by combining (1).w]piwn. the scientific name for cotton and the word simul;ition.


Photo credit U.S. Depatment of Agriculture,Agricultural Research Service.

Farmer and engineer check automated weather stationthat feeds daily weather information into the COMAXsystem to update its prediction for cotton yield and

harvest dates.

first application area. The system was supported throughCalifornia’s statewide integrated pest management pro-gram and delivered to farmers for testing and use. It isone of the best-documented attempts at delivering expertsystems to farmers for use in crop production (31). Be-cause the program was developed with State support, norevenue has been collected from its users and the projectcontinues to depend on State support.

Pennsylvania State University supports a laboratorydevoted to the development of expert systems and theirdelivery through the Cooperative Extension Service. TheUniversity has developed several expert systems using

an expert system shell (PENN-Shell) developed in-house.One of these expert systems, GRAPES, recommends pestcontrol options for insects and diseases in vineyards (83).Penn State’s expert systems all run on Apple Macintoshcomputers, and the University supports a statewide com-puter network for these machines.

USDA-ARS researchers (28) developed a knowledge-based system for management of insect pests in storedwheat. The system determines whether insects will be-come a problem and helps select the most appropriateprophylactic or remedial actions. Simulation models ofall five major insect pests in wheat have been developed;the model’s output feeds the expert system.

Evans and coworkers (26) at the University of Man-itoba have developed an expert system to serve as aFertilization Selection Adviser. The current system con-siders only one type of crop (wheat), four different mois-ture regimes (arid, dry, moist, and irrigation), one soilnutrient (nitrogen), and four different fertilizer com-pounds (urea, ammonium nitrate, urea ammonium ni-trate, and ammonia). It provides return on investmentinformation; a risk analysis module is under develop-ment. This system was developed in the LISP program-ming language for the Macintosh: however, work hasalready begun to develop a similar system using the Cprogramming language cm an IBM-compatible micro-computer.

In general, one can find expert system applications forcrop production for virtually all the major crops in thiscountry and in many countries around the world. Insectpest management, weed control, and disease identifica-tion are the most common domains. Other systems thathave received wide recognition in crop systems include:

●

●

●

●

●

EasyMacs, an expert system and database programdeveloped at Cornell University for recommendingpest management strategies for apple production;SOYBUG, an expert system developed in Floridathat helps farmers with insect pest control in soy-beans (2);SIRATAC, an expert system and simulation modeldeveloped in Australia for helping cotton farmerswith pest management decisions that has since beenmarketed internationally (36);TOM, an expert system for diagnosing tomato dis-eases developed in France (5); andWHAM, a wheat modeling expert system developedat the University of Melbourne, Australia (3).

Research Needs— Development of commercial expertsystem shells is being driven by forces outside agriculture


and is proceeding at a relatively rapid rate. However,agriculture applications generally will require expert sys-tem shells to operate in a microcomputer environmentwhereas industrial applications often reside on worksta-tions or minicomputers. Since this is a domain-independentproblem, it may be best addressed by computer scientistsoutside of the agricultural sector.

The main limitation to development of expert systemsis adoption of computer technology. To promote this areawill require more trained personnel and incentives todevelop and deliver computer systems.

Object-Oriented Simulation Systems

In addition to expert systems, another type of knowl-edge-based system that is useful for planning is object–oriented simulation. Traditional simulation systems modelthe behavior of a system by explicitly simulating indi-vidual processes. The structure of the real system usuallyis implicit in the model. Object-oriented simulation mod-els have an inverse structure; they explicitly model the

Figure 4-5—System Structure for an

Field (a parent object)

structure of the real system, and the behavior of thesystem is implicit in that structure.

Each component of the real-world system is repre-sented in the simulation as an object. Objects are unitsthat consist of self-descriptive data and procedures formanipulating that data. Objects can be represented in ahierarchy such that they inherit properties from moregeneral categories (i.e., their parents). For example, anobject-oriented simulation of a farm (figure 4-5) wouldcontain a general FIELD parent object that describes thegeneral features of all fields (e.g., a method to calculatethe area of the field). Individual fields (e.g., field 23)would be represented as unit objects that inherit the prop-erties of the parent FIELD object and may also containsome information specific to themselves (e. g., currentcrop planted in the field). Objects in object-oriented sys-tems communicate by exchanging messages. For ex-ample, if field 23 is to be harvested, a HARVEST messageis sent to the field 23 object. The field 23 object handlesthe details (internally resetting its own values) and returnsthe amount of crop harvested. This return message can

Object-Oriented Simulation System

Silo (a parent object)

PROCEDURE area - length x width/43,560PROCEDURE harvest - yield x area

reset yield to zero

message: area /

I /

II

response: 98.7 acres

PROCEDURE harvestvolume = voiume + harvest

SOURCE: Office of Technology Assessment, 1992.


be sent to a particular silo object which knows how toadd the crop to its inventory. Once the object-orientedsystem is developed, the simulation sends messages toappropriate objects in a fashion similar to farm managersgiving orders to their employees.

There are two main advantages to this type of simu-lation. First, the model closely corresponds in structureto a real system. This facilitates maintaining and ex-panding the model. Second. procedures in an object canbe represented in a symbolic fashion similar to expertsystems. Thus, object–oriented simulation can be usedto model processes that may not be quantitatively welldefined.

Object-oriented simulations have been under devel-opment since the early 1980s. An early object-orientedsimulation language (ROSS) was developed in the LISPprogramming language by the Rand Corp. for the AirForce (62). This language has been used in military ap-plications. Two early examples are SWIRL, an object-oriented air battle simulator, (47) and TWIRL, an object-oriented simulation for modeling ground combat betweentwo opposing tactical forces (48).

Object-oriented simulations are powerful tools formodeling the behavior of biological systems that are oth-erwise difficult to describe mathematically. Output fromthese systems can be used in planning and to determineimpacts of changing management procedures. However,most existing object-oriented simulation models cannoteasily be transferred to production agriculture.

Several object-oriented simulation projects have beendeveloped specifically for agriculture. Researchers at TexasA&M University developed an object-oriented model tosimulate animal/habitat interactions (82). The simulationwas specifically used to study the damage caused bymoose migrating through forest plantations. This systemwas developed on a Symbolics workstation using LISP.Another agricultural simulation was developed by USDA-ARS to model insect disease dynamics in a rangelandecosystem (9). This model is primarily a research toolfor studying the relationship between grasshoppers andtheir pathogens to assist in integrated pest managementprograms. This system was also developed on a Sym-bolics workstation using the FLAVORS object–orientedprogramming language. Another LISP-based system wasthe host-parasite model developed by Makela et al. (58)to study the interaction between the tobacco budwormand one of its parasites in cotton fields. More recently,Crosby and Clapham ( 18) used the Smalltalk languageto simulate nitrogen dynamics in plants; Stone (95) usedan object-oriented model of a mite predator-prey system

to show that chaotic dynamics rather than stable or pre-dictable cycles, might be the norm in agricultural sys-tems; and Sequeira et al. (87) developed an object-orientedcotton plant model for use in studying the interactionbetween localized pest feeding and cotton lint yield andquality. Another object-oriented simulation project is un-der development by Chang and Jones at Cornell Uni-versity for use in agriculture ( 10). This project uses aLISP-based, object-oriented programming language (B-object, Kessler, University of Utah) to model the oper-ation of a milking parlor. When completed, this modelwill be useful to dairy–farm managers and their consul-tants for parlor configurations and for identifying changesin performance when changes in parlor operation aremade.

Research Needs—The general paradigm of object-oriented programming is being incorporated into severaltraditional programming languages (e. g., C, PASCAL),but few inexpensive commercial shells exist in which todevelop object-oriented simulations. Smalltalk is a goodexample. It is a language and a development environmentin one, and it generally comes complete with many pre-define object classes developed specifically for simu-lation. Other expert system shells like KEE, Goldworks,NExpert-Object, and Level-V Object include the object-oriented paradigm and can be used for simulation. LISPoffers many advantages for prototype systems such asthe parlor project. However, LISP is not a language inwhich final products should be delivered, since it requirestoo much memory and is too slow for agricultural ap-plications. More research is needed to determine the po-tential value of object-oriented simulation for agriculture.

Knowledge-Acquisition

Knowledge-based systems are powerful computer toolsbecause they contain and apply a significant amount ofexpert knowledge to problem-solving; however, this alsoconstrains systems development. Knowledge acquisitionis a slow and tedious process, and problem-solving rulesand procedures are often hard to articulate.

Artificial intelligence can help automate one type ofknowledge acquisition (21, 66), that of rule formation.Machine learning, for example, is an artificial intelli-gence technique for automatically generating rules froma set of examples. This is sometimes called “learningfrom examples. ” It can be used to assist experts to de-velop rules or fill in where experts do not exist. Forinstance, rules for a crop disease diagnostic expert systemcan be generated using a machine learning system witha database of plant descriptions and associated diseases.


Michalski and others (65) compared rules derived byexperts and those generated by a machine learning al-gorithm (AQ11 ) when developing an expert system forsoybean disease diagnosis (PLANT/ds). The databaseconsisted of 630 examples based on 35 plant and envi-ronmental descriptors for 15 soybean diseases. One rulewas generated for each disease. When tested in an expertsystem, the machine-generated rules outperformed thosegenerated by experts. The machine rules properly diag-nosed 98 percent of the test cases while the expert derivedrules diagnosed 72 percent correctly.

A microcomputer-based machine learning system hasbeen developed for agricultural problems (27). This sys-tem was first used to generate rules for a grass identi-fication system (WEEDER). Other generic machine-learning algorithms are available as commercial products(e. g., Classification and Regression Trees, CaliforniaStatistical Software, Inc, Lafayette, CA; ID3, Knowl-edge Garden, Naussau, NY).

Due to the nature of rules generated from machinelearning (i.e., the rules indicate which variables are im-portant for describing certain results), machine learningcan also be used as a data analysis tool. Liepins et al.(57) investigated the use of three machine learning al-gorithms for analyzing natural resource data. They stud-ied the effect of storm damage on lake acidification usinga data set generated after a major storm stuck the Adi-rondack Park in upstate New York. Application of ma-chine learning to these data provided no new informationbut reinforced many of the discoveries made using tra-ditional statistics. Dill (23) also used a machine-learningalgorithm to analyze the sale price of cattle sold at publicauction. The data set contained all information availableto a buyer on sale day and the price for which the animalwas sold. Using machine learning. Dill was able to de-termine which variables influence the buyer’s decisionand now will be able to generate an automated appraisalsystem from these results.

Research Needs—There are several problems asso-ciated with machine learning. One concerns data thatcontain random errors (i. e., ‘‘noisy’ data). Some ma-chine-learning algorithms are unable to handle this typeof data while others perform poorly (57). Much of thedata in agriculture is noisy. Another problem is that manyof the machine-learning algorithms require discrete data(e.g., classification-based) while agricultural data is mainlycontinuous (e. g.. numeric). A third problem is that ma-chine learning requires a complete database with asso-ciated outcomes from which to operate. Few of thesedatabases exist in agriculture.

Despite these limitations, machine learning can be avery valuable knowledge acquisition tool in certain sit-uations. With continued development, these limitationswill likely be overcome.

Knowledge-Based Report Generation

One of the initial goals in artificial intelligence was todevelop systems capable of translating documents fromone computer language to another ( I I ). An integral com-ponent of machine translation is developing a knowledgerepresentation of the original document such that text canbe generated in another language. Though machine trans-lation will not have a major impact on American agri-culture, systems that are able to generate knowledge-based reports from a database will.

Farmers receive large volumes of production data withlittle or no interpretation; hence, they may be unable toconvert these data into useful information. Knowledge-based report generation is an emerging technology thatcan provide them with interpretive reports to better sup-port management decisions.

In many respects, programs for knowledge-based re-port generation are similar to expert systems. Reportgeneration programs contain four components:

1. a domain-independent knowledge base of linguisticand grammar rules,

2. a domain database from which the report is to begenerated,

3. a domain knowledge base for interpreting the datastructure, and

4. the text planning component for deciding what tosay and how to say it (69).

Once a system is complete, the domain knowledge canbe removed to create a shell that can be used in anotherdomain. Report generation is still largely in the researchstages and commercial shells have not been made available.

CoGenTex, Inc. has developed a proprietary linguisticshell for knowledge-based report generation. This shellhas been used to generate weather forecasts in both Eng-lish and French for the Canadian Government. A USDASmall Business Innovation Research proposal has beensubmitted to study the suitability of this approach forgenerating knowledge-based reports that interpret DHIArecords for dairy farmers (46).

Research Needs— To date, there have been no appli-cations of knowledge-based report generation in agri-culture. Research should be directed at investigating thepotential benefit of this technology to American agri-culture. Once the preliminary investigations are com-

Chapter 4—Advanced Computer Tehnologies ● 109

pleted, a better understanding of needs and benefits willbe established.

Interfacing Technologies

Farmers have been slow to adopt personal computers.Recent surveys indicate that only 15 to 27 percent offarm managers utilize computers in management (1, 55).Two factors that may have contributed to this slow adop-tion rate are the lack of high quality management software(71 ) and a computer phobia on the part of some farmmanagers. Farm managers have available to them onlya limited selection of computer programs. most of whichperform similar functions. The computer phobia is causedby a lack of exposure to computers but is exacerbatedby the type of user interfaces (both hardware and soft-ware) employed by most agricultural computer programs.

Hardware Issues

Currently, most microcomputer systems use a key-board as the major input device. Keyboard entry is clumsyfor agricultural software as many farm managers are slowtypists. Even for programs that require little input. a‘ ‘hunt and peck’ typing ability can frustrate the user tothe point of not using the system. Another problem withkeyboard entry is impaired dexterity from excessivephysical labor or injury that severely impairs the farmmanager’s ability to type. Consequently, software shouldbe developed allowing the use of alternative input devices.

Two relatively common input devices are the mouseand the light pen. However, neither of these capture theuser’s natural pointing instincts (77 ). A more intuitiveinput device is the touch-sensitive screen. Another al-ternative input device is speech.

Touch-sensitive screens are computer displays in whichportions of the display may be used as an input device.This technology has been available since the mid-1960s(41 ). Touch-sensitive screens are easy to learn, very du-rable, and require no additional work space. At the sametime they have the disadvantage of increased cost, in-creased development complexity, lack of software to takeadvantage of touch-sensitive screens, arm fatigue. andscreen smudging. A major complaint of touch-sensitivescreen users is the lack of precision; however. high-precision screens have recently been developed (86). Dueto their durability and user-friendliness, touch-sensitivescreens have been used in specialized applications suchas kiosk information systems in shopping malls and air-ports and for order processing in restaurants. Both ofthese applications have been developed to allow controlof a computer systems by nontechnical users.

A second area of research aimed at improving thephysical link between the computer and user is speechrecognition. This research has been glamorized by sci-ence fiction movies such as 200 l: A Space Odyssey, inwhich computers carry on a dialogue with the user. Thoughthis is the goal of research efforts, it is not the currentstate-of-the-art (52). A prominent researcher has pre-dicted that totally spontaneous, unrestricted speech rec-ognition is still as much as 30 years from fruition ( 105).However, speech recognition appears to be suitable forapplications with restricted discourses. Agriculture is onesuch application.

Speech recognition is based on the ability to distin-guish between words and on natural-language processingwhereby natural language input is transformed into aform that the computer can utilize. In a common methodfor speech recognition, template matching, each spokenword is matched against a predetermined lexicon. Thelexicon must be trained to recognize a user’s voice. therebyresulting in a user-specific system (52). High-perfor-mance, speaker-independent, continuous-speech recog-nition systems use another approach. that of statisticalmodeling. Commercial speech recognition systems rangefrom speaker-dependent, single-word recognition (64-wordvocabulary units) to speaker-independent, continuous-word recognition (40.000-word vocabulary units ) (75).

Speech recognition is not a perfect function. Mostliterature values for recognition accuracy range from 95to 99 percent (97); some articles report 8 to 12 percenterror rates (61 ). Several factors affect the error rate; theseinclude presence of background noise. phonetic similar-ity of words, and mood of the user as he/she alters voicequality (52). Furthermore, lack of a one-to-one corre-spondence of sounds to words distinguishes speech fromother inputs. For instance, when a key is pressed on thekeyboard, the output is unambiguous. With speech rec-ognition, the output is the most likely output which cor-responds to the input. Consequently, the performance ofcurrent systems degrades (in both time and accuracy) asthe vocabulary increases. When speech input was com-pared to traditional input methods, it was found to requirethe same amount of time as mouse input, 80 percent asmuch time as a single key stroke and 48 percent as muchtime as full-word typed commands (61).

A commercial speech recognition system recently wasadded to a medical diagnostic system for clinical dataentry (88). The system was an isolated-word, speaker-dependent system capable of recognizing eight contin-uous syllables. Utterances required a half a second to


take effect and 90 percent of all utterances were recog-nized correctly. For this application, speech recognitionproved an effective interface for improving the accep-tance of the diagnostic system.

Advances in hardware input devices to improve theusability of computers are being driven by multiple non-agricultural sources. For example, speech recognition isa goal of the Department of Defense ( 105) and of researchaimed at providing more environmental control to thephysically disabled (20). Since this technology is domainindependent, advances in other domains should also greatlyfacilitate the use of speech recognition in agriculture dur-ing the next decade.

Software Issues

The software design of the user interface is the mainfactor determining the effort required both to learn andto use a computer program. The most important functionof the user interface is to match the needs of the user.Novice users need interfaces that are easy to learn whileadvanced users prefer interfaces that are easy to use. Mosteasy to learn systems are not convenient to use. Thus,no one interface will meet the needs of all computer users(33).

In general, agricultural software has not been distin-guished by sophisticated user interface designs. This partlyreflects the fact that most agricultural software is writtenby people who understand agriculture. yet have little orno training in user interface design.

Currently. there are nearly a dozen different interfacedesigns that can be used with computer programs. Theserange from command languages to natural language.

Two common user interface designs in agriculture arecommand and question/answer systems. A command-driven user interface is similar to the DOS system wherea series of commands and arguments have to be knownby the user. For example. in the Cornell Remote Man-agement System, which is used to ascess DHI data. acommand such as AIM 1-S1-DH1MO094 is used to runa report. This type of user interstice is easy for an expertto use, but because it is not intuitive. it is difficult tolearn. Another type of command-driven user interfacecan be designed by mapping commands to special keys.This interface is used by WordPerfect ( WordPerfect Corp..Orem. UT) which uses multiple combinations of theSHIFT. ALT. and CTRL keys with function keys for-specific commands. Question/answer systems require theuser to enter a response. If the type of response is un-ambiguous, this design can be easy to use but also te-

dious. This type of user interface should be limited toresponses which are Yes-No (e. g.. Y/N) or numeric.

A type of computer interface that is more intuitive touse than command and question/answer systems is nat-ural language. With this type of interface, commands aregiven in normal spoken or written language instead of aformal command language. An example of a natural–language user interface is one that converts natural-lan-guage commands to DOS commands. For example, thenatural-language command ‘‘show me the files on driveb:” is converted to the command “dir b:*.*” (53). An-other example of a natural-language interface is one thatwas developed for signal processing (68). This systemallows users who are knowledgeable about signal pro-cessing but ignorant of any programming languages tomanipulate wave forms using English commands ori-ented toward mathematical operations. However. the mostcommon use of natural-language interface has been indatabase querying systems.

Natural language is attractive to the casual user andto the user who is unwilling to learn a formal commandstructure. However, natural-language user interfaces re-quire more typing than command-language interfaces.As discussed previously, typing requirements are an im-portant consideration for agricultural software. There-fore, natural-language is probably not a desirable userinterface for systems that can be driven with a limitedset of commands (e. g.. DOS).

Another popular user interface design is the menu sys-tem. In the simplest form. a menu is a list of choices.The user selects one choice by entering a number or letter.Another version includes a light bar that can be positionedover the menu using the keyboard. A more sophisticatedmenu design, known as the graphic user interface (GUI),is the icon and mouse system. This type of system rep-resents menu selections using a picture that is ‘"clicked-on’ with a mouse. The icon system was first developedfor the Xerox ‘Star’ workstation (90) to reduce the learn-ing time of the user interface. The user is expected im-mediately to know which icon is appropriate. Thus. theicon must be unambiguous and realistic. Distinguishableand meaningful icons may be difficult to develop forseveral similar items (96). Accompanying text is oftenadded to clarify the meaning of possibly ambiguous icons.

Another major factor of the user interface is data entry.For this factor. interfaces called “form-filling” designshave been developed. The user is presented with a seriesof fields in which data are entered. The display relatesto a written form and allows the user to see all of thefields together. Often. form-filling interfaces have data


validation and editing capabilities. For more complexdata entry needs, multiple forms arranged as overlaidwindows can be used. As data are entered into a field,it actuates the next form which displays with the appro-priate related fields. This type of user interface is rapid,easy to use, and easy to learn (96).

Design of sophisticated user interfaces has advancedto a point where they should now be considered for allagricultural software. Proper attention to user interfacedesign issues can result in agricultural software that ismore acceptable to use. For example, adaptive interfacesare aimed at satisfying the differing needs of both noviceusers and experienced users. An adaptive user interfacedetermines the skills of the user and changes the interfaceto meet those skills. In general, novice users are providedwith menus and question-answer systems, while ad-vanced users are given the option to use command lan-guages and special key strokes. A prototype adaptiveinterface has recently been developed (SAUCI); (101)for processing UNIX commands. Using the adaptive in-terface, users made about half as many errors and re-quired less time to perform tasks. Research in adaptiveinterfaces should result in systems that are more intuitiveto use and easier to learn.

Information Retrieval Systems

Information retrieval systems are a set of advancedcomputer technologies for accessing stored information.These technologies differ from decision support systemsin that they offer no recommendations. Three technol-ogies are emerging that may have a role in Americanagriculture in the next decade. These are natural-languageinterfaces, full-text retrieval systems, and hypertextsystems.

Natural-Language Inter-aces—Maintaining a com-plete set of production records is a critical component offarm management. More important is the ability to rap-idly and flexibly access information for management de-cisions. The best method of accessing production recordshas been through database management systems; how-ever, these systems generally have inflexible retrievalfacilities based on menus that present options of data toretrieve or predefine reports to run. Traditional systemsrequire the user to learn the hierarchical structure of themenu system and limit the type of reports available. Anatural-language interface for querying a database canoffer a more flexible retrieval system (43).

The current generation of natural-language interfaceswas made possible by a set of linguistic theories devel-

oped by Chomsky ( 12). These theories were first imple-mented in an efficient algorithm in a natural-languageinterface for retrieving information about lunar rock broughtback from the Apollo space missions (LUNAR) (107).

LUNAR is based on a three-compartment model ofdata retrieval. The first compartment is syntax analysis,which determines the grammatical structure of the sen-tence. The second compartment of LUNAR is the se-mantic module, which is responsible for determining themeaning of the syntactic structures. The meaning is trans-lated to a formal query language in this module. Thethird module of LUNAR is the retrieval component. Thismodule executes the formal query language, based onthe semantic analysis, to retrieve data from the appro-priate database. When LUNAR was tested, it answered78 percent of the questions presented to it ( 107).

The purpose of developing LUNAR was to assist sci-entists in retrieving data on lunar rocks. Its users wereprimarily interested in specific data as that data relatedto other scientific information that had been collected.However, this style of data retrieval is not appropriatefor production agriculture where management decisionsneed to be made. A natural-language interface for re-trieval of data for decisionmaking should put the data inthe proper context so that an informed decision can bemade. Consequently, a knowledge-based, natural-lan-guage interface was developed to formulate more com-plete, intelligent answers to users’ questions from anagricultural database (IDEA) (44).

IDEA is based on the LUNAR three-compartment modelbut utilizes a new approach for semantic representation.Unlike the formal query language used in LUNAR, IDEArepresents the query through a set of domain concepts,which contain ‘‘expert’ information. IDEA has the ca-pability of responding to a query and offering additionalpertinent information. An example of a query and answeris shown in box 4-D.

IDEA was developed for a dairy database to assistfarm managers in decisionmaking. It is capable of re-sponding to several different types of queries. The sim-plest query is about a single cow (e.g., “When is 5000due to calve’?” or, simply, “Is 5000 pregnant’?”). Morecomplicated questions can be asked about subgroups ofcows (e.g., “Which daughters of Thor are bred to Bell’?”):averages (e. g., “What is the average calving interval forcows in the north barn’?’); and counts (e. g., ‘‘How manyheifers are due to calve in June’?”). Replies are designedto contain important information that the user may nothave known was in the database or may not have evenasked for.


Just as generic, domain-independent shells have givenexpert systems widespread use; for natural-language in-terfaces to be successfully used in agriculture, a genericnatural-language shell capable of being transported toother databases is needed. However, unlike expert sys-tems, development of a generic shell for natural-languageinterfaces has proven difficult. Hendrix and Walter (37)point out that full synchronization is needed between thedatabase management system and the natural-languageinterface. This is difficult to achieve when the natural-language interface is added as an afterthought. For ex-ample, in the dairy database that IDEA accesses, thereproductive status of a cow is given as a numeric value(e. g., O means ‘‘not bred’ ‘). To access multiple data-bases, the natural-language interface must be able totranslate all their representations.

TEAM is a natural-language interface developed atSRI and designed to be transportable between databases(32). However, a database expert still is needed to adaptthe system to each new database. The expert suppliesinformation describing the database and domain-specific

words, a process that TEAM has automated through aseries of questions. Another source of difficulty in mak-ing a natural-language interface transportable is associ-ating meaning to phrases. For example, consider thephrases “bred to Bell” and “bred in May. ” The firstrefers to the ‘‘service sire’ while the second refers tothe “breeding date. ” Most natural-language interfacescannot handle these kinds of subtleties. Thus, for a nat-ural-language interface to be successfully transported toa new database, a database expert and a linguistic expertare needed ( 19).

Because of the problems in developing generic shells,natural language commercially lags behind many otherartificial intelligence technologies (70). One of the fewgeneric natural-language interfaces available commer-cially is the “Intelligent Assistant” interface for Q&A(Symantic Corp., Cupertino. CA), which was introducedin 1985. This system differs from most transportablenatural-language interfaces in that it has its own databasesystem. Users build their application directly in Q&A.This system also uses synonyms for acquiring new words.

Box 4-D—An Example Query and Answer to a Natural-Language Interface>> which cows are due to calve next week?

4897 was bred to STARMAN on 12/15/1987and is due to calve on 09/18/1988Projected calving interval: 347 days

>> when did 5281 calve?

5281 calved on 05/26/1988 with a heifer calf - #5535The calf was in good conditionGestation length = 278 days5281 had a retained placenta

>> is 5239 pregnant?

YES5239 was bred to TOPBRASS on 03/20/1988and is due to calve on 12/19/1988Projected calving interval: 466 days

>> is 5449 pregnant?

NO5449 calved on 12/1 1/1987 with a heifer calf - #5478The calf was in good conditionGestation length = 283 days5449 is 282 days in milk5449 was bred to LEVI on 02/21/19885449 was pregnancy checked on 03/30/1988 and was open

SOURCE: L.R. Jones and S.L. Spahr, “IDEA: Intelligent Data Retrieval in English for Agriculture,” A/ Applications in NaturalResource Management 5(1)56, 1991.


An attractive feature of this system for agriculture is thatit operates on standard IBM-compatible microcomputers.Another commercial natural-language interface is Nat-ural Language (Natural Language, Inc., Berkeley, CA).This system interfaces with any database that supportsStructured Query Language (i.e.. SQL).

Full-Text Retrieval Systems—A relatively new areaof human-computer interfaces that holds great promisein making information more accessible is full-text re-trieval. The goal of a full-text retrieval system is to searcha collection of documents to find relevant informationfor the user (4). These systems can be particularly usefulfor accessing a collection of documents that are authoredby several different people who potentially use differentwords to express the same thing. Such a collection ofdocuments, including most Agricultural Extension pub-lications, is unedited and generally not indexed.

Blair and Maron (4) evaluated the effectiveness ofSTAIRS (STorage And Information Retrieval System). afull-text retrieval system developed by IBM. They foundit to retrieve less than 20 percent of documents relevantto a particular search when the database contained roughly350,000 pages of text. They identified several pitfallsthat need to be considered in developing full-text retrievalsystems. STAIRS was efficient at retrieving documentsthat exactly matched the wording of the request, but itperformed poorly in retrieving documents that containedmisspelled words, and words that were synonymous withthose in the request. For example, the word ‘‘gauge’was spelled 9‘guage’ in an original document, preventingits retrieval. Full-text retrieval systems must be able toaccount for such situations and retrieve relevant docu-ments whose text may not match the exact wording ofthe request. A simple key-word search or an indexingscheme thus does not meet the needs for full-text re-trieval.

A full-text retrieval system developed by Gauch andSmith (30) contains an expert system and a thesaurus.The thesaurus contains domain-specific information forwords, a list of synonyms for each word. its parent word(s),and a list of children words. This structure allows aparticular search to be generalized or narrowed. Deci-sions as to the search pattern are made by the expertsystem. If the recall is low, it will broaden the search.If the precision is low (i.e., too many irrelevant passagesare retrieved) the expert system will use a more specificsearch. The query is formed by the user and then passedto a full-text retrieval system that has immediate accessto any passage in the text. The retrieval system requiresthat the text undergo two stages of preprocessing. In the

first stage, the text is formatted for enhanced display.Formatting includes insertion of format marks (line. tab.italics, line, label ) and context information (section. par-agraph, sentence. item). In the second stage of prepro-cessing. the file is converted to fixed-length records tot-fast access. Consequently. the system does not operateon the original documents. This is an undesirable featureas it precludes searching subsets of documents and re-quires additional storage.

A full-text retrieval system now commercially avail-able (Metamorph; Thunderstone, Chesterland. OH) shouldhave wide application in agriculture. Metamorph oper-ates on standard ASCII files using natural-language quer-ies to search and find relevant passages in documents.The natural-language input undergoes morphologicalanalysis to normalize each word. The normalization pro-cess converts words to morphemes—the smallest mean-ingful unit of a word. A set of morphemes that are relatedto, but not necessarily synonymous with. the originalmorpheme is generated. Metamorph then correlates theseequivalence sets to textual passages to determine pas-sages that relate to the natural-language query. At thefirst level of search, an equivalence must be present inthe passage for its retrieval. If this is unsuccessful, Me-tamorph will broaden the search. Another important fea-ture of the correlation procedure is that it utilizes anapproximate match to account for minor discrepancies inspelling. These features fulfill the conditions Blair andMaron (4) identified as necessary for a full-text retrievalsystem.

Numerous applications of full-text retrieval are pos-sible. A recent project used a commercial full-text re-trieval system to assist users in querying a specific DHIcomputer manual (29). Additionally, with the advent ofmass storage systems for microcomputers (e. g., CD-ROM), full-text retrieval systems can play a significantrole in providing expert information (e. g., extension bul-letins) to county extension offices and directly to farmmanagers. An effort is underway to develop a nationaldairy database (39) consisting of full-text documents cov-ering major dairy-management areas. This full-text da-tabase is expected to be delivered on a CD-ROM andaccessed using a full-text retrieval system.

Hypertext-Hypertext is a method of connecting re-lated passages of text, graphics, animation, or computerprograms in a multidimensional {i. e., hypercube) fashionsuch that they can be accessed in a nonlinear fashion.Each node can be connected to any number of other nodesthat provide additional related information. Hypertextsystems are analogous to footnotes or references in a

114. A New Technological Era for American Agriculture

document. For example, a footnote contains additionalinformation related to the text. The reader determineswhen or if the footnote is to be read. Computerized hy-pertext systems are based on the same principle.

Hypertext systems are relatively easy to implement butare difficult to build. They require the locations of therelated text to be stored with the location of the originaltext. This is essentially a database management problem.The difficult part of a hypertext system is to establishthe appropriate links between and among nodes. Thisusually requires a domain expert, but the process can beautomated through full-text retrieval tools.

As Extension documents begin to be disseminated inelectronic form, hypertext should be considered as a methodof increasing access to related subject matter. For ex-ample, an extension bulletin that describes the use oflactation curves for herd management should be linkedto other bulletins describing the use of butterfat and pro-tein curves. To demonstrate the benefits of hypertext inan agricultural setting, Rauscher and Johnson (80) de-livered the six feature papers contained in an issue of AlApplications: Natural Resources, Agriculture, and En-vironmental Sciences in hypertext form.

Integrated Systems

Management of an agricultural enterprise requires avariety of decisions and, hence, a variety of decision-support tools. Long-range research in the area of human-computer interface will be directed at integrating variousdecision–support programs into a single system. Currentresearch is aimed at integrating autonomous systems,developing intelligent user-interface managers, and in-tegrating systems through a common representation sharedby an intelligent dialogue manager.

An overall controlling software system that allows theuser to access different decision-support tools yet main-tains operational independence of tools themselves rep-resents the lowest level of systems integration. The generaoperating system of a computer is an example in that itallows the user to access multiple programs in the sameenvironment. More advanced integrated systems assistthe user in choosing the decision-support tool and providelogical links between tools. This type of integration canalso be used to develop multimedia applications such asfull-color, full-screen graphics; full-color, full-screen video;aural delivery of speech or music; and animation (50).

An example of an advanced multimedia system forintegrating several different decision-support tools is theWhole Earth Decision Support System (WEDS; reference(51. The WEDS project combines textual databases,

expert systems, simulation models, traditional programsand laser-video images within the agricultural domaininto a single integrated system. Each module is developedindependently and inserted into WEDS. For example, anexpert system for lactation curve analysis developed in-dependently from WEDS can be incorporated and linkedwith other components dealing with lactation curves (e. g.,documents in the textual database). In this system, theuser moves between the different modules guided bylogical connections. Systems such as WEDS should beable to provide a complete information resource to ex-tension agents, agri-service personnel, and farm man-agers for solving problems and formulating managementdecisions. The multimedia approach utilized in the WEDSproject should be encouraged for systems developed inthe 1990s since people remember more if they combineseeing, hearing, and doing during the learning process(60).

A more tightly coupled method of integrating softwareis to link different systems through a user-interface man-ager. The user-interface manager controls all user-inter-face functions for a set of application software (96) andvalidates all inputs for the application software. Screendisplays, including error messages and on-line help, arealso controlled by the user-interface manager. There aretwo major advantages to integrating software in this fash-ion. First, a system does not need to be redeveloped foreach piece of application software. Second, the user isalways presented with a consistent interface; thus, as theuser moves from one application to another, the userinterface remains the same. This is important for ac-ceptability of software by laymen. Development of ageneric user-interface manager awaits further research;however, several fourth-generation languages include fa-cilities that can assist in development of generic userinterfaces (%).

A more advanced method of integrating software isthrough an intelligent user interface; such an interfaceallows problems to be formulated and appropriate ap-plication software selected using natural language. A pro-totype system for integrating crop production decision-support systems is under development (see figure 4-6);(59). It uses an intelligent dialogue manager (IDM) withunrestricted natural-language communication to developa problem description. The IDM parses input into a se-mantic representation using knowledge of the types ofqueries that can be asked and the lexical entities that canbe discussed. The IDM also utilizes a model for inferringthe goal of the user’s input and relating it to the contextof the overall dialogue. The semantic representation ispassed from the IDM to an expertise module dispatcher


(EMD), which selects the application to respond to thequery and formulates the appropriate control structure forthe application software. The EMD is an expert systemwith knowledge of the problem-solving abilities of eachapplication software module. This system can providethe user with a variety of problem-solving tools. Fur-thermore, the user does not need to know the nature ofthe software, the details for using it, or the situations forwhich it is appropriate.

Other Computer Technologies

Three other emerging computer-oriented technologieswill impact American agriculture in the 1990s. The firstinvolves dispersal of information to those who need it indifferent geographic localities. The second. robotics, willimpact the labor problems associated with agriculture.The third area is sensor technology.

Networks and Telecommunications

American agriculture is decentralized and widely dis-tributed, making information dissemination problematic.However, electronics can be used to provide mass dis-tribution of information. Electronic information can betransmitted essential y at the speed of light and duplicatedat minimal cost. Two electronic forms of informationdelivery will dominate in the 1990s: a satellite-basedsystem and a wide-area computer network.

Satellite transmission of data has become a common-day occurrence for telephone and other communications.

A geosynchronous satellite receives a transmission fromEarth and rebroadcasts that message back to Earth overa wide area. Different frequencies are used to send mul-tiple simultaneous messages. Two common modes oftransmission are the Ku and C bands.

Interest in delivering agricultural information via sat-ellite is growing. Several distance-learning programs havebeen developed at the University of Utah for delivery inEcuador ( 13). Their developers are also preparing anundergraduate animal breeding and genetics class to bedelivered over the national AG*SAT satellite instruc-tional network, which routinely carries Extension pro-grams. An Extension series of interactive dairy programshas been developed and delivered by the University ofWashington (8) as well as by the University of Wisconsin(35). The American Farm Bureau also maintains a sat-ellite link to 46 States and 573 of their county offices(72). This satellite link is used to transmit data as wellas instructional programs.

Satellites not only make possible mass distribution ofinformation, they do so in a way that makes this infor-mation easily accessible to end users. They only need asatellite reception disk and a television. However, de-velopment of satellite-based instruction programs can beexpensive. Poor planning may also reduce attendance.Other problems include limited audience interaction andlow motivation on the part of the end user to view theprogram. The importance of in-person interactions withthe live speaker should not be underestimated. However,

Figure 4-6—Functional Components of the Crop Production Expert Advisor System

Deep reasoning

User IDM I State Simulationrepresentation

Management Expert system modulesproblem situations : domain-specific

The problem solving system component

SOURCE: L.R. Maran, “CPEAS: The Crop Production Expert Advisor System,” Knowledge Based Systems Research Laboratory, Department of Agronomy,University of Illinois, Urbana-Champaign, 1989.


if funds for education continue to dwindle, this mayremain the only feasible means to conduct an Extensionprogram.

Another method of rapidly delivering information isthrough a wide-area computer network. Much of the west-ern world currently is criss-crossed with multiple computernetworks. Two of the original computer networks areBITNET (figure 4-7) and ARPANET. BITNET was ini-tiated at the City [University of New York and was used toconnect major educational institutions. ARPANET was ini-tiated by the Department of Defense. Today there are na-tional computer networks for the government. commercialcompanies, and educational institutions. A number of re-gional networks have also been developed. These includenetworks such as Clemson University Forestry and Agri-cultural Network, CNET (Cornell University). and PEN-pages (Pennsylvania State University). Most of-these networksinterface through the national Internet system so that mes-sages can be sent from one network to another. Internet isfunded by several government agencies and numerous com-panies (50).

The main benefit of wide-area computer networks isthe ability to rapidly share information and expertise. forinstance, an industry situation report can be posted onthe network and broadcast to all interested readers withaccess to the network. County Extension agents on thenetwork can send and receive files in electronic format.In this way. interdisciplinary work can be conducted overlong distances. Varner and Cady (103) have establisheda bulletin-board type system, called DAIRY-L, throughwhich dairy professionals can request and receive infor-mation. DA IRY-L is only one of hundreds of bulletin-board systems, but a pioneer in the use of networkingfor Extension education.

DAIRY-L, which resides on the University of Mary-land mainframe computer. was initiated early in 1990.Since that time subscription has grown to 150 subscribersfrom 37 states and 20 foreign countries (figure 4-8).Message traffic also has increased, approaching an av-erage of 15 messages per month (figure 4-9). Messagesare submitted to a ‘‘list server’ which in turn transmitsthem to all participants of DAIRY-L; therefore. all sub-

Figure 4-7—Topology of BITNET Connections in the United States

SOURCE: J.R. Lambert, ‘(Networks, Telecommunications and Multimedia Information Bases for Agricultural Decision Support, ” commissioned backgroundpaper prepared for the Office of Technology Assessment, Washington, DC, 1990.


Figure 4-8—States with Participants in DAIRY-L.

New Hampshire

I

SOURCE Mark Varner, University of Maryland (M.A. Varner and R A Cady.Dairy Science 74(Supp 1): 201, 1991

Figure 4-9—Volume of DAIRY-L Messages.

50 I

SOURCE Mark Varner, University of Maryland (M A Varner and R ACady. Dairy-L A New Concept in Technology Transfer forExtension, Journal of Dairy Science 74(Supp. 1 ) 201, 1991

Dairy-L” A New Concept in Technology Transfer for Extension, Journal of

DA IRY-L has proven extremely useful to extensionspecialists needing knowledge in areas outside their in-stitution’s expertise. Because all members see all mes-sages, DA IRY-L is also a powerful ediucational tool.

Information exchange through wide-area computernetworks makes efficient use of personnel and resources.Therefore. a high priority should be given to maintainingand enhancing the backbone systems (i. e.. satellites andwide-area computer networks ) that provide rapid dissem-ination of information. Since these systems are nationalin scope, this initiative should occur at the Federal levelwith USDA-ES providing the leadership in agriculture.

Robotics

Robotics are machines that can be programmed to per-form a variety of labor intensive tasks in agriculture.Since 1968, when strew Dutch companies proposedmechanisms similar to robotics for harvesting citrus, re-searchers have proceeded though the poposal stage andcurrently are testing Laboratory and field prototypes forfruit harvesting. transplanting, tissue culture propaga-tion. and machine guidance (67) (table 4- I ).


Figure 4-10—Volume of DAIRY-L Requests for Remote Retrieval of Text Files and Software

Number of requests

Date posted File/software o 10 20

3/21/913/22/914/22/914/22/914/22/915/09/916/19/916/25/916/25/916/25/917/18/918/29/918/29/918/29/91&29/91

Cow price spreadsheetPoison plant fact sheet

TMR fact sheet #1TMR fact sheet #2

TMR fact sheet #3FTP instructions

water coliform fact sheetHeat stress fact s. #1Heat stress fact S. #2

Heat stress fact s. #3Staff design fact sheetDairy producer survey

Antibiotic residue fact sheetSomatic cell count fact sheetSomatic cell count fact sheet

SOURCE: Mark Varner, University of Maryland (M.A. Varner and R A Cady, ‘Dairy-L A New Concept in Technology Transfer for Extension, Journal ofDairy Science 74(Supp. 1): 201, 1991.

Most robotic applications under development are for-eign-based. The United States is noticeably lacking indevelopment efforts. Japan and Europe have much strongerprograms and are likely to capitalize on this technologymuch sooner.

Agricultural robotics research is proceeding in two di-rections. One involves sensor technology (see followingsection) and machine vision. This is because, unlike pro-duction line robots, agricultural robots will operate in en-vironments where interferences will be encountered. Forexample, a fruit-harvesting robot must be able to locateirregularly shaped fruit despite the obscuring effect of leavesand stems. A second research concern is robot end-effecters(i.e., grippers). These are the mechanisms through whichrobots conduct their work. Again, unlike industrial oper-ations, agricultural robots will generally be working withfragile products (e.g., bedding plants and fruit). Touch andforce feedback are necessary to avoid bruising or damagingplants, fruits, or animal products.

Three other areas of research are important for robotdevelopment but are not specific to agriculture. Manip-ulators are the physical linkages that move the end-ef-fectors. Breakthroughs in speed and cost of manipulatorsare necessary. Agricultural robots will likely require lessprecision than industrial robots and will not require cur-vilinear motion, thus reducing the cost, Easily adoptedrobot components from nonagricultural applications wouldreduce the engineering costs of agricultural robots. A

second research area is the development of computeralgorithms for robot control. Significant advances in theminiaturization and integration of control hardware areneeded. Integral feedback of the robot’s position is es-sential. More powerful integrated circuit chips to inter-face sensors and to control the manipulators are alsoneeded. New artificial intelligence approaches to taskselection will be important facets of robot control re-search. A final area of research, systems simulation,allows evaluation of alternative robot configurationsthrough animated computer simulations. Advances incomputer simulation would reduce the development costand time required in engineering a robot.

One major use of robots in agriculture will be for labor-intensive tasks. For example, there are two Dutch com-panies developing robots to milk dairy cows; one pro-totype is operating at the University of Maryland. Labor-saving robots will enable American farmers to remaincompetitive in world markets despite higher labor costsand a shortage of part-time, seasonal labor. They willalso help to stem the flow of young, struggling industriessuch as ornamental horticulture, bedding plants, and planttissue cultures to countries with low-priced labor. If ro-botics can help these industries survive, they will createor maintain jobs which would otherwise be lost.

Another major use of robots will be to micromanagecrops. For example, a robot with an image sensor todetect weeds could be used to spot-spray herbicides. This


Table 4-l—A Partial Catalog of Research Applications of Robots in Agriculture

Application Location Notes

Fruit harvesting

Apple harvesting France Able to harvest 500/. of fruitCitrus fruit harvesting University of Florida 1 fruit every 3 seconds, able to harvest

a fruit on 750/0 of its attemptsTomato harvesting Kyoto University 20 seconds per fruitCucumbers harvesting Japan In a laboratory study, the hand

successfully completed theharvesting motion for 42 of 53cucumbers.

Muskmelon harvesting Purdue University 5 seconds per fruitVolcani Institute, Israel

Plant material sensing and handling

Transplanting● pepper plants Louisiana State University Transplanting rates as low as 1 plant● marigolds and tomatoes Purdue University every 3 seconds have been achieved● move plugs from one flat to another Rutgers University with a 95°/0 success rate.Automated tissue propagation University of Georgia, Operations include retrieving the

University of Florida, cuttings from a conveyor, trimming toUniversity of Illinois, size, stripping selected petioles,New Zealand, Europe, applying rooting hormones, andIsrael, Japan, Switzerland sticking the finished product into a

plug flat cell.Mushroom harvester England Uses a vision system to locate and size

mushrooms and guide a selectiverobot harvester.

Forest thinning Performs automatically selective fellingwithin the tree ranks, bunching theharvested trees and carrying them toa process zone.

Animal

Robot milkers NetherlandsSheep shearing AustraliaEgg handling University of California. Facilitated candle inspection,

DavisPork protein sensing Purdue University Robot moves an electro-magnetic

scanner over a carcass.Pork carcass sectioning SwedenOyster shucking University of Maryland Machine vision application to locate

oyster hinges.

Machine guidance

Automated guided vehicles Michigan State University Based on machine vision sensing.Texas A&M University

Plowing robot FranceRice combine Japan Used edge-following to guide the

machine around a rectangular field.Direct spot spraying Purdue University Machine vision application to recognize

Corn detasselingplants.

Purdue University Machine vision application to recognizeplants

SOURCE Office of Technology Assessment. 1992

would encourage farmers to adopt conservation tillage robots to perform their tasks. Reliable sensors coupledand post-emergence spray programs. with knowledge-basccl decision support systems will pro-

vide important managenment tools.Sensor Technology

120. A New Technological Era for American Agriculture

Photo credits: Norman Pruitt, Maryland Agricultural Experiment Stat/on.

This research prototype automated milking system, developed in the Netherlands, allows scientists to studysystem automation and robotics that can benefit dairy farms.


or that require more vigorous sensing than we can pro-vide. Sensor technology provides information the humansenses cannot access.

There generally are six classes of sensors. The newestis machine vision which processes images (e. g., camerainput ) to detect patterns. Nuclear magnetic resonance(NMR) is a noninvasive technique of resonating high-frequency electromagnetic radiation in the presence ofhydrogen nuclei. This technology is widely used for di-agnosis in the medical field, but it is costly and difficultto apply in field situations. Neur-infrared (N/R) spec-troscopy is another noninvasive technique that measuresthe reflectance of NIR radiation by a substance. Because

organic compounds absorb and reflect NIR radiation dif-ferently this is a quantitative sensor. Acoustical mea-surements provide another class of sensors for measuringthe density of substances. Biosensors are sensors thatincorporate a biologically sensitive material (e. g.. im-mobilized enzyme). Electrical sensors can monitor theelectrical properties (e. g., conductance) of a substance.

Considerable work has been done in environmentalsensing (i. e., crops, weather), somewhat less in animalsensing (i. e., estrus detection) (40). A partial list of re-search efforts in sensor technology is presented in table4-2. Animal sensors are difficult to engineer due to bio-compatability problems and animal welfare constraints.

Photo credit: U.S. Department of Agriculture, Agricultural Research Service.

Drawing of pig (left) shows where cross section was made by magnetic resonance (MR) imaging. Spine, loin muscles, andkidneys are visible in upper part of MR image (right). Scientists can measure fat development under the skin

quickly without injury to the pig.

Table 4-2—A Partial Catalog of Research Applications of Sensors in Agriculture

Application Type of sensor

Electronic navigation system Used the Global Position Satellite SystemAutomated plowing system Photodetectors sensed the furrow edgeTractor guidance Computer visionMonitor organic matter in soil Light and NIR reflectanceApplication of spray material Electronic surface gridMonitor gaseous ammonia NIR spectroscopyMoisture sensors for irrigation Electrical resistancePlant stress Infrared leaf temperature sensorCrop growth Spectral reflectanceWeed identification Machine visionIdentification of plant embryo shapes Machine visionAnimal digestive system Radionuclide imagingEstrus detection Electrical conductivitySex determination of baby chicks Machine vision

SOURCE: Off Ice of Technology Assessment, 1992


Research on sensors for use in crop production generallyfocuses on the following objectives:

●

●

●

●

●

Improving operations in crop production by ma-chine guidance systems.Applying pesticide and fertilizer chemicals.Improving the management of irrigation water toconserve the resource and reduce production costs.Developing methods of monitoring crop growth toincorporate with computer models for improvingday-to-day crop management and strategic plan-ning.Developing sensors for assessing crop maturity andfruit location as basis for mechanical harvesting.

There remain numerous agricultural areas where sen-sors need to be developed (40). Doing so will require a

multidisciplinary approach with input from professionalswho understand the biology of the system in question aswell as professionals who understand sensor technology(e. g., engineers and physical scientists). Some of theareas that need to be addressed include:

●

●

●

●

Accurate three-dimensional fruit location sensor forcrop canopies. This will facilitate robotic fruit har-vesting.High-resolution navigation for field machines. Abil-ity to program machine locations within inches, notseveral feet, is needed.A chemical drift sensor to monitor fertilizer andpesticide application and production of air pollutinggases from animal units.Irrigation demand sensors that are not affected bysoil properties and climatic factors.

Photo credit: Gerald Isaacs, University of Florida

An experimental fruit picking robot uses a machinevision sensor and a computer to locate individualfruit for detachment. Approximately 3 seconds per

fruit are required.

●

●

●

Animal stress sensors that can remotely detect earlyanimal health problems.A fruit-ripeness sensor that can determine optimumharvest times and detect early stages of fruit andvegetable deterioration.Microbial sensors that can detect early developmentof spoilage or bacterial contamination in fresh meats,including poultry and seafood.

Photo credit: U.S. Department of Agriculture,Agricultural Research Service.

Animal physiologists test a sensor that will detect whenthis cow is ready to give birth.

An important component of the use of sensors in an-imal agriculture is telemetric data transfer and electronicidentification of animals. For sensors that are to be im-planted (e.g., tissue conductivity for estrus detection),telemetric data transfer must be accomplished within thesize constraints which make implantation feasible. Thisremains a research issue. Implantable electronic identi-fication systems have been developed and are currentlyunder review by the Food and Drug Administration. Con-cern centers on the possibility that implantable sensorsor identification units can enter the food chain.

The development of sensors will facilitate more formsof automatic control over various aspects of agriculturalproduction. The development of robots is closely tied tosuccess in the area of sensor technologies. A broaderimplication of sensor technology may be to provide adata acquisition system and a database from which de-cision support systems can operate. This should result intighter controls for management and higher profitabilityfor the enterprise. Another important impact of sensortechnology will be in the food safety arena. Sensors todetect food spoilage or contamination will greatly in-crease the safety of the American food supply.

SUMMARY/PROGNOSISComputer technologies change at such a rapid pace

that it is difficult to foresee their application in the nextdecade accurately. Irrespective of agricultural policics.computer technologies will continue to advance to sup-port the needs of other industries. Meanwhile. a numberof impendiments exist that are likely to slow adoption ofthese technologies in agriculture. These impediments canbe removed through changes in policy. Most projectionsof agricultural application of computer technologies havebeen overly optimistic. For example, Holts futuristicview of the application of computer technoolgies for farmmanagement (38 ) is still 20 years from fruition.

OTA has developed a scenario for the application ofcomputer technologies in agriculture assuming that newtechnologies have a 5-year development phase. That isto say that once a research project begins it takes 5 yearsbefore that technology is applied. It was also assumedthat incentives to bring new computer technologies outof- the research laboratory and into production agriculturewould exist. There are almost no incentives to do sotoday . Thus. American agriculture will not be affectedby these technologies in a major way for at least10 years.

The Current State

By and large. computers have had little impact onproduction agriculture to date. Predictions that every farmerwould own a computer by 1990 have not come true. FeW

farmers have computers and those who do use them pri-marily for bookkeeping and routine calculations (e. g..ration balancing ).

Computers have had somewhat more impact on agri-culture support industries. Using computer networks andtracking systems. equipment dealers are better able toprovide faster service and feed dealers are better able tomanage feed inventories. Most of these advances havecome from directly adopting general business softwarewith little or no input from the agricultural academiccommunity.

Another technology that currently is being adopted byfarmers is fax machines. This allows for rapid exchangeof printed material. An example of the use of- this tech-nology is in ration balancing. A nutritionist can receivethe results of a feed analysis by fax from the laboratory.formulate a ration. and fax that to the farmer all with-ina few minutes. There is limited use of networks for ex-change of information among Extension personnel ( i .e ,Dairy-L) and among protoype full-text databases (i. e.,National Diary Database).

Mid-1990s

Within the next few years. many technologies cur-rently under development should find their way into ap-

plication. By the mid 1990s, the performance ofmicrocomputers will likely double, eroding some of thecurrent constraints to farmer adoption of computer technology. However. it still is unlikely that a high proportionof farmers will own a personal computer by that time.

The primary application of advanced computer tech-nology in the mid-1990s will be in the form of ad hocexpert systems to solve well-defined problems. Thesewill be primarily problem diagnosis expert systems thatare currently under development. Farmers will have acadre ofexpert systems at their disposal to diagnosediseases and to evaluate animal and crop performance.These systems will generally not be integrated with eachother and each will condisider one aspect of a problem.Integrated systems that solve producton problems whileconsidering economic consequences will not becomeavailable until later in the decade.

The primary use of expert systems within the next 5years may be by agribusiness personnel, as they will beable to leverage the cost of’ adopting these technologiesacross more farms. Using expert systems to provide ad-ditional service to farmers may cause a shift in the roleof some professionals. For example. expert systemshelp veterinarians take an epidemiological approach tosolving problems (85 ). It will also cause some diversi-fication in services provided. For example. nutritionistsmay be more likely to become involved in consulting forthe crop program when armed with an expertsystem.

Sensors will see limited application for collecting real-time data for expert systems. The primary use of sensorswill be for monitoring weather and field conditions forcrop management. Expert systems will help farmers tointerpret these data and suggest appropriate managementstrategies such as irrigation, fertilization, or pesticidetreatment.

Another technology likely to see application within thenext 5 years is full-text retrieval systems. It will be pos-sible for farmers and Extension personnel to have a CD-ROM with all of- the latest publications at their fingertips.Using a full-text retrieval system they will be able toretrieve pertinent information that will help them makebetter decisions. For example. when a farm experiencesa corn mycotoxin problem, the manager can access aninformation base to find relevant literature. Large infor-mation bases, such as the national dairy database, willlikey be developed and delivered by 1995.


Robots for highly specialized, labor-intensive taskswill begin to be applied to agriculture in the late 1990s.This would include robot transplanting of seedlings andpork carcass sectioning. Robots for milking cows couldreach application by the mid- 1990s.

2000The turn of the century should bring with it significant

new applications of computer technologies in Americanagriculture. Ten years will provide sufficient time for theacceptance by farmers of computer technologies as avalid management tool and for the development of in-tegrated management programs. It will also allow timefor universities to become comfortable with these tech-nologies and for personnel to be properly trained in de-veloping these technologies.

By 2000, whole-farm advisors, or integrated “man-agement workstations, should be developed. A man-agement workstation will consist of integrated decisionsupport tools with a multimedia presentation of infor-mation. The workstation can thus serve as a diagnostictool, an information source, an advisor, and a planningsystem. The expert systems will consider the holisticview of an enterprise when making recommendations.The systems will also share data so that information usedin one system will be available to other systems. Thisgeneration of expert systems should operate as monitorsthat can alert producers to potential problems, as opposedto current expert systems which are situation-driven: thatis, the producer must perceive a problem and decide toexecute the system. The management workstation willalso contain an advanced user interface consisting ofspeech recognition and touch-sensitive screens.

The future dairy management workstation might con-tain decision support systems that monitor the financialrecords, the herd production records and the crop pro-duction records. Cropping decisions would be integratedwith the dairy needs, the financial situation, and the landresources available. Currently, these decisions are allmade independently. When the farmer is alerted to aproblem (e. g., pest infestation). he or she can use themultimedia features of the workstation to retrieve videosegments to learn how to identify the pest and the propertechniques for applying a pesticide.

Robots for harvesting fruits and vegetables and forautomatically guided vehicles should become availableby 2000. Their application will depend on the cost as-sociated with using human labor for the same job.

1,

2,

3. 4

4.

5.

6.

7.

8.

9.

1 ().

I 1.

12.

13.

14.

15.

16.

CHAPTER 4 REFERENCESBatte, M. T., Jones. E.. and Schnitkey, G. D.,‘bComputer Use by Ohio Commercial Farmers, ”American Agricultural E(onomic.v Asso(’iation72:935, 1990.Beck, H. W., Jones, P., and Jones, J. W., “SOY-BUG: An Expert System foe Soybean Insect PestManagement, Agrit,u[[[ir(ll S>,j[(,mif 30:269_~~6,1989.Berger, R.. “WHAM: A Wheat Modeling ExpertSystem. User Manual, Technical Report Depart-ment of Computer Science, University of Mel-bourne, Australia. 1987.Blair. D.C. and Maron, M. E., b’An Evaluation ofRetrieval Effectiveness for a Full-Text Docun~ent-Retrieval System. ’ C(Jttlt~ll(tli(~ltiotl.~ (!/’ Ihe ACI1428(3):289. 1985.Blancard, D., Bonnet. A., and Coleno. A., “TOM:Un Systeme Expert en Maladies des Tomatoes.P.H.M. RCIIIC Holti(olc. 261 :7– 13. 1986.Buick, R.D. et al.. “CROPS, A Crop RotationPlanning System for Management ot Whole-FarnlResources Implementing Low-Input SustainableAgriculture. Al Appli(wti~m.~ 6: 1992.Buxton, R., ‘6Modelling Uncertainty in ExpertSystems. ltlt(’~.}t(ltiot~[[l J()[{rnal l)t’~(itl-~(1(.llitl(~Studies 3 I:4 15, 1989.Cady. R.A. et al., ‘‘The Role of Televideoconfer-ences in Extension Programs. JoI/rnal (f L)t/ir\’Stienfc 74( Supp. 1 ):290. 1991.Carruthers, R.1.. Larkin, T. S.. and Soper. R. S.,b’Simulation of Insect Disease Dynamics: An Ap-plication of SERB to a Rangeland Ecosystem. .

Sim[(l[ltiotl 5 l: 101, 1988.Chang. W. and Jones. L. R., “An Object-0 -ientedApproach for Modeling Milking Parlor Perfor-mance,” Journal oj’Dai\> Scien(c 74( Supp. 1 ):236.I 991.Charniak, E. and McDernlott. D.. ‘ ‘Introductionto Artificial lntell igence’” (Reading. MA: AddisonWesley Publishing Co., 1985 ).Chomsky, N., “Syntactic structures. ” Mouton, TheHague, 1957.Christensen, C. et al.. ‘ ‘Preparing and 1nlple-menting Distance Learning by Utah State Univer-sity, “ Journal ~?j”Dait:} Siictlie 74( Supp. I ):289,199 I .Citrenbaum. R.. Geissmtm. J. R., and Shultz, R.“Selecting u Shell, ” Al Appli(wtion.s in N[lt[iralRe.~our(c klanagctnen~ 2( 1 ):3. 1988.Conlin. B.J. et al.. “Machine XPERT. ’ Proceed-ings of the National Mastitis Council. 1989.Conlin. B .J., Stcuernagel. G.R. and Peters ,R.R. ,“Design of an Expert System for ‘rr(~ublcShooting Dairy Hml Mtinugernent. ’ J(wt-tra/ {J/”L)(/it-\’ .Stietl(c 72( Supp. I ):461. 1989.

17.18.

19.

20.

21.

7?---

23.

24.

25.

26.

?7.

28.

29.

30.

31.

Conhn. personal conlnlunications. Sept. 17. 1991.Crosby’. C.J. tind Cluphwn. W’. M.. ‘‘A SimultitionModeling Tool for Ni(rogcn D~mmlics Using Ob-ject-oriented Programming. Al App/ittl(i(~ll.\4(2):94- 100. 1990.Dtimerau. F. J.. ‘ ‘Prospects for Knomlcdgc-BusedCustt~nlizati~~n ot Natural Language Query S~Is-ten ls . ‘ ‘ 111/i~rttl{[ti{jtl Pro~c.~.ii\l(q & kl~llllltqelllcllt24:65 1. 1988.Damper. R.].. ‘ ‘Voice-input Aids For The Ph~si-cully Disublcd. ’ l~lt[~l”}~[iti~~}l~~l JtMIrIItIl {J/ kfLlII-M{[t}linc .%tl{dic.v 21 :54 1. 1984.Dicttcrich, T. G.. b ‘Mtichine Learn ing. ’ AIII?IItI/Rc\ic\\ ~J/’ C’(wIp[IttIr .$~iril~c 4:255, 1990.Diikhuizcn, A.A. and Huirrw. R. B. M.. “’CHESS—An Integrated Decision Support Sjstcm T() Antilyr/cIndividual Ct~\*-Herd Pcrt\mnancc. 1lll(’<qr(ll(’dDt’ci.~i(~il S[(pp(Jr( .%>.$tctlt.~ in t!,ql-icl~ltl{t-t’-.$ll[-tv.~sfi[l Pr[i~tital Appli~wiio\t.~. Ikutschc Land-sw i rt hsc h afts-Ge se 11 sch~ fts ( cd. ). Fran k fu rt,Germany. 1990. p. 221.Dill. D. E.. ‘‘An Electronic Model for DeterminingValue and Predicting Stile Price of Holstein CattleSold at Public Auction. Ph. D. Thesis. Universityof Illinois. Urbtina. 1990.Dologite. D. G.. ‘ ‘Developing a Knowrledyc-BasedSystem on u Personal Computer Using an ExpertS~stenl Shell.’” J(H{I-HIII (f .$>st(ws M(iml<qcmcnt30, 1987.Eckhousc, R.. “opcrtiting En\’ironnwnts for the80386, ’ ‘ IEEE Compl(tcr 8:82, 1988.Evans. M., Mtmdor. R.. A Flaten. D. b‘ ExpertSystems and Farm Mm.igement. (’[llwdiall ./(mr-l~[tl of” Agri~l[lll{r~il E((Jtl~Jl)li(”.~ 37:639. 1989.Fcrrntinian. T“. W.. Michtilski. R. S.. and Ktitz. B..‘ bAgAssistant—A Tool For Dclreloping ExpertS)stcms. KII()\\led%qe B{~.\e(l .Y>.itcnl.$ i)! Agri(l{l -tl{i-[~—Pl”().~]~t’c”t.~ ,fiw Appli(wtim. Jleutsche Land-w’ifischafts-Gesellschaft (cd. ). Frtinkfurt. Gemumyz.1988. p. 252.Flinn, P.W. and Hagstrum, D.W. ,“Stored Gra inAdvisor: A Knowledge-Based System for Mtin-agement of Insect Pests of Stored Grain. Al Ap-plitwti~)]t.~ ill Natliral Re.WH(r(aC kfanllgo~~l~)llt ~(~ ):~~.1990.Fourdraine, R. H.. Tomaszewski. M. A.. and Tay-lor. J. F.. ‘‘Using a Natural Language Interface toAssist DH1 Supervisors, ” J~)[{rnal (!/’D{lil-> .~~ic}l(’c73(supp. 1 ):209, 1990.Gauch, S. and Smith. J. B.. “Intelligent Search ofFull-Text Databases, ” TR87-035, Department ofComputer Science. University of North Carolina.Chapel Hill, 1987.Goodell et al., CALEX Cotton: An Integrated Ex-pert System for Crop Production and Managementin California Cotton, Cal{/iu-nitl A,qriculfut-e 44:(5 ):I 8–20. 1990.

~-). -.

33.

M.

35.

36.

37.

38.

39.

40.

41.

42.

43.

44.

45.

46

47.

48.

Grosz. B.J. et al.. . ‘TEAM: An Experiment in theDesign of Transportable Nuturtil-Ltinguugc lntm_-faces,. Arti/i~i[ll lntclli<qctl~c 32: 173. \9%7.Grudin. J.. ‘‘The Cuse Agtiinst Uwr Intcrtticc Con-sistence). (’{~t~lt~lltllic[~tiott.~ of t~le A(’M 32:1 164.1989.Hunsen. J. P.. ‘ “DAIRY—an integrated decisionsupport sjstcm for planning of tied suppl) and larduse on the duirj’ tarm. Illtc,grdted Dc~i.$i~jtl .Sllp-port S>’.stt~ttl.s ij~ A<gric[{ltl{r[’: .5[{c~e.\.~/i[l Practi~[llApplit[~ti(w.s. Deutschc Lundwirtschaits-Gesells-chat’t ( ed. ). Frtinkfurt. Germany. 1990. p. 182.

Hat-wood. E.Il.. Smith. T. R.. and Conlin. B. J..“Devek~pment and Production ot’ Daig-LIVE Vidct~Teleconference Programming. J~~I(rIl(I/ ~j/’ D(iirjSticll(c 74( Supp. 1 ):201. 199 I .Hearn. A. B.. SIRTAC: ‘‘A llccisi(~n Suppt~rt S}Is-tern for Cotton Management. Rc\’ic\t of Markct-i)l,q aIId A,qri~[{l(l(ral lic(ulot]li~.~ 55( 2 ): 170-173,1987.Hendrix. G.G. and Walter. B. A.. “The IntelligentAssistant. BYTE Deccnlber:251. 1987.Ho]t. D. A.. b’Conlputcrs in Production ,4gricul-t u r e . ~~’~.i~,]l~,[~ ~~~:~~~ . I 985.Hutiens. M.F. und Eastwood. B. R.. ‘‘An Updateon the National Dair\r Databtiw Program.. J()[IrIlal~?/” D~/it-\’ .St”iclr(e 7 l~Supp. I ): 144. 1988.lsaacs, G.W. . .Sensor Tcchnologj’. ” conlmis-sioned background paper preptirecl for the Ofticeof Technology Assessment. Washington. DC. 1991.Johnson. E. A.. ‘‘Touch Displays, A ProgrammedMan-Machine Interface. ” ~-i.,q(~tlot?li(,j 10:271. 1967.Jones, L. R., personul communication. Sept. 17.199 i .Jones, L. R.. ‘‘Emerging Agricultural Technologiesin Human-Conlputer Interactions and Their impacton American Agriculture, commissioned back-ground paper prepared for the Office of Technology~ssessment, Washington, DC. 1991.Jones, L.R. and Spahr, S. L.. “IDEA: IntelligentData Retrieval in English for Agriculture. ” Al Ap-plicwtiotls itl Natl{ral Re.w)urtv Mana,qemcnt 5( I )56.1991.Kalter. R.J. and Skidrnore. A. L.. “DAIRYPERT:An Expert Systems Approach for Improving Dair~rFarm Management and Assisting Technical Trans-f e r , A.E. Res., Department of Agricultural Eco-nomics, Cornell University, Ithaca, NY, 1991.Kittridge. R. I., Proposal: “An Artificial intelli-gence Approach to Providing Herd ManagementInformation. “’CoGenTex. Inc.. Ithaca. NY, IQ91 .Klahr. P., McArthur, D. and Narain, S.. ‘ bSWIRL:An Object-Oriented Air Battle Simulator.” Proc.~[1 Al~jl. Nllt. Ccj}lf; OH Artif; Intel.. Pittsburgh. PA.1982, p. 331. ‘Klahr, P. et al.. “TWIRL: Tactical Warfare in theROSS Language, ” Rand Corp. R-3 158-AF, 1984.


49.

50.

51,

~?. -.

53.

5-I.

55.

56.

57.

58.

59.

60.

61.

62.

63.

64.

Kline. D.E. et u]., ‘‘Machincr} Selection UsingExpert Systems and linear Programming. CcJnl-p[(ter,j (iItd Elc(.troili(.s ill A,qri([(lt[tre 3 :45.1988.Lambert. J. I?.. ‘“Networks, Tclcc~~l~lnlunic;ltit~nsiind Multi nwdi~l lnt~)rnlatit~n Buses for AgriculturtilDecision Support ,“ cwmm issit~ncd background pa-per prcpwui for the OIficc 01” Tcchnolog) Assess-ment. W:ishingtt~n. DC. 1 9 9 1 .L~inlbcrt, J. Il. :Ind Deter. R. S.. ‘ ‘Whole Earth De-cision Supp(wt Sjstenl. J{)l(I.IILII ofl>air]’ .Scicltcc7Nsuppl I ):255. 1990.Ltinduu. J .A.. Norwich. K. H.. :md E\ri.ins. S. J.,‘‘Autom;itic Speech Rcc(~gnition-Can It ]mprokcthe Man-M:lchinc lntcrt’:icc in Mcdiciil Expert Sys-tcms’!’ lltt[’1.lttltit~}~ill J(Jl(rn~Il ot’[)iol]lt’tli~[ll” C’OIII-pl{[it?,q 24: I I I . I 989.Lunc. A.. . ‘DOS in English.’. BYTE Dccenl-bcr:261. 1987.Luyer. tl. A., ‘‘Microcomputer History and Prc-history-An Archueologictil Beginning, ’ Anncl/.Y()/” tht~ Hi.~i(jr>’ ()/’ Cl)\ttpl{ti~l(q 1 l; 127. 1989.Luzw-us, W.F. and Smith, T. I-l.. .’A&~ption ofComputers tind Consulting Su-\’ices by New YorkDtiiD’ F:mmxs.’ J(mrIILIl (!/’D~~i\;\ .hiemv 7 I; I(j(j7,1988.Lcmmon, H .E., ‘ bCOMAX: An Expert System forCrop Mantigcment. ’ S<-iellce 223:29-33. 1986.Licpins, G.. Gocltz. R.. tind Rush. R.. b’MachineLc~irning Techniques t-or Nuturti] Resource Ihtu.Anu]ysis, Al Applit “[lti(~n.~ ill Natl(ral Rc.~ol(rccIvl(ln(i,qmen( 4(3):9. 1990.Makclu. M. E.. Vins(ln. S. B.. tmd Stone. N. D..‘‘ f+t)st-Pw-asitoid Pt)pulation Dynitmius in a Het-erogeneous Enl’ir(Jnnlcnt. in Pr(M’. .$(x’. [’~)IH-putcr Si\ttl{l[iti~)\t ,Mllltit{~\l/i~l.t’tl~[’ oil Art~/i~i[illit(clli,qctt~c (Ind .~inllll~iti(m: TiIc [~i~vr.~it> (){” Ap -plitatio~t.$, Sun Diego. C,A. Feb. 3-5 . 1988. p p .n+~j~ .Mm-an, L. R., ‘‘CPEAS: The Crop Production Ex-pert Acl\iw}r S}’stcm. ’ Knowltxlge Bused Sjstenl\Resciirch l.uborut~~r), Dcp:u-tment of Agronoml,Univcr\it) t~t- lllim~is. Llrb:illu-Chai~lptiign, 1989.Mw-cus. A. :md vun D:lnl. A., ‘ bUser lntcrfuccDciclopmcnts t’t)r the Nineties.’. IEEE C(N~/pI(tcrScptcmber:49. 1991.Martin. G. L., “The Utilit~r of Spetch Input in Uscr-Computcr ln[crt-:lccs.. ltltt~)-ltil~iolltll J(NII-IW! @’M~(t~-Machi}lc .7tl({iie~ 30:355, 1989.h4cArthur. D., Kl~lhr, P.. and Narain, S., ROSS:An Object-oriented Language for (lmstructingSinlul~it(ws. RLmd Ct~rp. R-3160-AF: 1984.,M~.Grunn. J .iM.. Kurkt}sh. K., and F:dconcr. L..A~ri~u]tur~l] Financiill Anulysis Expert S!’stcnls.Texas A~!riuu]tur:l] Experiment Stution, ‘l-CXUS A&h4(-,lni~~r~it~ . (’(~]lcgc St;ltion. TX. 1989.McKinitm, J.ltl :Ind l.enlmtm. H. E.. ‘ Expert S~\-

65.

66.

67.

68.

69.

70.

71.

7~

73.

74.

75.

76.

77.

78.

79.

80.

terns for Agriculture, Computers und Electronicsin Agriculture I :3 1. 1985.Michalski, R. S., b’Lem-ning by Being Told undLearning From Examples: An Experimental Conl-parison of’ the TWO Methods of Knowledge .Ac-quisition in the Context ot Dcvelt~ping tin ExpertSystem for Soybeun Disease Diagnosis. /nter-mltiont{l J{~lIrnLIl ()/’ Poli~~” Anal~’.ji.s tIIId ln/i~rljI~I-Iion S~.\tem.\ 4: 125. 1980.Michalski. R. S.. Curbonel]. J. G.. and Mitchell.T. M.. kla~.hinc Ltvirnincq. AI1 Arti/i~i~~l lntclli<gen~eAppr(MIfh (Lt}s Altos, CA: Morgan Kuutr-mmn Pub-lishers, Inc.). 1983.Miles. G. E., ‘ ‘Robotics tind Intelligent Mti-chines. commissioned b:wkground puper pre-pared for the Office of Technology Assessment.Washington. DC. 1991.Morris, D.T. and Asumu. [I. E.. “intelligent Nat-ural Ltinyuagc Intcrt.ace for u Signal Proces s ingSystem. IEE P}-{)f([~tlitl,q.\ I 37(5):37 I. 1990.M}kowiccka, A.. ‘ .Text Pl:mning-How T~J MakeComputers T~llk in Nutural-Language, ’ /iItcrIM-tioni{l JOIII.11111 ~)[” l-liil~-kl~i~’llijje Studie.$ 34:575,1991.Newquist. H .P.. A Comprehensii’e Annual Reporton the Artificial Intel I igence Industry, DM Datu,Inc., Scottsdtile, AZ, 1987.Novuk, J.L. and Stcgelin, F.Ii.. “ U . S . Fiirmer’sAttitudes Toward Microcomputer Uses. ” C~Mn-pllter.s and Electr{)ttic.~ in A,qri~llltllre ..3:81, 1991.Olson. K. E., ‘‘Stitellitc Technt~logy-A Farm Or-gtinization Perspective, J~)l(rnal {!/’ Daij-> .Y~icn~.c74(supp. I ):290. 1991.Oltenticu. PA., Fcrguson, J.Il., and Lcdnor, A. J.,“A Data-Driven Expert System to Evaluate Man-agement. En\rironnlentiil and Cow Fuctor lntluenc-ing Reproducti\c Eff iciency in Dairy Iierds.Jol~rIlttl (!fDait;\ .~cietl~c 7.3( Supp. l): 145, 1990.Oltlen, J. W. et al.. .9 Integrated Expert System forCulling M:intigement t~t’ Beet COW’ S,’ Cfvlrplit~~r.\attd Electr~~ni~.s ill A,qri~l{lturr 4:333, 1990.Peacocke. R.D. and Grtif. D.H. “An Introductionto Speech and Spei.ikcr Recognition, /EEL - C()/Il-puter August:26. 1990.Petirl, J., ‘‘Rcast~ning Under Unccrttiinty.. AnnlIIIlRe\’ic\\ of Cotttp[lter S~ictt~c 4:37, 1990.Pickering, J .A.. ‘‘Touch-Scnsiti\e Scrccns: TheTcchnolt~g} und Their Application. ” //?/i’/-/(ifJi?t/?t//J(HII.IILI1 {!f”Mt~tl-hl[itltittt~ Stl{dic.s 25:249. 1986.[~lckerlng. J. et ~11 ., “RAIN: A Novel Approuch toConlputer-Aided Decisitmmaking in Agriculture undForestr}’.. Cot)lp[(tcr.~ atd Ele~tr~)ni~j in A,qritl[l-tllr(’ 4:275- 285. 1990.Plunt. R. E., ‘‘An Integrated Expert System Deci-sion Support S}stem for Agricultural System. A,y -ricl!ltltr{ll .5)YttIIn.~ 29;49–66. 1989.R:iuscher. 14 .N4. tind Jc~hns(~n. S., ‘ .Autht)ring Hy -

Chapter 4—Ad}7anced Computer Technologies ● 127

81.

82.

83.

84.

85.

86.

87.

88.

89.

90.

91.

()? .

93.

pcrtext Documents Using HypcrWriter. ’ Al Ap-pli~[itit)tt,s in N(itiirtil Rc.~~)iIr~e IWLi II[itqCmCIIt5(2): I 16. 1991.Rotich. J. W. et al.. ‘‘POMME: A Computer-BtisedConsultation S}rstcm for Apple orchard Mwwge-ment Using PROLOG. E.rpcr[ .S)’stct)is 2(2 ):56–69. 1985.Swu-tmmtia. H. et al.. ..An Artiticitil lntclligcnccMt~delling Approach to Simultiting Animul Htibitutlntertiutitms... E~olo(qi~[ilM() ~lt~llill<q” 44: 125.1988.Suundcrs et u].. ‘‘GRAPES: An Expert S)stcm torViticulture in Pcnnsjl\runia, ’ A/ App/i~(i[i(uis illN(iti{r(ii Rc.Yoiir(S(J A4[in[igc\tic}it I (2): 13-20. 1987.Schmisseur. E. und Ptinkratz. J.. ‘X LAYER: AnExpert S>stcnl tor Lti~m Managcmcnt... Poi{/ti-\Sticllte 68: 10447. 1989.Suhrcinemtikers. J.F. et al.. . ‘The Intr(xiuction ofExpert S}stcm\ in Animul Husbumir}. The \rc/-crilifit~]’ Qii{irterl] i 0( -1 ):28 i. I 988.Scars. A. and Shnciticrnum. B. ..liigh PrecisionTouchscrceris: Design Strategies und Compurist~n\With A Mouw. ’. lilt~’1.iititi~)il~il .Ioi{rtitil of” IVl~iII -

M(i(lli)te .Sti((lic.j 34:593. i 99 i .Sequcir:i. R.A. et al.. . .Ob,lcct-oriented Simulii-ti(~n: Pl:int Growth A Discrctc Org:in to (lytinIntcrtict it~ns. Ectjl. Jl(dcllilt,q 58:55-89. i 99 i .Shift-man. S. ct Lil.. . .13uilding :1 Spaxh intcrf’:iucto u Mcdiciil DiugntJstic S)stcm.. /h’E”L’ E“.I~wrlFcbruar)(:4 i. 1991.Sht~rtlit’t’c, E.. 11.. Computer-ll:iwi Mtxiic.:il C~m-sult;iti(ms: MYC”IN. Amcricun Elsmricr. Nc\\ Y(wk.i 976.Smith. D.C. ct :ii.. .’Dc\igning the S’I”Ail L’wrintcrt’acc.. BYTE 7:?-$2. 1982.Sonk:i, S. T.. . .Expcrt S)\tcms t~~r Busincis Dc-c ision M :ik in:.. ct~mnlisii(~nc~i b:ickgrt)lin(i p~ipcr

prepared t’(~r the (N’t’icc (~t” Tcchnd(lgy ASwS\mcnt.I 99 i .Spiihr. S. 1.., Jones. L. R.. :imi i~ill. D. E.. ‘. E\pcrtSy\tcm-Their Uw in [liiir~ 1 lcrci Nliin:igcmcnt ...Joi{rii[il of I)(iir> S’(i(’}1(() 7 I :879. 1988.Storw. N. 1). ct :il.. ‘ .C~~tf]cx: A hf(xili[:i[” E\pcrtS}stcnl thiit S~nthc\i/c\ Bi~~l(~yic:il iin(i Ec(~nt)nli-C:II Aniilj\is: ‘l-hc f’cst hl LIn:igcnlcn( Ad\ iv)r A\ AnExiinlplc.‘‘ i n Pr(jcct~[litig.$ (~/ IIic’ B1’ltliqii{() < ‘olloIi”Pro[ii{~ti(j\i (iti(l fic.~c~ircII {’(~Ii/crctitc. N:it i(~n;tlCott(m Council of” Amcric:i. hlcmphi~. TN. 19X7.

94.

95.

%.

97.

98

99

i(M).

Ioi .

I 02.

103.

104.

i05.

106.

107.

Stone. N.1).. ‘‘Knowiedge-Based S>stcms torC r o p s , commissioned background paper preparedfor the otfice of Technology Assessment. U.S.Congress, Washington. DC. i99i.Stone, N. D.. b’Chat~s in an lndi\idual-Lc\ci Pre-dator-Prey Model ..’ N(iti(r(il Rr.\oiircc M{)tieli}t<q4(4): 539-553. iwo.Sutciiffc. A.. Hi{tti~ilt-C’(~t ~il~i{tt’t. lntcrtii~c Dt)sicq}i(New York. NY: Spririger-vcrltig NCW York. inc..i989).

Taibot. M.. The M)th @Spw(h Ttdm(dogy, .YpwclI[ind L[iII(qi{(i<qt’-B[i.~t’(l I}itcrtic tio}i,~ \\’itli hl~i~liille,~,J.A. Wutcrww-th ~inci M. T:ilb(~t (cd\.). Ellis Hor-WIood. Chichcstcr. England. i 987.Thiemc. R.H. et tii.. ‘ .Expcrt S)stem “techniquesAppiitxi To Grtiin M:irketing Anui}sis.. (’oi~)-pi(tcr.s (iiId Elettr(}\iic”~ iii A,qrici(lliirc 1 :299. \ 987.Tomaszcwski. M. A.. . ‘~i~c ~~f Expert S)stems inAnimtil Agricuiturc, uommiwioncd btickgrt}lindptipcr prcpurcci for the Office of” Tcchnt~lt~gy ,A\-sessmcnt. I 9!) i .

Tomuszewski. M. A.. pcrson:il c(~lllll]llnic:itit~n. Sept.i7. i99i .‘r~rlcr. S.W. und Trcu. S.. . An intcrt’~icc Archi-tecture To Pro\idc A~iiipti\c T:isk-Spccit’ic (“(mtcxtfor the Usci-. I}tt(’1.liii{it)tiiil .loiIrIitil of” I~l(iII-i\l(i-[liiltc .~li(dic.~ 30:303”. i 989.V:in H(~rn. Al.. [,lti(l(~l”.~t(iti(lijtq E.rperI .Y>.~[cIIi.~.Banttim i30t~hs. Ncu Yoi+k. NY. i 9 8 6 .Viirncr. M .A. iind C:iCi) . R. A., b.l)tiirj -L: ,4 NewConcept in Technology Trtin\t-cr i’(~r l-lxtcnsit~n....Ioi(rn(il ()/ ll[iir~ .%~ic}i~c 74( Supp. I ):20 i . I 99 IWtiin, N.. Iklilicr. C. D. F.. :imi lli~ri~. R. l}.. . ‘ARuIc-B;IvA Inlkrencc Sjstcm l~)r Anim:il f%cduc-tion Miinwymcnt. (.oi)tpi(lci.$ (in(l [llt’clrolli~~ iliAcqri~i~lti{rc 2;277. i 988.W;ildrop. M.M . . . A l.:incim:irk in Speech Rcu-t~gn it ion. .Sticii(c 2-I(): i 615. 1988.Whittakcr. A.D. et :11.. . ‘Iliir) 1 {crci NLitritioniil

An~il~sis using Kmm Iccigc Sj’itcms “rCUhIll~LIC\.

A,qrici{ltiirtil .$].~tcjtl.~ 3 I :83. i 989.W(N)cis. W.A. . . .Lun:ir R(~ch\ in N:itul”ti] Engliih:Explorations in N:itur:ii i.linyu:i~c ~Licsti(m An-s ~+ c r I n g.” L illtq iii.~ t i~ .51ii(~ti(rt f>rf)ci~.~iliy. A .Z~impt~lli (Mi. ) ( Ncm Yt~rk. NY: North-Hollundp~l[l[lihlll,~ C(). . i 977). p. 52 I ..-