Google Duplex, the latest machine learning experiment to come out of Mountain View, isn’t your average chatbot. It’s a robocaller trained to book reservations and hold phone conversations on your behalf.
The catch: the bot sounds all too human.
Duplex: Hi, I’m calling to book a woman’s haircut for a client. I’m looking for something on May 3rd.
Staff: Sure, give me one second.
The neophyte AI assistant plays the part of a human so naturally when it places a call that the person on the other end of the conversation might actually think they were speaking to a human.
On the back end, Duplex is powered by Google’s advancements in natural language processing, deep learning, and text-to-speech technology.
On the B2C front, this super bot demonstrates so-called speech disfluencies – the pauses in conversation marked with uhh’s, umm’s and hmm’s – associated with moments of careful thought.
This has been deliberately baked into the system for the AI to “signal in a natural way that it is still processing,” the engineers behind the project said.
When Google CEO Sundar Pichai unveiled Duplex at the company’s I/O developer conference, complete with a demo of two pre-recorded phone calls made by the bot, the crowd cheered him on.
Online, however, a number of industry observers have expressed concern over the eeriness of hearing an artificial proxy independently making calls and setting appointments.
More alarmingly for some, the way service workers at the receiving end of the call had been duped by Duplex raises ethical questions.
“[Google] seems to have unleashed the AI onto unsuspecting business staff who were just going about their day jobs. Can you see the ethical disconnect?” asked Natasha Lomas, senior reporter for TechCrunch.
Others have echoed a similar sentiment.
“As I walked out of the conference yesterday, I couldn’t stop thinking about the person on the other end of the line,” said Natt Garun, technology editor for The Verge. “When did human service workers become Google’s to experiment on?”
The rise of tech that can be customized to cater to every need no matter how small has enabled a feeling of entitlement and a disregard for the humans fulfilling such demands, Garun said.
Lauren Goode of Wired, along with other observers, pointed out the robocaller had failed to identify itself as non-human at the start of both calls.
“[Google Duplex] straddles a fine line between being enormously convenient and eerily deceptive,” she said.
Another point of contention over Duplex is the fact that the automated calls come from Google’s back end, not from the user’s own phone. The user will not have access to a recording, but just a notification that the task – in the case of the demo, setting an appointment – has been completed.
If the conversation ever gets messy, the call will be turned over to a human, the developers said.
Google’s experiment may have set a precedent for AI developers to keep programming bots that increasingly behave like humans, even in the way they stutter or slow down in their conversations.
Duplex will still be tested publicly in the coming months and, as CNET reported, Google wants to tread with caution. Pichai made clear the team is working to “get the user experience and the expectations right for both businesses and users.”