Professional Documents
Culture Documents
Articles
CARMA
A Case-Based
Rangeland Management Adviser
■ CARMA is an advisory system for rangeland In Wyoming, the estimated total annual loss
grasshopper infestations that demonstrates how AI to grasshoppers is roughly $19 million. The
technology can deliver expert advice to compen- southeastern quadrant of the state is particu-
sate for cutbacks in public services. CARMA uses two larly prone to grasshopper infestations, with
knowledge sources for the key task of predicting significant areas of high-grasshopper densities
forage consumption by grasshoppers: (1) cases in 30 of the last 34 years.
obtained by asking a group of experts to solve rep-
Various chemical and biological pesticides
resentative hypothetical problems and (2) a
are available for treatment of grasshopper
numeric model of rangeland ecosystems. These
knowledge sources are integrated through the
infestations, but the cost of using these agents
technique of model-based adaptation, in which often outweighs the value of the forage saved
case-based reasoning is used to find an approxi- by their application. Moreover, indiscriminate
mate solution, and the model is used to adapt this pesticide application can be damaging to
approximate solution into a more precise solution. rangeland ecology. Despite their potential for
CARMA has been used in Wyoming counties since damage, the majority of grasshopper species
1996. The combination of a simple interface, flex- are usually innocuous or even beneficial to
ible control strategy, and integration of multiple grassland ecosystems. Of more than 400
knowledge sources makes CARMA accessible to inex- species of grasshoppers in the western United
perienced users and capable of producing advice
States, perhaps only 15 can be considered seri-
comparable to that produced by human experts.
ous pests; many of the other species are bene-
Moreover, because CARMA embodies diverse forms
of expertise, it has been used in ways that its devel- ficial in terms of weed control, nutrient
opers did not anticipate, including pest manage- cycling, and food for wildlife (Lockwood
ment research, development of industry strategies, 1993a, 1993b). The decision whether to use
and in-state and federal pest-management policy insecticides or other control measures is a com-
decisions. plex task depending on a multiplicity of fac-
tors, including not only short-term economic
costs and benefits but also preserving
G
rasshoppers are the most serious range- grasshoppers’ natural enemies (Joern and
land pest in the western United States, Gaines 1990), safeguarding biodiversity, and
consuming 21 to 23 percent of range- protecting environmental and human health.
land forage and causing an estimated $400 mil- Before 1996, the United States Department
lion in losses (Hewitt and Onsager 1983). Fig- of Agriculture (USDA) paid the entire cost of
ure 1 illustrates grasshopper infestation treatment on federal land, one-half the cost on
densities in the western United States during state land, and one-third the cost on private
2000, a fairly typical year. In years of heavy land. In addition, the USDA provided inten-
infestation, grasshopper densities and econom- sive surveys and pest-management advice to
ic losses might be much higher. For example, ranchers about treatment selection. Subse-
during the 1986 to 1987 outbreak, over 20 mil- quently, however, the USDA stopped provid-
lion acres of rangeland were treated for ing these subsidies (except for infestations on
grasshoppers in the western United States at a federal rangelands that represent an immedi-
cost of more than $75 million. ate threat to adjacent crops) and the level of
Copyright © 2002, American Association for Artificial Intelligence. All rights reserved. 0738-4602-2002 / $2.00 SUMMER 2002 49
Articles
survey and logistical support was substantially ecosystem. Various approaches to behavioral
decreased. This change in policy has increas- prediction are possible. In systems for which a
ingly shifted both the cost of treatment and precise model exists and accurate values of
the task of determining when treatment is state variables can be determined, simulation
desirable to ranchers themselves. CARMA was can be used to predict the system’s behavior.
developed to help compensate for the Alternatively, if there are sufficient historical
decreased availability of federal assistance by data, empirical methods such as case-based rea-
helping ranchers identify and balance the fac- soning (CBR) (Aamodt and Plaza 1994), deci-
tors relevant to pest-control decisions. sion tree induction (Quinlan 1993), or statisti-
cal techniques can be lead to accurate
prediction.
Task Description Precise models exist for the behavior of
many simple physical systems. However, mod-
CARMA’s task is to help ranchers determine the
els of agricultural, ecological, and other biolog-
most cost-effective responses to rangeland
ical systems are often incomplete, either
grasshopper infestations within user-defined because a complete state description for such
environmental constraints. Determining the systems cannot be determined or because the
most cost-effective responses requires, at a number and type of interactions between sys-
minimum, estimating (1) the value of the for- tem elements are poorly understood. More-
age that is likely to be consumed by grasshop- over, although historical data often exist for
pers if no action is taken, (2) the value of the such systems, they are often insufficient for
portion of this forage that would be saved in accurate prediction using empirical methods.
current and future years under each treatment As illustrated in figure 2, biological systems
option, and (3) the cost of each option. often occupy an intermediate point in the con-
Estimating grasshopper forage consumption tinuum between highly analytic domains,
requires predicting the behavior of a rangeland such as celestial mechanics and the prediction
50 AI MAGAZINE
Articles
SUMMER 2002 51
Articles
Model-based
case adaptation
Consumption Adapted
Cost-benefit
estimate under consumption
analysis of
acceptable estimate without
treatment Pesticide
Markov model treatment treatment
options efficacy and
of reinfestation options
probability grasshopper
consumption
models
measure based on market price. The expert their conclusions. Moreover, entomologists
then advises the rancher to take the most eco- can generate causal predictions of the effects of
nomical action, either applying the most cost- incremental variations on case facts.
effective control measure or doing nothing. Opportunism: Human experts can use a
Experts can justify their advice by appealing to variety of different strategies to solve a single
an underlying causal model but seem to use given problem depending on the available
this model only in explaining and adapting the information. Human experts don’t address the
predictions associated with prototypes and not subgoals that arise in decision making in an
in performing any sort of simulation. invariant order but adapt their problem-solv-
The protocol analysis identified four impor- ing behavior to the particular facts of a given
tant characteristics of human expert problem case.
solving in this field: (1) graceful degradation, In summary, the protocol analysis indicated
(2) speed, (3) causal explanations, and (4) that experts in rangeland pest management
opportunism. use an eclectic approach that includes case-
Graceful degradation: Human experts can based reasoning for consumption prediction,
use, but do not require, highly precise informa- rules for inferring case features and acceptable
tion of the type required for accurate model- control measures, and causal reasoning for
based reasoning. Less accurate information can adaptation and explanation. Moreover, expert
degrade the quality of advice an expert can problem solving is fast and tolerant of inaccu-
give but doesn’t preclude useful advice. In the racies in data.
worst case, human experts can provide plausi-
ble advice based merely on the location of the
rangeland and the date.
Application Description
Speed: Human experts can provide useful CARMA is designed to model the problem-solv-
advice quickly, suggesting, consistent with our ing behavior of experts in managing grasshop-
process description, that human experts can per infestations, as described in the previous
use highly compiled knowledge in the form of section. CARMA emulates expert human advice
prototypical cases. by providing treatment recommendations sup-
Explanations in terms of a causal model: ported by explanations in terms of causal, eco-
Although the speed and graceful degradation nomic, and pragmatic factors, including a
of human expert performance suggest that numeric estimate of the proportion of forage
experts can use compiled knowledge, they can consumed and a cost-benefit analysis of the
also readily provide causal explanations for various treatment options.
52 AI MAGAZINE
Articles
Model-based reasoner
Rule-based reasoner Consultation manager
Projection module
Feature inference module Infer relevant case features
Critical-period adapter
Feature inference rules Predict forage consumption
Feature weights
Treatment exclusion rules Determine treatments options
Rangeland ecosystem model
General rules Provide cost-benefit analysis
Pesticide efficacy model
Case-based
Geographical
reasoner
information Probabilistic reasoner Case matcher
Infestation history map Reinfestation probability Feature adapter
Rangeland value map calculator Prototypical cases
Weather station info. Markov infestation model New case
SUMMER 2002 53
Articles
54 AI MAGAZINE
Articles
ability resulting from the imprecise nature of necessary for determining exclusion is already
rangeland ecosystems, this estimate is convert- known from the case features (for example, the
ed to a qualitative range (for example, high, presence of grasshoppers in the first nymphal
meaning that approximately 60 to 100 percent instar—the earliest, readily observable develop-
of the available forage will be lost). The win- mental phase that hatches from the egg pod,
dow explaining estimated forage loss, shown which lies beneath the soil surface, into the
in figure 6, gives both aggravating and mitigat- above-ground environment—indicates an
ing factors (that is, factors tending to increase ongoing hatch, which precludes malathion
and factors tending to reduce estimated forage and carbaryl bait from consideration). Other
loss). Explanation text in this and other CARMA conditions must be determined from further
windows is produced using conventional user input (for example, will it be hot at the
schema-based techniques (Moore 1995). If the time of treatment? If so, exclude malathion.).
proportion of available forage that will be lost
to grasshoppers, and the proportion needed for Treatment Recommendation
livestock (and wildlife) exceeds 100% of the For each acceptable treatment option, CARMA
forage available, CARMA concludes that grass- provides estimates of the reduced probability
hoppers will cause economic losses. of future reinfestation and current-year and
long-term savings. From the estimated savings,
Determining Treatment Options carma recommends the treatment or treat-
If grasshoppers will cause economic losses, CAR- ments that are most economical.
MA applies a set of rules to determine the treat- CARMA calculates the total reduced probabili-
ment options that are excluded by the condi- ty of future reinfestation for each treatment
tions of the case. Some of the information type using a Markov model of infestation prob-
SUMMER 2002 55
Articles
ability for each location derived from historical save the most under a best-case scenario. Usu-
data collected by the USDA and synthesized by ally, the worst- and best-case scenarios produce
the University of Wyoming Entomology Sec- the same recommended treatment. Following
tion (Lockwood and Kemp 1987). CARMA com- the treatment recommendation, the initial
putes the current-year savings as the difference phase of the consultation is complete.
between the value of forage saved and the Optionally, the user can then rerun the con-
treatment cost. CARMA calculates the savings for sultation with one or more case facts or treat-
future years for each treatment type by multi- ment parameters altered. To facilitate the alter-
plying the reduced probabilities of reinfesta- ing of treatment parameters, CARMA includes a
tion by the estimated forage loss for each sub- treatment-matrix window, shown in figure 8,
sequent year. that permits the user to change the default val-
A typical treatment recommendation win- ues of any of the variables (for example, cost of
dow, including estimates of future reinfesta- chemical, cost of carrier, rate of application,
tion and economic savings, appears in figure 7. cost of application, expected efficacy) that
CARMA lists both worst- and best-case scenarios determine the cost of an acre protected of each
for most calculations. Note that this analysis treatment option or the cost of adding entirely
includes no treatment as an option and that new treatments. Although calculation of cost
negative savings indicate a loss. of an acre protected is straightforward, experi-
CARMA recommends the treatment that is ence has shown that it is a common source of
estimated to save the most under a worst-case errors, particularly when users are attempting
scenario and the treatment that is estimated to to determine the costs of reduced agent-area
56 AI MAGAZINE
Articles
treatments (RAATs), which involve applying based adaptation is useful in domains in which
insecticide to only a fraction of the infested both cases and models are available, but nei-
area. The treatment matrix permits entirely ther is individually sufficient for accurate pre-
new parameters and products to be added to diction. Such domains are typified by chemical
CARMA as they are developed, an important fea- or biological systems with well-developed, but
ture in view of rapid recent refinements in imperfect, models. Model-based adaptation
grasshopper management methods. Easy mod- has been applied for bioprocess recipe plan-
ification of treatment parameters permits ning in SOPHIST (Aarts and Rousu 1996; Rousu
“gaming” to explore which combination of and Aarts 1996) for selecting colorants for plas-
insecticide, rate, carrier, and coverage provides tic coloring in FORMTOOL (Cheetham and Graf
the best return on investment. For example, 1997) and in design reuse (Goel 1991).
such gaming can permit ranchers and pesticide Model-based adaptation is appropriate for
vendors to negotiate a price for a pesticide that CARMA’s advisory task because both empirical
is economical given the current infestation lev- knowledge, in the form of cases, and a grass-
el and the price of substitute forage. land ecology model are available, but neither is
individually sufficient for accurate prediction
of forage consumption, given the information
Uses of AI Technology that ranchers can typically provide.
CARMA uses AI technology in two distinct ways.
First, as described earlier, CARMA’s control strat- Case-Based Reasoning in CARMA
egy emulates human experts’ speed, oppor- The initial impetus for using CBR for forage
tunism, explanation capability, flexibility in consumption prediction was cognitive veri-
eliciting relevant case features through a vari- similitude. The protocol analysis suggested
ety of alternative heuristic rules, and ability to that human experts in this domain reason
integrate multiple knowledge sources. Second, using prototypes, which is consistent with var-
CARMA uses model-based adaptation for the key ious cognitive studies that have demonstrated
reasoning step of predicting the amount of for- that examples or prototypes often play a cen-
age that will be consumed by grasshoppers. tral role in human concept structure (Klein and
Model-based adaptation consists of using Calderwood 1988; Smith and Medin 1981).
CBR to find an approximate solution and mod- During the development of CARMA, however,
el-based reasoning to adapt this approximate CBR’s ability to facilitate knowledge acquisi-
solution into a more precise solution. Model- tion grew in importance. Few precise records of
SUMMER 2002 57
Articles
PC'
Density
NC
PC
Hatch Adult
Average phase
Figure 9. Projection of a Prototypical Case PC to PC’ to Align Its Developmental Phase with New Case NC.
58 AI MAGAZINE
Articles
13.3 CARMA:
14.2 Specific weights
Wyoming 21.1
Global weights
expert sets 34.9
25.6
0 5 10 15 20 25 30 35 40
Figure 10. Root-Mean-Squared Error Rate (in Percent) for Leave-One-Out Test Results.
Figure 9 illustrates how the population in son during which forage losses cannot fully be
prototypical case PC must be projected back- replaced by forage growth, termed the critical
ward in time to PC’ to match the average devel- period. The forage loss predicted by a prototyp-
opmental phase of new subcase NC. Projection ical case must be adapted if the proportion of
backward in time increases grasshopper densi- the life span of the grasshoppers overlapping
ty by removing the effect of attrition over the the critical period in the new case differs from
interval of the projection, whereas projection that in the prototypical case. This adaptation
forward in time decreases grasshopper density requires determining, for both the new case
by adding attrition during this interval. The and the prototypical case, the proportion of
vertical bar corresponding to PC and PC’ indi- the grasshopper population’s lifetime con-
cates the confidence range for grasshopper sumption occurring in the critical period. For a
density, which always increases (indicating more complete description of model-based
greater uncertainty) as a function of the inter- adaptation in CARMA, see Branting, Hastings,
val projected. and Lockwood (1997).
In feature adaptation, the forage loss predict-
ed by the best-matching prototypical case is Experimental Evaluation
modified to account for any feature differences of Model-Based Adaptation
(other than average developmental phase) The design of CARMA’s forage-consumption
between it and the subcase. The modification is component was based on the hypothesis that
a linear function of the feature differences. The an integration of model-based and case-based
coefficients of the linear function are deter- reasoning can lead to more accurate forage-
mined by a form of introspective learning consumption predictions than the use of either
(Hanney and Keane 1997; Leake, Kinley, and technique individually. This hypothesis was
Wilson 1995), consisting of hill climbing based on the observation that neither the
through parameter space to optimize leave- causal model nor the empirical data available
one-out predictive accuracy within the case for rangelands are individually sufficient for
library (Branting, Hasting, and Lockwood accurate prediction. To test this hypothesis, an
1997). ablation study was performed in which CARMA’s
Critical-period adaptation is needed because empirical and model-based knowledge compo-
grasshopper consumption is most damaging if nents were each tested in isolation and the
it occurs during the portion of the growing sea- results compared to the performance of the full
SUMMER 2002 59
Articles
CARMA prediction system under both global and approaches—21.1 percent for factored-NN,
case-specific adaptation weight modes. 34.9 percent for ID3, and 25.6 percent for linear
Each predictive method was tested using a regression—and for the purely model-based
series of leave-one-out tests in which a set of approach—29.6 percent. CARMA-SPECIFIC and
cases (S) from a single expert was split into one CARMA-GLOBAL were also more accurate than the
test case (C) and one training set (S - C). The alternative methods on the Wyoming median
methods were trained on the forage-loss pre- set, although linear regression was only slight-
dictions of the training set and tested on the ly less accurate.
test case. This method was repeated for each The initial confirmation of the hypothesis
case within the set (S). that integrating model-based and case-based
CARMA’s empirical component was evaluated reasoning through model-based adaptation
by performing leave-one-out tests for CARMA’s leads to more accurate forage-consumption
forage-consumption module with all model- predictions than the use of either technique
based adaptation disabled. CARMA’s forage-con- individually is tentative because the relatively
sumption module with model-based adapta- low level of agreement among experts and the
tion disabled is termed factored nearest-neighbor absence of any external standard give rise to
prediction (factored-NN) because under this uncertainty about what constitutes a correct
approach, prediction is based simply on the prediction. A detailed description of the empir-
sum of nearest-neighbor predictions for each ical evaluation of CARMA is set forth in Branting,
subcase. Two other empirical methods were Hastings, and Lockwood (1997).
evaluated as well: (1) decision tree induction
using ID3 (Quinlan 1986) and (2) linear regres-
sion using QR factorization (Hager 1988) to
Application Use and Payoff
find a least squares fit to the feature values and In June 1996, CARMA 2.0 was distributed to the
associated predictions of the training cases. University of Wyoming Cooperative Extension
The predictive ability of CARMA’s model- Offices and Weed and Pest District Offices in
based component in isolation was evaluated by each of Wyoming’s 23 counties and was made
developing a numeric simulation based on CAR- available for download from a University of
MA’s model of rangeland ecology. This simula- Wyoming web site. CARMA 2.0 was used by
tion required explicit representation of two Wyoming ranchers and pest managers every
forms of knowledge implicit in CARMA’s cases: summer from 1996 to 2001. CARMA has been
(1) the forage to an acre based on the range- endorsed and advocated for use by pest man-
land value of the location and (2) the forage agers by the United States National Grasshop-
typically eaten in a day by each grasshopper for per Management Board (NGMB 2001). Perhaps
each distinct grasshopper overwintering type the greatest interest in the system has been
and developmental phase. expressed by the county-level Weed and Pest
The accuracy of each approach was evaluat- District supervisors, who—with the withdrawal
ed using leave-one-out testing for the respons- of USDA support—have become the “front-
es from each of the eight Wyoming experts and line” agency in grasshopper pest management.
for a data set consisting of the median of the Workshops to train these individuals in the
predictions of the Wyoming experts on each optimal use of CARMA were developed and
case. The full CARMA prediction system was test- delivered at the request of the agency.
ed using both global-adaptation weights (CAR- Although CARMA was designed as an advisory
MA - GLOBAL ) and case-specific adaptation system for ranchers, CARMA’s ability to robustly
weights (CARMA-SPECIFIC). integrate a variety of knowledge sources led it
The root-mean-squared error rates for each to be applied in several ways that were not
of the methods are set forth in figure 10. These imagined when the program was developed.
rates provide initial confirmation for the First, CARMA’s economic analysis has been used
hypothesis that integrating model-based and to justify pest management policy decisions. In
case-based reasoning through model-based 1998, CARMA’s economic analysis was used to
adaptation leads to more accurate forage-con- generate a declaration of grasshopper disaster
sumption predictions than the use of either areas by Wyoming County Commissions, lead-
technique individually. The smallest root- ing to low-interest, federal loans by the Farm
mean-squared error rate was obtained by Service Administration. CARMA’s economic
CARMA-SPECIFIC. On the Wyoming expert sets, analysis played a role in the NGMB’s recom-
the root-mean-squared error rate was 13.3 per- mendation of a new treatment approach,
cent for CARMA-SPECIFIC and 14.2 percent for RAATs (Nelson 1999), a strategy now adopted
CARMA - GLOBAL . The root-mean-squared error in six states.
rate was higher both for the empirical Second, CARMA’s analysis was incorporated
60 AI MAGAZINE
Articles
into industry strategies. Uniroyal (CK Witco) diminished explanation capability that would
developed recommendations for the use of have resulted from a purely simulation-based
Dimilin, a new chemical pesticide, using approach.
CARMA ’s analysis. Similarly, RhônePoulenc A key factor in CARMA’s acceptance among
(Aventis) developed recommendations for the users is its simple interface and speed, which
use of Fipronil based on CARMA’s analysis. make using CARMA straightforward. The combi-
Finally, CARMA-based economic analysis was nation of a simple interface, flexible control
incorporated into pest management research strategy, and integration of multiple knowl-
in Lockwood et al. (1999) and Lockwood and edge sources makes CARMA accessible to inexpe-
Schell (1997). rienced users and capable of producing advice
comparable to that produced by human
experts.
Development, Deployment,
and Maintenance Notes
1. www.sidney.ars.usda.gov/grasshopper.
CARMA was developed as a dissertation project
2. www.sdvc.uwyo.edu/grasshopper/carma.htm.
(Hastings 1996). The out-of-pocket develop-
ment costs were small, consisting of several
years of graduate research assistant support Acknowledgments
and the license fees for Franz Allegro Common
This research was supported by a grant from
Lisp, the language in which CARMA was devel-
the University of Wyoming College of Agricul-
oped. However, the path to the development
ture, a faculty grant in aid from the University
of CARMA was quite circuitous, with a variety of
of Wyoming Office of Research, National Sci-
different approaches to grasshopper advising
ence Foundation career grant IRI-9502152, and
having been developed, tested, and rejected.
a grant by Uniroyal (CK Witco). We gratefully
Thus, the development costs would have been
acknowledge the contributions of Scott Schell
much higher outside an academic environ-
to the development of CARMA’s task analysis
ment.
and knowledge base.
In the years since the distribution of CARMA
2.0, there have been a number of changes in
pest-treatment practices. In 2001, CARMA 2.0 References
was updated to CARMA 3.3 to reflect these Aamodt, A., and Plaza, E. 1994. Case-Based Reason-
changes and include the treatment matrix dis- ing: Foundational Issues, Methodological Variations,
cussed earlier for calculating treatment costs and System Approaches. AI Communications 7(1):
for each acre under various alternative eco- 39–59.
nomic conditions. CARMA’s declarative knowl- Aarts, R., and Rousu, J. 1996. Toward CBR for Bio-
edge representation made revising the program process Planning. Paper presented at the Third Euro-
straightforward. These changes were funded by pean Workshop on Case-Based Reasoning (EWCBR-
a grant from a producer of a pesticide intro- 96), 14–16 November, Lausanne, Switzerland.
duced after the distribution of CARMA 2.0 and Allen, T. F. H., and Hoekstra, T. W. 1992. Toward a
therefore not included as a treatment option in Unified Ecology. New York: Columbia University Press.
the earlier version. CARMA can be downloaded Branting, K.; Hastings, J.; and Lockwood, J. 1997.
from the USDA’s grasshopper-control web site Integrating Cases and Models for Prediction in Bio-
or from the University of Wyoming’s Grass- logical Systems. AI Applications 11(1): 29–48.
hoppers of Wyoming and the West web site.1,2 Cheetham, W., and Graf, J. 1997. Case-Based Reason-
ing in Color Matching. In Proceedings of the Second
International Conference on Case-Based Reasoning,
Conclusion 1–12. New York: Springer.
Goel, A. 1991. A Model-Based Approach to Case
CARMA demonstrates how AI technology can be
Adaptation. Paper presented at the Thirteenth Annu-
used to deliver expert advice to compensate for al Conference of the Cognitive Science Society, 7–10
cutbacks in public services. CBR proved to be August, Chicago, Illinois.
an appropriate AI technique for the forage-pre- Hager, W. 1988. Applied Numerical Linear Algebra.
diction component both because experts in Englewood Cliffs, N.J.: Prentice Hall.
this domain appear to reason with cases and Hanney, K., and Keane, M. 1997. The Adaptation
because asking experts to solve example cases Knowledge Bottleneck: How to Ease It by Learning
was an effective knowledge-acquisition tech- from Cases. In Proceedings of the Second International
nique. Model-based adaptation provided a Conference on Case-Based Reasoning, 359–370. New
mechanism for incorporating rangeland York: Springer.
ecosystem models into the system without the Hastings, J. 1996. A Mixed-Paradigm Reasoning
slow performance, sensitivity to noise, and Approach to Problem Solving in Incomplete Causal
SUMMER 2002 61
Articles
Theory Domains. Ph.D. thesis, Department International Journal of Pest Management His current research interests include
of Computer Science, University of 46(1): 29–42. empirical natural language processing,
Wyoming. Moore, J. 1995. Participating in Explanatory legal applications of AI, and integration of
Dialogues. Cambridge, Mass.: MIT Press. case-based reasoning with other problem-
Hastings, J.; Branting, K.; and Lockwood, J.
solving paradigms. His e-mail address is
1996. A Multiparadigm Reasoning System Nelson, D., ed. 1999. Proceedings of the
karl.branting@livewirelogic.com.
for Rangeland Management. Computers and National Grasshopper Management Board
Electronics in Agriculture 16(1): 4–67. Meeting, Colorado, 12–13 January, Denver, Jeffrey Lockwood is a
Colorado. professor of entomology
Hewitt, G., and Onsager, J. 1983. Control of
NGMB. 2001. National Grasshopper Man- in the Department of
Grasshoppers on Rangeland in the United
agement Board Position Statement. Paper Renewable Resources at
States: A Perspective. Journal of Range Man-
presented at the National Grasshopper the University of Wy-
agement 36(2): 202–207.
Management Board Meeting, 16–17 Janu- oming. His research is
Joern, A., and Gaines, S. B. 1990. Popula- concentrated on the
tion Dynamics and Regulation in ary, Denver, Colorado.
ecology of superabun-
Grasshoppers. In Biology of Grasshoppers, Pimm, S. L. 1991. The Balance of Nature:
dant (pest management)
eds. R. F. Chapman and A. Joern, 415–482. Ecological Issues in the Conservation of Species
and rare (conservation biology) grasshop-
New York: Wiley. and Communities. Chicago: University of
per populations. He is the executive direc-
Klein, G. A., and Calderwood, R. 1988. How Chicago Press.
tor of The Orthopterists Society (a 350-
Do People Use Analogues to Make Deci- Quinlan, J. R. 1986. Induction of Decision member international organization
sions? In Proceedings of the DARPA Workshop Trees. Machine Learning 1(1): 81–106. devoted to the study of grasshoppers, crick-
on Case-Based Reasoning, 209–218. San Quinlan, R. 1993. C4.5: Programs for ets, and katydids) and the director of the
Francisco, Calif.: Morgan Kaufmann. Machine Learning. San Francisco, Calif.: Association for Applied Acridology Interna-
Leake, D.; Kinley, A.; and Wilson, D. 1995. Morgan Kaufmann. tional (the first and only humanitarian-
Learning to Improve Case Adaptation by Rousu, J., and Aarts, R. 1996. Adaptation based, nongovernmental organization of
Introspective Reasoning and CBR. In Case- Cost as a Criterion for Solution Evaluation. entomologists in the world, providing
Based Reasoning Research and Development: Paper presented at the Third European expert advice, training, and applied res-
First International Conference on Case-Based Workshop on Case-Based Reasoning earch to people and nations in need). Lock-
Reasoning (ICCBR-95), 229–240. Lecture (EWCBR-96), 14–16 November, Lausanne, wood teaches courses in insect population
Notes in Artificial Intelligence 1010. New Switzerland. biology, insect anatomy and physiology,
York: Springer. Smith, E. E., and Medin, D. L. 1981. Cate- and biodiversity. He has published in the
Lockwood, J. A. 1993a. Benefits and Costs gories and Concepts. Cambridge, Mass.: Har- areas of pest management, population
of Controlling Rangeland Grasshoppers vard University Press. modeling, biological control, agricultural
(Orthoptera: Acrididae) with Exotic Organ- ethics, and conservation biology; he serves
Wettschereck, D., and Dietterich, T. 1995.
isms: Search for a Null Hypothesis and Reg- on the editorial board of the Journal of
An Experimental Comparison of the Near-
ulatory Compromise. Environmental Ento- Insect Conservation. His e-mail address is
est-Neighbor and Nearest-Hyperrectangle
mology 22(5): 904–914. lockwood@uwyo.edu.
Algorithms. Machine Learning 19(1): 5–28.
Lockwood, J. A. 1993b. Environmental
Issues Involved in Biological Control of John Hastings is an
Rangeland Grasshoppers (Orthoptera: Acri- assistant professor of
didae) with Exotic Agents. Environmental computer science at the
Entomology 22(3): 503–518. University of Nebraska at
Lockwood, J., and Kemp, W. 1987. Proba- Kearney (UNK). He re-
bilities of Rangeland Grasshopper Out- ceived his B.S. (1989,
breaks in Wyoming Counties. Wyoming with honors), M.S.
Agricultural Experiment Station Bulletin (1991), and Ph.D. (1996)
B(896). in computer science
from the University of Wyoming. After
Lockwood, J., and Lockwood, D. 1991.
working several years as an expert systems
Rangeland Grasshopper (Orthoptera: Acridi-
developer and consultant, he joined the
dae) Population Dynamics: Insights from
faculty at UNK in 2001. His research inter-
Catastrophe Theory. Environmental Ento-
ests include case-based reasoning, machine
mology 20(4): 970–980.
learning, and applications of AI to natural
Lockwood, J., and Schell, S. 1997. Decreas- resource management. His e-mail address is
ing Economic and Environmental Costs hastingsjd@unk.edu.
through Reduced Area and Agent Insecti-
Karl Branting is a princi-
cide Treatments (RAATs) for the Control of
ple research scientist at
Rangeland Grasshoppers: Empirical Results
LiveWire Logic, Inc., in
and Their Implications for Pest Manage-
Morrisville, North Car-
ment. Journal of Orthoptera Research 6(1):
olina. He received a B.A.
19–32.
magna cum laude in phi-
Lockwood, J.; Schell, S.; Foster, R.; Reuter, losophy from the Uni-
C.; and Rachadi, T. 1999. Reduced Agent- versity of Colorado, a
Area Treatments (RAATs) for Management J.D. from Georgetown
of Rangeland Grasshoppers: Efficacy and University, and a Ph.D. in computer sci-
Economics under Operational Conditions. ence at the University of Texas at Austin.
62 AI MAGAZINE