From Artificial Intelligence to Decision Support Systems

Optimal agricultural production management relies on close observations of animals and fields and, if problems arise, on a timely reaction based on expert knowledge. To assist this process, management information and decision support systems have been used for some time - but not always with sufficient quality and capacity.
Artificial intelligence seemed an interesting option to improve the dissemination of expert knowledge. From the start the so-called rule based expert systems were promising. Here the domain expert is modelled by elicitation of the rules he or she uses to reach a solution. However, it soon became evident that such systems were of limited use, because they could not adequately handle the uncertainty in agricultural production.
Then techniques related to the Bayesian network methodology came into focus. Instead of modelling the expert, his knowledge of the domain (e.g., animal production) is used in the model. The uncertainty is an integrated part of the model, and observations can be directly incorporated to improve the decisions. Promising prototypes included bovine paternity testing, optimal winter wheat management, and mating management of sows. These prototypes were met with a certain amount of scepticism, but indicated the potential of the techniques.
Currently, new agricultural applications are on the verge of being implemented for decision support within sow and slaughter pig production. Furthermore the techniques have become an integrated part of several research projects related to production management. The agricultural application areas have indicated methodological limitations. New developments to solve these problems are sought in the Dina collaboration merging the expertise of researchers with different theoretical background.