Stable prescriptive policy construction

A computer program product comprising:

• one or more computer readable storage media and program instructions stored on the one or more computer readable storage media, the program instructions comprising instructions to:
• identifying a decision tree corresponding to an artificial intelligence model, wherein the decision tree comprises an old rule, a set of precomputed coefficients, and an incidence matrix that maps a transaction to the old rule;
• executing a counterfactual estimator to produce a sorted list of features from a training data, wherein a feature graph for a feature in the sorted list includes a node corresponding to a value of the feature and a last node as a SKIP node wherein a path corresponding to a particular rule passing through the SKIP node indicates that the feature is not a part of the particular rule;
• detecting new data associated with an update to the identified decision tree, wherein the new data comprises (i) an update to a known feature in the sorted list of features, (ii) an intervention, and (iii) a beneficiary data;
• further executing the counterfactual estimator to output counterfactual data corresponding to the new data by analyzing the new data according to the sorted list of features and producing an expected outcome, the expected outcome being an effect on the beneficiary data in an absence of the intervention;
• generating an updated decision tree based on the identified new data and the identified counterfactual data, wherein the generating comprises generating a new rule executable by the artificial intelligence model, wherein the new rule corresponds to the expected outcome;
• and
• updating the artificial intelligence model from the old rule to the new rule.

A computer system comprising:

• one or more computer processors;
• one or more computer-readable storage media;
• program instructions stored on the computer-readable storage media for execution by at least one of the one or more processors, the program instructions comprising instructions to:
• identifying a decision tree corresponding to an artificial intelligence model, wherein the decision tree comprises an old rule, a set of precomputed coefficients, and an incidence matrix that maps a transaction to the old rule;
• executing a counterfactual estimator to produce a sorted list of features from a training data, wherein a feature graph for a feature in the sorted list includes a node corresponding to a value of the feature and a last node as a SKIP node wherein a path corresponding to a particular rule passing through the SKIP node indicates that the feature is not a part of the particular rule;
• detecting new data associated with an update to the identified decision tree, wherein the new data comprises (i) an update to a known feature in the sorted list of features, (ii) an intervention, and (iii) a beneficiary data;
• further executing the counterfactual estimator to output counterfactual data corresponding to the new data by analyzing the new data according to the sorted list of features and producing an expected outcome, the expected outcome being an effect on the beneficiary data in an absence of the intervention;
• generating an updated decision tree based on the identified new data and the identified counterfactual data, wherein the generating comprises generating a new rule executable by the artificial intelligence model, wherein the new rule corresponds to the expected outcome;
• and
• updating the artificial intelligence model from the old rule to the new rule.