Applied Machine Learning and AI – “You said what?”
Machine learning and Artificial Intelligence are rapidly moving from the realm of research to business and consumer application to power critical functions of businesses like Google, Facebook, and Amazon.
AI can generally be defined as “software that perceives its environment and adapts to new information, predicts outcomes to make decisions and take action, and continually learns from every interaction to optimize towards its goals”.
The type of AI we hear about today generally falls into the category of Weak or Narrow AI, which is “software that equals or exceeds human effectiveness or efficiency for specific tasks”. In the context of robotic process automation, this can be extremely powerful, since almost all tasks that make up a process are domain bound and deal with problems in a very specific context.
This article focuses on how Natural Language Processing is powering RPA.
Natural Language Processing
Natural Language Processing (NLP) refers to a machine’s ability to understand spoken language and Natural Language Generation (NLG) is the machine’s ability to generate natural language. The technology has made significant advances, however, it is still not mature enough to be trusted to accurately determine the intent of a request and function without human validation. Furthermore, today’s NLP technology is geared towards only serving a specific domain.
A simple test I often do to demonstrate NLP capabilities and context awareness is to ask Siri or Alexa “I have 3 apples and my friend gave me 2 more apples. How many do I have now?” Although a simple question that a 4-year-old can answer, Siri or Alexa were not designed for this context and hence have a problem determining what is being asked of them.
The lack of straight through processing is evident in the use of ChatBots where upon any ambiguity the ChatBot tries to confirm the intent of the user by asking “You want to … ?” type questions. Within RPA this would fall into the human intervention bucket.
Despite these shortcomings, NLP has been very helpful in specific areas in processing various incoming unstructured text. Two examples of this are given below.
AI and Machine learning examples
For a customer who needed to improve their helpdesk SLAs for password reset type requests, we used a commercially available NLP processing service and trained this service with a number of previous requests identifying a “password reset” request. We learned that it was also very important to train the service on requests that may contain the words password and reset and may not be a password reset request, e.g. “Please do not reset my password”. The service evaluated every incoming ticket and if it determined with an 80% level of certainty that the request may be a password reset request, it was routed to the top of the queue for human operator verification. Once verified, the execution bot was able to perform the actions required to reset the user’s credentials.
Sentiment analysis capabilities of NLP services can also be very useful in determining tones and emotions, especially in customer service scenarios. Sentiment Analysis can be used today for dynamic Brand Management by monitoring Company’s Consumer facing Digital Channels. For a particular customer, we needed to determine if the response to customer service request was positive or negative. We were able to use a commercially available NLP service and determine the sentiment of the customer response to a ticket resolution and if the results showed a negative sentiment, the issue was automatically escalated to the top of the queue for further investigation. By scanning responses faster than any human ever could, we were able to reduce the response times drastically.
AutomateWork is on a mission to help their Clients on how they can automate Business Processes using RPA, BPM, AI and Cognitive technologies. AutomateWork focuses both on RPA and “Smart” RPA Technologies (http://www.automatework.com)