Thomas Friedman on Artificial Intelligence: It Still Needs Human Intelligence For Now!

Dear Commons Community,

After fifteen years, New York Times columnist, Thomas Friedman revisited Bangalore  (India’s Silicon Valley) and found that a lot has changed especially with how artificial intelligence (A.I.) is being deployed in various activities.  He describes how one company that specialized in customer service has retooled dramatically as A.I. has taken over much of the actual interactions with callers needing assistance.  Here is an excerpt.

“BANGALORE, India — Fifteen years ago I came to Bangalore, India’s Silicon Valley, to do a documentary on outsourcing. One of our first stops was a company called 24/7 whose main business was answering customer service calls and selling products, like credit cards, for U.S. companies half a world away.

The beating heart of 24/7 back then was a vast floor of young phone operators, most with only high school degrees, save for a small pool of techies who provided “help desk” advice. These young Indians spoke in the best American English, perfected in a class that we filmed, where everyone had to practice enunciating “Peter Piper picked a peck of pickled peppers” — and make it sound like they were from Kansas not Kolkata.

The operations floor was so noisy from hundreds of simultaneous phone conversations that 24/7 installed a white-noise machine to muffle the din, but even then you could still occasionally hear piercing through the cacophony some techie saying to someone in America, the likes of: “What, Ma’am? Your computer is on fire?”

Well, 24/7’s founders — P.V. Kannan and Shanmugam Nagarajan — invited me back last week for an update. Their company is now called [24] and their shop floor is so quiet that the operators are encouraged to play their own music. The only noise is from the tapping on keyboards, because every query — from customers of U.S. retailers, banks and media companies — is coming in by text messaging from smartphones, tablets, desktops and laptops.

These text queries are usually answered first by a [24] chatbot, or “virtual agent,” powered by A.I. (artificial intelligence) and only get handed over to a person using H.I. (human intelligence) if the chatbot gets stuck and can’t answer. The transformation of [24] from perfecting its accents to perfecting its insights illustrates in miniature how A.I. is transforming the whole work landscape.

In a nutshell, the U.S. and Indian middle classes were built on something called the high-wage, middle-skilled job. In an A.I.-driven world, such jobs are becoming extinct. Now there are mostly high-skilled, high-wage jobs and low-skilled, low-wage jobs, and a dwindling number in between.

Virtually all of the [24] human operators today have college degrees, because they need to be able to text with good grammar in English, understand the interaction between the chatbot and the person calling for service and communicate with expertise and empathy when the chatbot runs out of answers.

At the training class I sat in on last week, Peter Piper was gone. He was replaced by a competition among trainees over who could grasp first exactly when the chatbot — which [24] calls by the woman’s name Aiva, for Artificially Intelligent Virtual Assistant — could no longer understand the “intent” of the customer and what that intent actually was.

It’s at that critical point that the human agent not only has to step in and answer the question that Aiva couldn’t, but also to “tag” the customer’s queries that stumped the bot and feed them to [24]’s data scientists, who then turn them into a new, deeper layer of artificial intelligence that enables Aiva to answer this more complex query the next time. (Kannan is about to publish a book on A.I. called “The Age of Intent.” )

The data scientists who figure out the upgrades for chatbots that handle text are called “digital conversation designers.” For another, small part of the business, data scientists for chatbots that speak in computer-generated natural language are called “voice conversation designers.”

“It’s a cool job,” Santhosh Kumar, a 45-year-old conversation designer, who came up through the 24/7 system, said to me. “You are designing what the chatbot should be saying to the customers.” It is all about “how to make a computer sound like a human.” Banks want their bots to be formal; retailers prefer more conversational bots.

Another new term I learned here was “containment.” That measures how deep into a conversation your chatbot can go without having to hand the customer over to a human agent. A company’s “containment rate” is like its A.I. batting average.

Today, [24]’s containment rate ranges from 20 percent to 50 percent of queries, depending on the company it is serving. Its goal is 80 percent. As the bots grasp more of each customer’s intent, the skilled humans are redeployed to more complex services and sales, and that, said Kannan, “turns into better sales and keeping customer satisfaction high.”

His chatbots, Kannan explained, are built with a “negative sentiment detector” to identify angry customers, so “we auto-generate sympathy when we can,” but for the most part “complexity and empathy” are left to the humans.

Hollywood and Bollywood movies lately “have created a really bad impression that robots are going to take over,” said Irene Clara, a trainer. “I don’t think that fear is justified. I think we grow together. When you’re teaching Aiva, you’re getting skilled yourself, and without that Aiva becomes incompetent.”

So — for now — if you have critical thinking and empathy skills, Aiva is your friend. But I wonder what happened to all those Indian high school grads I met 15 years ago. Because if you don’t have those skills — and just have a high school degree or less, which applies to hundreds of millions of Indians — or you are doing routine tasks that will be easily roboticized, well, Aiva the robotic fruit picker, Aiva the file clerk or Aiva the trucker will not be your friend.”

Friedman’s conclusion:  A.I. has a lot to offer and for now it depends upon the human touch but this may only be an intermediate stage.


Comments are closed.