Counter fraud system

A computer program product stored on a non-transitory computer readable medium, which when executed by a computing system utilizes artificial intelligence (AI) to generate novel fraud schemes that are configured to detect perpetrated fraud using AI planning and computational logics operating at a theory-of-mind level, the program product comprising:

• program code configured to generate a semantic information modeler and a fraud scheme generator, the program code causing a processor to perform processes including:
• implementing, in the semantic information modeler, a mathematical model of a financial or commercial domain which is vulnerable to fraud, wherein the domain is governed by laws and regulations designed to prevent fraud, wherein the model is based on a set of data sources, wherein the model comprises semantic information, expressed in a formal cognitive event calculus (CEC) language which is highly expressive and in which intensional operators of the CEC are applied to domain participants and allowed to range over subsequent events, wherein the intensional operators relate to what domain participants believe, know, intend, and perceive and comprise theory-of-mind logic involving targeted cognitive states of human agents, and wherein the model is expressed in quantificational formulae so that full models of plans on which fraud schemes are based can be developed, the semantic information including:
• rules and processes of the domain expressed as formulae, wherein the rules include at least one of a law or regulation,
• semantic models of known fraud schemes pertaining to the domain, and
• knowledge bases of domain participants, wherein the domain participants comprise actors capable of perpetrating fraud and observers trained to prevent fraud, wherein the knowledge bases are expressed as theory-of-mind logic and are represented using cognitive calculi that express cognitive states of the domain participants;
• and
• automatically, in the fraud scheme generator, generating a fraud scheme expressed using cognitive calculi within the domain using the semantic information that is carried out by a domain participant, wherein the fraud scheme is configured to create an illicit monetary gain, and wherein generating the fraud scheme includes:
• receiving and expressing an inputted goal for the domain as a goal state Sg using cognitive calculi;
• providing a start state Ss of the domain expressed using cognitive calculi;
• discovering a chain of actions that change the domain from the start state Ss to the goal state Sg, wherein each action in the chain of actions is selected from a set of actions that formulaically map a current state of the domain to a new state, wherein the chain of actions comprises a series of events that are related temporally or causally;
• ensuring that the chain of actions forms a novel fraud scheme that is mathematically provable to break at least one law or regulation of the domain, to be outside the purview of relevant domain participants and not be a known fraud scheme, wherein proving includes logically evaluating a plan as a proof and testing and ensuring the plan achieves the goal state;
• in response to the generated fraud scheme not being a novel fraud scheme, iteratively generating a new fraud scheme with a new start state and a new goal;
• and
• in response to the generated fraud scheme being a novel fraud scheme, outputting the novel fraud scheme in a natural language output that allows the novel fraud scheme to be tested in a real environment of the domain to determine if the novel fraud scheme has been perpetrated.

A computerized method that utilizes artificial intelligence (AI) to generate novel fraud schemes that are configured to detect perpetrated fraud using AI planning and computational logics operating at a theory-of-mind level, the method comprising:

• implementing, in a semantic information modeler, a mathematical model of a financial or commercial domain which is vulnerable to fraud, wherein the domain is governed by laws and regulations designed to prevent fraud, wherein the model is based on a set of data sources, wherein the model comprises semantic information, expressed in a formal cognitive event calculus (CEC) language which is highly expressive and in which intensional operators of the CEC are applied to domain participants and allowed to range over subsequent events, wherein the intensional operators relate to what domain participants believe, know, intend, and perceive and comprise theory-of-mind logic involving targeted cognitive states of human agents, and wherein the model is expressed in quantificational formulae so that full models of plans on which fraud schemes are based can be developed, the semantic information including:
• rules and processes of the domain expressed as formulae, wherein the rules include at least one of a law or regulation,
• semantic models of known fraud schemes pertaining to the domain, and
• knowledge bases of domain participants, wherein the domain participants comprise actors capable of perpetrating fraud and observers trained to prevent fraud, wherein the knowledge bases are expressed as theory-of-mind logic and are represented using cognitive calculi that express cognitive states of the domain participants;
• and
• automatically, in a fraud scheme generator, generating a fraud scheme expressed using cognitive calculi within the domain using the semantic information that is carried out by a domain participant, wherein the fraud scheme is configured to create an illicit monetary gain, wherein generating the fraud scheme includes:
• receiving and expressing an inputted goal for the domain as a goal state Sg using cognitive calculi;
• providing a start state Ss of the domain expressed using cognitive calculi;
• discovering a chain of actions that change the domain from the start state Ss to the goal state Sg, wherein each action in the chain of actions is selected from a set of actions that formulaically map a current state of the domain to a new state, wherein the chain of actions comprises a series of events that are related temporally or causally;
• and
• ensuring that the chain of actions forms a novel fraud scheme that is mathematically provable to break at least one law or regulation of the domain, to be outside the purview of relevant domain participants and not be a known fraud scheme, wherein proving includes logically evaluating a plan as a proof and testing and ensuring the plan achieves the goal state;
• in response to the generated fraud scheme not being a novel fraud scheme, iteratively generating a new fraud scheme with a new start state and a new goal;
• in response to the generated fraud scheme being a novel fraud scheme, outputting the novel fraud scheme in natural language that allows the novel fraud scheme to be tested in a real environment of the domain to determine if the novel fraud scheme has been perpetrated.