AI Powered Chatbots: Reality vs Expectations

Smartphone with finance and market icons and symbols concept

There is a lot of excitement surrounding chatbots, about how they can revolutionize the way banks and financial institutions serve their clientele. Some fintech bloggers are making predictions about widespread use of chatbots in the fintech industry for the coming year.

However, as it is usual with evolving tech, the capabilities of an AI powered chatbot are in the early wave of inflated expectations of a hype cycle.

Chabots are often advertised as sentient “virtual assistants” that can cater to your requests like a human being. But the truth is, currently existing chatbots will get confused if you ask a question in a slightly different manner than what it was programmed to answer it.

For example: If you ask CLEO a question that he is not prepared to answer, he simply replies, “OK”.

And then there’s cleverbot, one of the most popular “AI-powered chatbots” ever to grace the internet. Cleverbot was created by British AI scientist Rollo Carpenter and went online in 1997.

Ever since then, cleverbot has had hundreds of millions of conversations. That is a massive database to learn from. And yet, even after 20 years of being online, cleverbot’s conversational skills leave a lot to be desired. You don’t have to take my word for it. Try talking to cleverbot yourself.

According to John Lim, CTO of a malaysian fintech company that has expertise in building both chatbots and AI; the excitement over chatbots is somewhat justified. Especially when they are used in banking. “Currently, most chatbots work using a rule-based technology. They are programmed to respond to thousands of hardwired rules. Any query outside those rules, or asked in a manner that doesn’t fit into those rules, will elicit a generic response like “OK” or “Is that so?” or something completely nonsensical from a human perspective. The whole point of having an AI is to facilitate “human-like” conversations. Bots that work using rule-based technology are not intelligent enough to accomplish that. In fact, last year Microsoft’s Twitter chatbot “Tay” caused quite a scandal by retweeting racist, misogynist, and anti-semitic Nazi sentiments. Chatbots still have a long way to go before they can intelligently discuss a concept that they have never encountered before. They still fumble when faced with a question or statement that was not included in their hardwired rule set.”

“However, we have reached a certain level of effectiveness with our chatbots in specific domains, like banking. And the potential for these bots is significant in providing banking related information to customers on demand. We can do this to scale for services like micro-loans, where it is more viable to use a chatbot to deal with large number of customer enquiries.”

“As for visually identifying documents and people using AI, we still have a long way to go. Even Apple’s advanced facial recognition system can’t differentiate between twins and family members who have similar facial features. It’s important to have realistic expectations where AI is concerned, and not be influenced by hype blogs and sci-fi movies”

John believes that the biggest roadblock in creating a truly intelligent chat-bot is the availability a of a sufficiently large set of conversations that capture all scenarios of human interaction. Giants like Amazon and Google have access to such big data generated by millions of people every day. For them it is viable to build self-learning models that will eventually be used to build intelligent chatbots.

But for other companies that don’t have access to such large volumes of data to work with, creating an intelligent chatbot that can carry a conversation that is superior to cleverbot’s can be difficult and expensive.

“In the future, the real potential for chatbots lies in their ability to comprehend voice commands. The ability to execute voice commands will democratize access to digital banking services to consumers who are not tech savvy, and deliver a superior customer service experience. Therefore, companies with massive voice recording samples at their disposal are naturally at an advantage. Google Assistant and Microsoft’s Cortana and even Apple’s Siri all have a competitive edge over any emerging chatbot start-up. Billion-dollar corporations like SAP cannot hope to compete with them, unless they spend a fortune on hundreds of thousands of voice actors to generate samples for their AIs. Also, the success of voice based chatbots depends a lot on companies such as Google, Amazon, Microsoft, Tencent and Apple opening their APIs. From my experience, I can safely predict that these giants will make their APIs accessible to independent developers, in the hopes of developing more intelligent chatbots.”

By | 2018-04-25T17:43:51+00:00 2nd February, 2018|Insights|