How Machine Learning applies to Virtual Agents?
Machine Learning is a vast field and its importance has grown significantly over the last few years. It now plays a key role in the performance of Virtual Agents. However, people are still confused about the relation between Artificial Intelligence, Machine Learning and Deep Learning. Going back to a clear definition is a sure start.
Artificial intelligence in a nutshell
Artificial Intelligence started as an official academic discipline in the mid 1950s, with the early stages of computer sciences. But the idea of “thinking machines” has been a mythological and philosophical quest since the dawn of humanity. This heritage blurred our understanding of what can be expected of a machine with “intelligence”.
Sticking to the factual behaviour of an AI, its purpose and practical applications helps us to go beyond the myths. Artificial Intelligence in fact describes a machine-intelligence that mimics "cognitive" functions that humans associate with the human mind, such as "learning" and "problem-solving".
This means that business applications of AI can only derive from a clear identification and understanding of the problem that needs solving. Because a machine does not (yet) have a “mind of its own” to decide to apply its skills to problem-solving!
Examples of problem-solving with AI and Virtual Agents
Artificial Intelligence is already solving problems in many fields, including transport (autonomous vehicles), health (detect cancer cells at earlier stages), agriculture (crop and soil monitoring), Finance (market analysis and data mining), and the military (intelligence data collection).
One of the most interesting applications of AI machine learning technologies are Intelligent Virtual Agents. Those conversational bots are capable of interacting with humans thanks to language understanding. Virtual Agents can engage in conversation with users, through live- chat, email but also voice, understand human language and the intents behind requests.
What is at stake here is to generate productivity gains and improve users satisfaction.
The most common case is to allow users to solve a password reset or account lockout problem, by chatting with or calling an Intelligent Virtual Agent. Since 40% of call centers interventions relate to passwords issues (according to Gartner), letting a Virtual Agent handle them can save time and budget for the IT Service Desk.
Advanced conversational skills and intent understanding also enable a Virtual Agent to handle more complex IT requests and Troubleshooting. This capacity to perform a semantic analysis and understand a user’s incident and troubleshoot it is at the core of the IT Service Desk’s mission.
But first, the Virtual Agent needs to be able to understand if the user’s issue is in its scope of expertise or not. He does this by matching the request with its Knowledge Bases, FAQs and Enterprise content it has been allowed access to.
Then, the Virtual Agent can guide the user through a complex procedure within a conversation (for instance requests to “connect to a VPN” or “Free up hard drive space”). The agent can also propose choices and clarify vague or not precise user statements.
If the request is outside the scope of its computer program, the Virtual Agent can still propose solutions to help the users progress with their requests. For instance offering to email an HR contact for an HR related question.
We see here that understanding languages and intents enables the Virtual Agent to do more than just fetch an answer in a database.
The idea is to engage in a conversation that flows naturally, adapt the answers to the users requests and context. It is also important to reassure the users at every step of the way, to guarantee their satisfaction.
Let’s now focus on the technology that supports those conversational skills : Machine Learning and Deep Learning.
The difference between Machine Learning and Deep Learning
Machine Learning (ML) is a subset of Artificial Intelligence. ML is the study of computer algorithms that improve automatically through experience and training on data thanks to learning models and learning techniques. .
Intelligent Virtual Agents use a subset of Machine Learning called Deep Learning. DL is based on Neural Networks, modelled loosely after the human brain that works with an overlay of neurons, and used to detect patterns.
This Neural Networks approach for Deep Machine Learning is the foundation for Natural Language Processing (NLP), Understanding (NLU) and Natural Language Generation (NLG).
How Virtual Agents understand Natural Language
What are those patterns that Neural Networks can identify during a written or vocal conversation with a human ? Patterns are “semantic similarities”, recognizable combinations of words and phrases that the Virtual Agent will use to understand a user’s query and carry forward an intent-based communication.
The Virtual Agent may need to ask additional questions to really understand what the user means or needs, and match this request with answers coming from its Knowledge Bases.
Coupled with RPA (Robotic Process Automation), the Virtual Agent can even orchestrate multiple actions like sending emails or generating passwords, Wi-Fi codes...
Because the idea is not to plan and classify every possible query that could come from a human being. The goal is to create an NLU-NLG engine that will improve its understanding of human conversations over time thanks to learning algorithms. It will learn from a wide variety of data sources like existing knowledge bases or real conversations with users.
Deep learning as a foundation for growth and change
The strengths of the Deep Learning model comes from this capacity to learn from its dataset and accept a wide variety of data.
It enables a Virtual Agent to understand and answer questions, but also to handle context-switching and digressions for better speech recognition when talking to a human. This means that even if a user skips some questions, the Virtual Agent will not be stuck in a scenario but adapt in real time to the conversation.
That’s why Virtual agents are considered like a “digital workforce”, working in partnership with humans. In today’s IT and Digital Workplace environments (like Microsoft 365) where new tools and software are regularly deployed, end-users need to adapt quickly to the digital transformation.. The learning curve is very fast for a Virtual Agent already pretrained on Knowledge bases (like Microsoft’s or ServiceNow’s) and on real conversations with end users.