Required Reading: Breaking Down Conversational Artificial Intelligence

The idea of interacting with a computer using voice or text goes way back, but it only recently became a reality with the emergence of digital personal assistants, smart speakers, and chatbots. In his new book, Conversational AI: Dialogue Systems, Conversational Agents, and Chatbots, Michael McTear, a professor emeritus of knowledge engineering at Ulster University in Northern Ireland, breaks down the underlying technology, how it is developed, and challenges for future research. CRM editor Leonard Klie spoke with him to get more details.

CRM: What is conversational AI, and what promise does it hold for businesses today?

McTear: Conversational AI refers to software that can engage in natural conversational interactions with humans. Previously these systems were known as spoken dialogue systems or voice user interfaces. Chatbots and personal digital assistants are other terms to describe conversational systems. In scientific research, conversational AI is generally restricted to systems trained using statistical, data-driven methods such as reinforcement learning or deep learning.

Around 2016, leaders of several large tech companies declared that conversational interfaces would revolutionize how people interact with computers, providing a more intuitive and natural interaction than graphical user interfaces. For businesses, conversational systems enable customers to state their problems in natural language rather than navigating menus or lists of FAQs. Automated chatbots can also handle large numbers of routine inquiries and free human agents for more complex problems.

What is driving the rise of the technology today?

The major tech companies have seen the potential of conversational systems and invested massive amounts of money for R&D. Smaller companies specializing in the technology have also emerged. And the uptake of smart speakers in homes has made conversational interaction an everyday phenomenon.

All of this has been made possible by advances in technology. Previously speech recognition was not accurate enough for everyday use, but now deep learning has resulted in dramatic reductions in word error rates and improvements in natural language understanding. These advances have been made possible by greater computer processing power to support the massive parallel computations required to run deep neural networks and the vast amounts of data to train them.

What should businesses look at when considering conversational AI?

First, they need to assess whether conversational AI meets the needs of the business and its customers, what is the expected return on investment, and how the system can be integrated with or replace current systems. Advanced conversational systems typically operate in the cloud, so there might be privacy and data protection issues to consider. Safety is also an issue, as the outputs can be unpredictable and in some cases even harmful. It is also important to understand why a particular response or recommendation was given.

Many businesses are intimidated by conversational AI. Are those fears unfounded?

There is a lot of hype around conversational AI. It is often not explained, and the technical literature is not easy to penetrate; it typically requires a background in advanced mathematical concepts and techniques. It is important for businesses wishing to adopt conversational AI to engage with vendors that understand the technology and can explain and assess its application in a nontechnical way.

What is needed to put conversational AI in place?

Conversational AI is already in place in the voice user interfaces that have been used in contact centers for more than 20 years. These systems, as well as chatbots that are being used increasingly for customer support, have been designed using best practices accumulated over many years.

Hybrid solutions that use machine learning in some parts and handcrafted rules in other parts are a good option. In most toolkits for developing conversational systems, the natural language understanding module is trained using machine learning to classify user utterances as intents and extract important information from the utterances. However, other parts of the system, such as decision processes, are usually implemented using handcrafted rules.

In some instances, use cases are sufficiently constrained so they can be handled using traditional rules-based methods and do not require the large and costly resources associated with machine learning approaches.

If there’s one thing you want readers to take away from this book, what would it be?

Conversational AI is an interesting topic with a long history, and there are many ways to create software that can engage in natural conversational interactions with humans. I aim to make conversational AI more accessible by providing a readable introduction to the various concepts, issues, and technologies.