Businesses are delving into the world of artificial intelligence with virtual assistants that can help customers do everything from matching an outfit to calling an Uber – without waiting for the next available representative.
These days, when consumers use an app to ask an H&M stylist for advice or a Chase Bank teller to check their balance, chances are the “people” on the other end are chatbots.
These virtual assistants are programs that converse like humans in writing. They’re all the rage in customer support. Chatbots can be faster and more accurate than people, and they’re always available.
“Sometimes customers actually don’t want to pick up the phone and talk to a human — it’s the middle of the night, and they just want to get something done quickly,” said Mark Beccue, principal analyst at research firm Tractica.
These assistants offer quick and natural responses to conversational requests. For example, H&M’s chatbot can suggest accessories that complement a shirt, and other bots can use Facebook’s Messenger chat platform to call an Uber, or use WeChat to send payments via several Asian banks.
Bank of America and MasterCard recently announced they would use chatbots to allow customers to ask financial questions, initiate transactions and get advice via text messages or services like Facebook Messenger and Amazon’s Echo tower, reported the New York Times.
Advances in natural-language processing and machine-learning programs — plus the availability of an increasing amount of data to train them — are making chatbots more effective at these kinds of jobs. Although there is increasing hype about the chatbots, Beccue said the technology has improved greatly over the last year.
Chatbots are highly visible examples of artificial intelligence (AI) in use by businesses today. They aren’t what some call “strong” or “full” AI systems that can hold on-going conversations that are indistinguishable from human discourse. But they’re doing real work and taking on a broader range of tasks. They provide a window into the promise — and the challenges — AI holds for businesses, according to Bob Rogers, chief data scientist at Intel.
“We see a lot of places like this where AI is really primed to augment what people can do, and make life easier and more interesting, and take better advantage of our own human capabilities,” said Rogers.
Challenges for Chatbots
Consumer tools like Apple’s Siri now claiming to achieve an accuracy rate of about 95 percent. But there’s still a lot to learn.
Language is complex and nuanced. Even in written form, which lacks complications like accents, muttering and background noise, accurately interpreting language requires detecting misspellings and sarcasm, as well as analyzing cultural context.
Humans are good at picking up on these nuances. Machines, however, face a steep learning curve, according to David Reitter, co-director of the Applied Cognitive Science Lab at Pennsylvania State University.
“People understand natural language because they understand the world and the environment in which words are spoken,” said Reitter. “Language simply doesn’t provide enough detail and precision to be understandable or learnable without knowing so much more.”
One of the most important ways chatbots improve is through the experience they gain in every interaction through machine learning.
“The feedback loop is critical,” explained Rogers. “Companies structure these systems to observe how the user responds to the answer they’re providing and use that data to improve the answer the next time.”
Of course, there’s also a potential downside: Twitter users, for example, recently “trained” Tay, Microsoft’s experimental Twitter chatbot, to spout Nazi propaganda.
Social engineering and hacking are two reasons businesses need to think twice before making a chatbot an official company voice.
In a study of chatbot performance, Wannemacher and his colleagues found that about one-third of the time, chatbots “either failed to complete the consumer’s request or provided a clunky, awkward experience.”
When money is at stake, he wrote, “people are less forgiving.”
Having a Niche is Key
Chatbots perform better when they’re focused on a specific set of tasks — finding a hat to complement a pair of boots, for example — rather than trying to comprehend everything a human might say or ask, according to Beecue.
“Where the chatbots and assistants will grab hold is where they can focus on a narrow domain,” Beccue said. “That bodes well for enterprise applications like customer care, where you handle a lot of common end-user questions.”
Listen and Learn
To improve chatbot effectiveness, Beccue said some companies are looking at how to intermingle virtual and live agents.
“Let’s say you start with a virtual agent, and it gets stuck,” Beccue said. “Like a robotic customer service agent on the phone, it can say, ‘I see we’re not getting there; would you like to talk to a live agent?’”
The virtual assistant could continue to listen to and extract information from the human agent’s solution.
“Then that answer can be submitted to an editor and approved, and we’ll add that human-approved answer to the AI’s domain expertise,” Beccue said, explaining that this is how the chatbot will improve its responses over time.
It’s a good illustration of AI’s potential to work together with humans for better results than either gets alone. If the stylist chatbot doesn’t know the ins and outs of power clashing, certainly the fashionista on the other end of the phone will.