UW prof worries AI could induce 'rampant disrespect and awfulness'
A woman looks at a Google Pixel 2 phone in San Francisco, on Oct. 4, 2017. (Jeff Chiu / THE CANADIAN PRESS / AP)
Michael Oliveira, The Canadian Press
Published Friday, May 11, 2018 12:24PM EDT
Aimee Morrison doesn't look forward to the not-too-distant future when she might have to constantly doubt whether she's speaking on the phone with robots powered by artificial intelligence, or reading emails composed by algorithms.
Earlier this week, Google unveiled demos of new A.I. services that had the web abuzz, including Duplex, which would allow users to outsource the drudgery of booking appointments with businesses by phone to a virtual personal assistant.
Google released recordings of calls it says were placed to businesses -- including booking a restaurant reservation and a hair salon appointment -- in which the employees answering the phone seemed to have no clue they were interacting with a robot.
In calling about the restaurant reservation, Google's A.I. was able to seamlessly handle a series of questions in a nearly minute-long conversation and was not flummoxed when told a booking wasn't necessary since the eatery wouldn't be busy. In both calls, the computerized voices occasionally dropped some "umms" and "mm-hmms" in the script to appear more life-like.
"The Google Duplex technology is built to sound natural, to make the conversation experience comfortable. It's important to us that users and businesses have a good experience with this service, and transparency is a key part of that. We want to be clear about the intent of the call so businesses understand the context. We'll be experimenting with the right approach over the coming months," Google wrote in a blog post about the technology, adding "it cannot carry out general conversations."
"I live in fear of a Google Duplex world where I have to make a hair appointment and the person on the other end treats me like a robot and is awful to me and how am I going to prove that I'm a human?" said Morrison, an associate professor at the University of Waterloo who studies technology's impact on culture.
"Is it going to make all of our conversations with each other deeply suspicious and mistrustful? Is it going to encourage rampant disrespect and awfulness as we learn that some human voices are not really human?"
University of Toronto Prof. Gerald Penn called the execution of the Google demos "brilliant" but said that's partly because the A.I. targeted an unsuspecting target. Someone who knew they could be talking to Google's software would likely be able to outsmart it, added Penn, who studies natural language processing.
"There are going to be some rigid boundaries as to what Duplex can do, and if you're interested in tricking the system and determining if it's a human on the line or not, you could certainly do that," he said.
The technology could eventually have a significant impact on the customer service industry, which would have to begin catering to robot callers, he added.
"What's going to happen to small retailers when a significant part of the phone calls they're receiving are not from people? Because that could turn into a kind of negative spiral where they're disincentivized to provide any kind of decent phone support, because they're just talking to robots anyway."
Google also unveiled a feature for its Gmail service called Smart Compose, which provides users with suggested strings of text for emails. Beyond simply predicting the word the user is typing, the software tries to anticipate their whole thought and attempts to offer complete suggested sentences.
"Because they're running Gmail they get all the data, all the emails that people compose, and they can probably train a system to provide replies that would mimic how others have composed their emails before and provide that to users who can simply select some potential replies," said Prof. Pascal Poupart of the University of Waterloo, who studies artificial intelligence and machine learning.
"And the beauty of this system is whenever people are going to select something, that provides a feedback that Google can then use to further improve (the service) because it will know which types of replies are effective."
He found Smart Compose less controversial than Duplex, noting "it's common already that very, very busy people will often have an assistant who replies to emails on their behalf."
"There's already this kind of delegation happening and now the only difference is we're delegating that to a machine," he said.
But Morrison is disturbed by the potential outcomes of the feature becoming popular.
"Duplex is trying to make machines pass as humans and what the email assistant is trying to do is make humans sound more like machines. So they're taking the human out of the conversation at both ends," she said.
"What those things have a tendency to do is really collapse the variety and joy of human conversation into a serious of stock phrases ... and literally erase other ways of expressing yourself."