Self-correcting bot

Patent metadata

Field	Value
Publication number	US-2024419576-A1
Application number	US-202418813202-A
Country	US
Kind code	A1
Filing date	Aug 23, 2024
Priority date	Jan 6, 2023
Publication date	Dec 19, 2024
Grant date	—

Abstract

Bots are typically programmed to automate tasks and provide statistically expected results. However, a bot may malfunction and generate aberrant outputs. It is technically challenging to detect the aberrant outputs and determine whether the outputs are due to an error in how the bot processes inputs or because the bot has received unusual or unexpected input data. Apparatus and methods are provided for auto-determining why a bot has generated output outside expected results. An auto-correct bot will detect problems and auto-identify potential solutions, simulate those solutions and apply those solutions to remediate the malfunctioning bot.

First claim

What is claimed is:

1 . An artificial intelligence (“AI”) method for autonomously diagnosing a malfunction of a bot, the method comprising extracting computer readable instructions stored on a non-transitory medium and executing the computer readable instructions on a processor, wherein execution of the computer readable instructions by the processor:
monitors outputs generated by the bot;
detects a threshold number of aberrant outputs during a time-window;
identifies a potential solution for reducing the threshold number of aberrant outputs;
simulates application of the potential solution; and
based on results of the simulating, autonomously adjusts at least one processing parameter of the bot;
parses inputs associated with each of the aberrant outputs;
determines whether the threshold number of aberrant outputs are due to aberrant inputs or aberrant processing of the inputs by the bot; and
in response to determining that the threshold number of aberrant outputs are due to aberrant processing of the inputs by the bot, adjusts the at least one processing parameter of the bot based on a timestamp associated with each of the inputs.

2 . The AI method of claim 1 , wherein execution of the computer readable instructions by the processor:
generates simulated inputs; generates known outputs associated with the simulated inputs; submits the simulated inputs to the bot; determines whether outputs generated by the bot in response to the simulated inputs correspond to the known outputs; registers the bot as being associated with a processing error when the outputs generated by the bot in response to the simulated inputs do not correspond to the known outputs; and registers the inputs as being erroneous when the outputs generated by the bot in response to the simulated inputs correspond to the known outputs.

3 . An artificial intelligence (“AI”) method for autonomously diagnosing a malfunction of a bot, the method comprising extracting computer readable instructions stored on a non-transitory medium and executing the computer readable instructions on a processor, wherein execution of the computer readable instructions by the processor:
monitors outputs generated by the bot;
detects a threshold number of aberrant outputs during a time-window;
identifies a potential solution for reducing the threshold number of aberrant outputs;
simulates application of the potential solution; and
based on results of the simulating, autonomously adjusts at least one processing parameter of the bot;
parses inputs associated with each of the aberrant outputs;
determines whether the threshold number of aberrant outputs are due to aberrant inputs or aberrant processing of the inputs by the bot; and
in response to determining that the threshold number of aberrant outputs are due to aberrant processing of the inputs by the bot, decommissions the bot.

4 . The AI method of claim 3 , wherein execution of the computer readable instructions by the processor:
generates simulated inputs; generates known outputs associated with the simulated inputs; submits the simulated inputs to the bot; determines whether outputs generated by the bot in response to the simulated inputs correspond to the known outputs; registers the bot as being associated with a processing error when the outputs generated by the bot in response to the simulated inputs do not correspond to the known outputs; and registers the inputs as being erroneous when the outputs generated by the bot in response to the simulated inputs correspond to the known outputs.

5 . An artificial intelligence (“AI”) system for autonomously diagnosing a malfunction with a bot, the system comprising:
one or more computer servers including a processor circuit;
a first bot that is programmed to receive user inputs and generate automated outputs to the user inputs;
a second bot that is programmed to:
monitor the user inputs and automated outputs generated by the first bot during a first time-window;
detect a threshold number of aberrant automated outputs generated by the first bot;
diagnose a malfunction that is causing the first bot to generate the threshold number of aberrant automated outputs;
autonomously reprogram the first bot to change a processing parameter applied by the first bot;

wherein the second bot is further programmed to simulate the change to the processing parameter before autonomously reprogramming the first bot;
generate a test input;
input the test input to a digital twin of the first bot; and
trace each operational step performed by the digital twin to generate an automated response in response to the test input.

6 . The AI system of claim 5 wherein:
the second bot is further programmed to monitor the user inputs and automated outputs generated by the first bot during a second time-window; and
the second time-window begins after the first bot is reprogrammed to change the processing parameter.

7 . The AI system of claim 5 wherein the second bot is programmed to determine the processing parameter based on applying a machine learning analysis to automated outputs generated by a plurality of bots.

8 . The AI system of claim 5 wherein the second bot is programmed to determine the processing parameter based on applying a machine learning analysis to automated outputs generated by the first bot before and during the first time-window.

9 . The AI system of claim 5 wherein the second bot is programmed to detect the threshold number of aberrant automated outputs generated by the first bot based on:
a length of an interaction of the first bot with a user that submits at least one of the user inputs;
linguistic patterns extracted from the at least one of the user inputs;
a total number of interactions serviced by the first bot during the first time-window; and
a recurrence rate associated with the total number of interactions.

Assignees

• Bank Of America

Inventors

• Dhawan Lalit
• Ravi Swetha
• Delson Kevin A
• Dande Pratap

Classifications

• G06F11/3698

  Environments for analysis, debugging or testing of software · CPC title

• G06N20/00

  Machine learning · CPC title

• G06F11/3684

  for test design, e.g. generating new test cases · CPC title

• G06F11/3692

  for test results analysis · CPC title

• G06F11/3688

  for test execution, e.g. scheduling of test suites · CPC title