We are using the term ‘Bot Worker’ to mean a composition of multiple individual bots, each with different skills or abilities. A Bot Worker has the ability to use any combination of individual bot skills to provide coarser abilities to provide hyper (end to end) automation of complex processes, guided by a set of SOP (Standard Operating Process) diagrams provided using BPMN 2.0 standards.
A Bot Worker which performs activities like document scrutiny, reading / extracting data from documents, recognizing image objects / text, taking decisions, check list executions, etc. is the Maker Bot Worker. The Focus of this post is the Checker Bot Worker which checks the work performed by a Maker Bot Worker and thus confirm or reject the work done by the Maker Bot Worker. The Checker Bot Worker can be an associate of the Maker Bot Worker to improve the quality of work done and minimize Bot Worker errors. A Human Checker can still be used for high value and high-risk transactions.
We have developed a Pattern Definition Language (PDL) to describe an image pattern or a data pattern to recognize an image object / character / data scenario. The Pattern Definition provides a generic description of the component image features and their relative placements. A large set of Pattern Definitions can be used by Pattern Matching Bots to recognize / decide and this has created skill-based Maker Bot workers, distinct from Machine Learning Bot Workers, which require extensive labelled data for training.
We have now, extended the Pattern Definition Language to describe an Anti – Pattern. An Anti-Pattern describes why a given image object or a data context is not what a Maker Bot Worker has recognized / decided. This is called as the ‘Not This’ flagging ability / skill. We now started defining extensive anti-pattern descriptions to construct a Checker Bot Worker. A Checker Bot Worker uses a set of Anti Patterns to either confirm a Maker Bot Worker decision / recognition or to reject it.
This ‘Not This’ skill reduces the Maker Bot Worker Errors to less than 1 Error in a Million Decisions. Even this level of errors can be reduced by additional feedback anti-pattern definitions.
Patterns has started offering Anti-Pattern definition driven, skill-based Checker Bot Workers as companion Bot Worker with all Bot Worker variants. 2022 has been an exciting year for the Patterns Cognitive team with several exciting innovations and more innovations in the final implementation stage.
Aug
06
