Bots Are Learning Their ABCs, But They’ve Got a Ways To Go
You learned your ABCs long ago in elementary school, and honed your reading and writing skills in the ensuing years. Now chatbots are at a similar stage in their development. While AI technology is improving quickly, it has a ways to go before it can match a human’s ability to understand and communicate with written and spoken words.
If you haven’t noticed, the bots are out in force. No longer just a fad, chatbots are being increasingly adopted by enterprises in nearly every consumer-facing industry as one of the key ways they interact with their customers.
According to a recent survey of 7,000 consumers conducted by Forrester Consulting on behalf of Amdocs, 35% of people interact with chatbots on a weekly basis. That number will grow in the coming years, according to Amdocs, which says 85% of customer interactions will be with “software robots” in five years.
However, it’s not all digital roses and unicorns. Today’s chatbots tend to be “clunky” and fail to understand nuances of their human charges, according to Amdocs, which provides software for telecommunications software. The company’s survey concluded than 83 percent of consumers would rather to speak to a human.
If you’ve ever “zeroed out” to get to a human, you likely have experienced this same situation. The biggest foibles of chatbots, Amdocs found, are a lack of understanding of complex requests, an inability to deal with multiple questions, a lack of personalization, and no understanding of human emotions.
NLP to AI
However, technological improvements in the bots brought via AI will make those engagements go much smoother, says Doron Youngerwood, Amdocs’ product marketing manager of big data and AI.
“On the one hand, people are keen to use chatbots,” he tells Datanami. “They may have had a bad experience in the past, but with technology coming to fruition now” the experiences will be much better.
The majority of chatbots in use today are rules-based bots that can respond to a narrow band of possible situations, Youngerwood says. “You ask a simple question and get a very simple answer back. That’s it,” he says. “They’re very limited in their capability.”
As more advanced AI technology makes it into the field, the chatbots will get more lifelike, which should result in a more pleasant experience for human customers. Sentiment analysis, for example, is one area where AI technology can really improve the conversation over the current generation of natural language processing (NLP).
“NLP been around a number of years. It’s hardly an exciting functionality,” Youngerwood says. “You bring on board sentiment analysis and personalized recommendations…that’s something that artificial intelligence can do on the backend. You bring all the data that you have on a customer, and use something called NBA, or next best action, to personalize the response to the customer.”
But don’t expect any one AI-powered platform to rule the chatbot roost. Amdocs says that each chatbot system will need a deep level of industry-specific capability to make it useful for customers. That’s another factor that’s hurting the image of today’s chatbot systems, the company says.
“I think many of our customers today are getting fed up with the bots that don’t have specific industry intent,” Youngerwood says. “If somebody is engaging with a bot and says something along the lines of ‘I dropped my Galaxy,’ the bot has to understand that Galaxy is a phone. But many bots wouldn’t necessarily understand what Galaxy means and be able to respond to the customer inquiry.”
NLG to ML
Another technology firm that’s looking to improve human communications through robotic automation is Narrative Science. The Chicago-based company develops a platform called Quill that combines data analytics with natural language generation (NLG) capability.
While NLG has been around for decades, it’s still a fairly nascent technology, at least in the public eye, says Narrative Science CEO Stuart Frankel.
“NLG has been around for a long time. The National Weather Service, since the early 70s, has used NLG to generate little snippet weather reports and provide other information,” Frankel tells Datanami. “Really the big breakthrough for us was, instead of just taking data and turning it into language…we’re actually putting a step in between and we’re applying the analytics.”
That’s the hard part, it turns out. It would be a simple matter to generate words from every little piece of changed data. But what Narrative Science excels at is figuring out how to identify when an important piece of data changes in a meaningful way.
“It’s not knowing how to say something,” Frankel says. “It’s knowing what to say, and identifying those important things to say and then rendering those in a natural language.”
Since the company was founded in 2009, it has accumulated 80 customers, including some large enterprises like MasterCard, Credit Suisse, and American Century Investments that use Quill for things like generating suspicious activity reports, writing narratives on mutual fund performance, and provide customers with personalized account reviews.
“Companies are starting to adopt Quill to personalize and enhance communication to customers using data,” Frankel says. “It gives employees more time back and allows them to spend it on higher value-add activities.”
No AI Yet
Quill mostly uses well-understood mechanisms, such as statistical measures or user configured flags, to identify what changes are important enough to write a report about it. The system starts with the idea of intent, Frankel says.
“It starts with an understanding of what is the desired communication, what are we talking about,” he says. “By giving Quill the input, it has a significant amount of information that it then uses to figure out what are the important facts and insights that are relevant to the user at that point.
Narrative Science is just starting to experiment with machine learning and other AI techniques. However, the company is not ready to take the leap into more advanced AI approaches, such as deep learning.
“I think certainly you’ll see, over time, Quill take advantage of things like deep learning,” Frankel says. “We’re already adding certain compounds of machine learning to the system. But we also want to make sure that users have control, that they understand what the output is going to be, rather than hoping that Quill is going to write something accurate, appropriate, and compliant.”
Frankel says he hasn’t seen many success stories of companies using AI platforms outside of the halls of Web giants like Google, Facebook, and Microsoft. “We can explain what we do, how Quill works, and importantly, measure the ROI very specifically,” he says. “Companies aren’t going to invest heavily in these types of technologies and not a see an immediate return on investment.”
We constantly hear how AI is going to change everything, including how humans communicate with each other and with organizations. The technology is definitely promising, but we’ve still got a way to go before we get there.
Machine Learning, Deep Learning, and AI: What’s the Difference?
Training Your AI With As Little Manually Labeled Data As Possible
Gaining Control Over AI, Machine Learning