While there are still many areas where the technologies can improve, artificial intelligence (AI) and high performance computing are strong enough that smart companies can finally begin to imitate true intelligence, CogniCor cofounder and CEO Sindhu Joseph said.
CogniCor intends to lead the way and Ms. Joseph’s background gives them a good starting point. She earned her PhD in natural language and cognitive computing and worked in academia for 15 years, generating six patents along the way. CogniCor was founded based on Ms. Joseph’s experiences while earning her PhD.
“I looked at customer engagement and thought it was possible to use AI in that area back in 2012,” Ms. Joseph said.
There wasn’t much industry talk about chatbots or emotional assistants seven years ago, but Ms. Joseph strongly believed the technology had potential well beyond conversation-centered environments, that they could expand into cognition and determine intent. But it wouldn’t be easy.
“Understanding conversations and intent are the hardest problems in AI,” Ms. Joseph said. ”They’re the holy grail.”
Standard industry technology so far hasn’t sought to understand intent as much as they replicate it. Designers take some common intents and map out the technology’s stored knowledge to associated intents.
Take the example of ordering a pizza, Ms. Jospeh said. People want a pizza, then they have to indicate the size and number. Then come the toppings and delivery instructions. Imagine the mapping involved from this one action, and don’t forget to include all of the different ways of asking that question.
That method can only go so far, Ms. Jospeh said, because it fails to address the key driver of most human conversations.
“What’s missing is shared context – that is special,” Ms. Joseph said.
As humans interact with each other they derive new intents that take the conversation in entirely new directions, but developers struggle to replicate that ability. CogniCor addresses that by doing more than simply writing intent, Ms. Joseph said They begin by developing cognitive knowledge graphs, which encode a model of expert knowledge of every domain within a context. This forms a semantic basis of context from which they can respond to more complex requests.
Understanding intent becomes more manageable with the more information you can access but developers struggle with capturing the information they already have, Ms Joseph said. CogniCor created patent-pending technology allowing chatbots to ingest enterprise data from 90 percent of sources. That is especially important as businesses grow and their chatbots have to understand more unique intents.
“With most businesses the biggest problem they have is not creating a chatbot but is feeding it information,” Ms. Joesph explained. “Crafting and maintaining intents becomes harder as the business grows.”
CogniCor’s algorithm focuses on how to immediately use enterprise technology to absorb knowledge into chatbots such as PDF service manuals and marketing brochures. That is unique in the industry, Ms. Joseph said.
“This is something very powerful. It is quick to go live,” Ms. Joseph said. “Maintenance is easy. If a document changes you just update it.”
What is the ideal blend between humans and chatbots? Ms. Joseph explained in sales and marketing environments chatbots can qualify leads by asking questions and getting people to compare different products. Once the chatbot determines the customer is ready for deeper engagement they can collect the information and transfer someone to a human agent. An Asian bank increased revenue by $100 million annually by deploying such principles.
Simple procedures also lend themselves to chatbots, Ms. Joseph said. An Asian insurance company automatically renews up to 5,000 policies every month without involving a human at any step.
“It works very well because you can save time and efficiency by handling huge volumes,” Ms. Joseph said.
This Newsmaker has been deemed by this Newsroom as having a specialized knowledge of the subject covered in this article.