informa
/
4 MIN READ
News

Building a Chatbot That Humans Will Actually Like

A new generation of friendlier, more intuitive chatbots is on the way. Will users learn to love them?

Pity the poor chatbots. People love to hate them. Angry and irritated users have been known to say things to a chatbot that, in the real world, would likely result in a lawsuit or a punch in the nose.

Badly designed chatbots are irritating. “We all have high expectations for human interaction, and we all think in slightly different ways,” observes Wayne Butterfield, a partner with ISG Automation, a unit of technology research and advisory firm ISG. “If these differences aren't taken into account in the design of a chatbot, it will likely lead to a frustrating experience.” He notes that user frustration with chatbots has long been more common than user satisfaction. “There's an almost universal dislike right now of this capability,” Butterfield states.

While chatbots do a generally good job of handling simple requests, the technology shows its limitations when encountering complex queries that haven't been pre-defined by the provider. “This jumpstarts an endless cycle of ‘Sorry, I don’t understand that request’ responses, which leads to customer frustration,” says Vamsi Kora, chief data strategy officer at Apexon, a Silicon Valley digital engineering professional services firm.

Another reason why so many people dislike chatbots is the technology's inherent lack of empathy. “Humans seek empathetic responses to their questions that quell their anxieties, which chatbots today are not capable of providing,” Kora explains.

While chatbots are generally AI-driven, they're still not as intelligent as human beings when it comes to one-on-one communication, says Beibei Li, an associate professor of IT and management at Carnegie Mellon University’s Heinz College of Information Systems and Public Policy. “Moreover, they are perceived as less trustworthy, due to privacy concerns,” she notes. “Sometimes, humans speak idioms or jargons that are hard for computers to understand.”

Some scholars have even accused chatbots of exhibiting bigoted behavior. “Language models that drive conversational AI have shown a strong presence of biases and prejudices, based on the training data, which can lead to negative and even harmful effects,” Kora warns.

According to a recent University of Washington research paper, when language models grow too large, they become difficult to comprehend. “This results in models that encode stereotypical and derogatory associations along gender, race, ethnicity, and disability status” Kora says.

Necessary Chatbot Attributes

A strong, friendly conversational ability, and speedy request fulfillment, are two key traits in a successful chatbot, Butterfield says. “A good chatbot will make sure customers get what they need or has the ability to quickly pass customers on to a person who can fulfill that need,” he explains. “Customers who get their questions answered and their requests fulfilled are almost twice as satisfied as those who only get generic answers to their questions.”

Yet in their quest for chatbot excellence, enterprises often underestimate just how many design and development resources will be needed to build a great chatbot. “You need to figure out how to continue the conversation and solve the customer's problem,” says Kerry Robinson, vice president for conversational AI strategy at contact center solutions provider Waterfield Tech. “We don’t yet have technology that can learn that kind of thing from data, so it needs to be carefully designed and programmed into the system.”

What makes human agents effective isn't just their ability to listen and understand, Robinson notes. “A great agent knows how to guide a conversation and solve customers’ problems.” Chatbots have a long way to go to before equaling that capability.

Chatbots in Transition

Chatbots can be improved by moving away from rule-based systems and toward more natural conversational AI. “This technology enables chatbots to answer more complex queries that replicate human interaction,” Kora says.

As they consider a transition to chatbots, enterprises will need to think differently, Robinson says. “With chatbots and other forms of conversational AI, you’re not just building a piece of tech, you’re deploying a bot workforce, and they need leading and managing just like human agents do.”

As chatbots move into the realm of generative AI, the technology is likely to advance significantly, allowing very specific answers to be generated directly from existing data sources, versus today's highly scripted responses, Butterfield predicts. Generative AI, plus multi-modal chatbots, which can use more than one type of AI, will allow more contextual support, even to the point of recognizing and responding to the customer's emotional state, he adds.

Looking Ahead

In the days ahead, chatbots will continue moving toward more natural conversational AI, allowing increasingly complex, human-like conversation. Future chatbots will incorporate voice-based models inspired by virtual assistants, such as Alexa and Siri, Kora says. “With natural language processing, natural user interfaces, and cognitive AI, we’ll see chatbots handle more proactive engagements with end users.”

What to Read Next:

Conversational AI: How It Works and Where It's Headed

What Some Customers and Employees Hate About Chatbots

Chatbots: Approaching 60 Sure Looks Good on You

Editor's Choice
Brian T. Horowitz, Contributing Reporter
Samuel Greengard, Contributing Reporter
Nathan Eddy, Freelance Writer
Brandon Taylor, Digital Editorial Program Manager
Jessica Davis, Senior Editor
Cynthia Harvey, Freelance Journalist, InformationWeek
Sara Peters, Editor-in-Chief, InformationWeek / Network Computing