Current incarnations of enterprise virtual assistants can help online consumer users find information faster and more painlessly than ever before. But today’s virtual assistants are still limited in their abilities. If machine-learning technologies continue to advance as expected, a future generation of enterprise virtual assistants is likely to be much smarter.

Most web self-service agents don’t have any trouble answering straightforward and predictable questions such as: “What’s your refund policy?” Tomorrow’s agents will go further, potentially being able to provide good answers to complex questions like: “What shoes do you have on sale that would go well with this skirt and that I can wear to work?” or “What’s the quickest and cheapest way to get from Toronto to Paris, but with a one-day stop over in London?”

Technology companies are working overtime to crack the code on machine-learning approaches and algorithms that will enable this kind of reasoned question answering. While there’s nothing wrong with sitting back and waiting for the outcome, it’s interesting to observe some of the different players and approaches involved in the challenge.

Two relevant companies that have been in the news lately are Cycorp and IBM. Neither of these companies is considered a vendor of enterprise virtual assistants and their current products are overkill for the types of use cases that today’s assistants’ handle. But will their technology contribute to the smarter intelligent assistants of the future?

Cyc Knows How the World Works
Cycorp started working towards the smartest artificial intelligence all the way back in 1984. Though they’ve kept a low profile all these years, CEO Doug Lenat recently gave several interviews to talk about how far their technology has advanced. In a Google Tech Talk from May 2006, Lenat explained both the philosophy and the details behind the company’s artificial intelligence product called Cyc. Lenat’s aim is to get the computer to truly understand the question, not just to be able to search and find an answer. To accomplish this feat, the Cycorp approach has been to construct a knowledge-modeling framework and then to painstakingly populate the framework with all-encompassing knowledge of the world.

The Cyc knowledge repository contains everything from domain specific knowledge (rain is composed of water molecules and water molecules consist of two hydrogen atoms attached to one oxygen atom) to common sense facts (you will get wet if you don’t come in from the rain). All the knowledge is represented in organized ontologies to facilitate Cyc’s comprehension of how the world works. Armed with this understanding, Cyc can apply reason when navigating its way through the vast knowledge base. Building Cyc into a system that can execute complex searches and answer multifaceted questions has been a decades-long effort involving countless hours of manual entry and categorization. All of this effort seems to be paying off.

IBM has been in the news a lot with its IBM Watson cognitive computing platform. Since proving out the underlying technology with its convincing defeat of human Jeopardy! champions in 2010, IBM has made large investments in the Watson platform and launched an ecosystem for developers and partners.

As with Cycorp, IBM has developed a machine-learning architecture for Watson. Watson’s architecture, however, is based on the open-source Open Advancement of Question Answering project (OAQA). Watson doesn’t rely on a fully populated knowledge base representing the world. Instead, the system is trained to analyze and understand questions using a combination of techniques. It then leverages a massively parallel computing architecture to quickly search through huge amounts of unstructured data that it believes relates to the question.

Finally, it applies ranking algorithms to determine which answers are the most probable. As with Cyc, though, Watson requires training in domain-specific concepts before it can apply its autonomous searching and ranking algorithms. Once that training is established, Watson can be fed knowledge from an infinite number of sources. It’s interesting to see that IBM actually advertises for “Content Ingestion Engineers” to work on Watson.

A Higher Level of Intelligent Assistance
Will one of these technologies produce a virtual assistant that’s clever enough to understand and respond to complex questions? The jury is still out. Google, Facebook and others are also investing heavily in machine learning that could lead to smarter assistants. We can bet that the market leaders in web self-service virtual assistants aren’t standing still either. Regardless of how smart the technology becomes, though, chances are good that deploying a virtual assistant won’t be as easy as flipping a switch. It’s likely that businesses will still need to invest in some level of manual training and configuration to equip the virtual assistant with the information and concepts it needs to provide the best customer support possible.

The transformative potential of these technologies can hardly be over estimated. Truly smart virtual assistants will be able to assist customers in ways that humans can’t match. Think of the sample complex question we asked earlier about the quickest, cheapest method to get from Toronto to Paris, but with a daylong stopover in London. Even an experienced travel agent is going to need time to research the options and come back with an answer. This type of research could take an hour or more. The future virtual assistant will most likely be able to understand the intent of the question, access data about all possible travel options, compare and contrast them, and return the highest ranking options within minutes, if not seconds. The assistant will probably even be able to factor in the historical on-time average when selecting the best option. With that kind of power in the hands of the enterprise, the possibilities are limitless.

Amy Stapleton writes the Virtual Agent Blog providing an independent viewpoint on intelligent virtual assistant technologies, products, and markets. Stapleton is an information technology professional with over two decades of experience in enterprise applications. Her background includes stints at SAP, IBM, and U.S. federal agencies as a software project manager and consultant.

Categories: Conversational Intelligence, Intelligent Assistants, Articles

Tags: intelligent assistant, customer care, IBM