Would you care if your customer support representative was a machine rather than a person as long as you knew you were getting the best service? And would you be able to tell the difference?
IBM IBM -1.21% is betting you wouldn’t, on both counts.
The U.S. tech giant’s research lab in Haifa, Israel, is testing new software which it says can recognize the basic emotions of customers typing into a company’s automated online chat services or even tweeting with a company representative. The software is trained on texts from Twitter TWTR -0.25%, chat formats and email, as people express themselves differently in those mediums. In the near future, there are plans to develop the software to include voice calls. It’s an attempt at improving customer support software so that big clients can cut costs on employing human staff.
IBM says the plan is to next month roll out prototypes of the software for some of its customers. Many companies use live chat services on their web sites. The services involve windows that pop up on sites offering a live chat with a representative for support, advice or service.
Often, that representative is a software program, not a human, even though many sites will place a person’s face and name alongside their chat windows.
IBM says its new software adds a layer of emotional analysis on top of existing programs.
“Our goal is to show that we will get at least the same level of customer satisfaction that you have with good, human customer support person,” said Dr. David Konopnicki, who manages the Information Retrieval group for IBM Research in Haifa. Konopnicki’s team in Haifa has been working on these “computerized dialogue agents” for the past year.
The software looks at a variety of data. They include the words typed, their grammatical structure, the context of the conversation, and even metalanguage like pronounced use of exclamation marks (!!!!!!), writing in capital letters, slang like OMG, LOL, and that old standby for really irate digital communicators, WTF, and even emojis. The software then makes inferences on the emotional state of the user. In an instant messaging format, the software registers a change in the speed of typing. If typing speed increases, it could mean the user’s emotional experience is intensifying. Significant pauses between sentences could mean the opposite.
Based on these parameters the computer program will adapt its approach and even its language, and will decide whether it can respond to the customer itself, or whether to escalate the chat to a human.
“Sometimes you have to help people calm down by explaining to them the efforts that you’re going to exert to solve their problem. This includes apologizing to them. We’ve created an agent that is empathetic to humans,” Konopnicki told The Wall Street Journal.
For instance, the smart agent will know that if a user is calling for the third time, this user’s emotional context could be charged. An upset customer could be put through to a higher level support operator, possibly a human, quicker.
“The objective right now is not to fully automate and replace the human but to make the automated system more emotional,” said IBM Global Labs spokesperson Jonathan Batty.
IBM’s work takes existing smart digital assistants, like Apple’s Siri and Microsoft ‘sMSFT -0.66% Cortana, a step forward, in that the latter don’t yet use their users’ vocal intonations and grammar to infer emotions.
That could be next. Microsoft, Google GOOGL -1.44%, Facebook FB -1.50% and others are investing in natural-language recognition and human emotion analysis programs, an increasingly popular and lucrative field of artificial intelligence. Facebook, for example, is teaching its software to analyze speech, facial expressions, body language and social context in images and videos.
IBM’s Konopnicki thinks companies should still be up front with users about whether they’re talking to a human or a machine. “If you’re not happy with the machine we can escalate you to a human,” he said. “If the machine is good enough people will become used to it and it will become natural.”