Google Duplex is well on its way to becoming the most impressive voice assistant out there. This human-sounding virtual secretary debuted at Google I/O in May and made everyone who tuned in ask, “Is this the general-purpose A.I. we’ve been waiting for?”

Sadly or fortunately, depending on where you stand on the A.I. debate, this is not an all-purpose robot, Google executives tell Inverse. During a recent live demo Google’s VP of Engineering Scott Huffman reiterated that Duplex will only be able to carry out three specific tasks.

This press event was largely a push for transparency. The notion of a human-sounding A.I. raises few ethical quandaries: Don’t people deserve to know on some level who they’re speaking to? Google will do its best to quell those fears as it begins testing Duplex over the next few months.

Duplex's initial demo at Google I/O.

Here’s everything we know about Duplex, at least for now.

Google Duplex: It Will Announce Itself

One of the glaring issues brought up by Duplex’s first demo was that it could easily be used to dupe someone into thinking they’re speaking to a person. The company’s fix for this was simply to make Duplex announce itself as A.I.

“Hi! I’m the Google Assistant calling to make a reservation for a client,” a feminine voice told me as I pretended to be a restaurant receptionist. “This automated call will be recorded. Can I book a table for tonight at 7 p.m.?”

Google records these calls to further train Duplex and provide a measure of quality assurance, like a customer service hotline. Currently, the only way to refuse to be recorded is to hang up. This could potentially run into roadblocks with laws requiring two-party consent to record calls, active in twelve states.

Google Duplex: Three Specific Use-Cases

Duplex was designed to sound natural, but it was created to only perform three tasks: Check for store hours during holidays, make restaurant reservations, and book salon appointments.

google duplex assistant end to end graph
A user asks the Google Assistant for an appointment, which the Assistant then schedules by having Duplex call the business.

“This is not some kind of general A.I.,” says Huffman. “We made this with specific use-cases in mind. So if you ask it how the weather is like it won’t give you an answer.”

Google said it would begin limited testing of this first task with a select number of restaurants and “trusted testers” in July. Trials for its other two capabilities will begin in the following months, which will likely be followed by regular updates. A firm release date has yet to be set.

“While we’re not widely launching this feature yet, we’re sharing more information about this technology to provide transparency and encourage feedback,” stated a company press release. “It’s important that we get the experience right both for people and for businesses, and we’re taking a slow and measured approach as we incorporate learnings and feedback from our tests.”

Google Duplex: Where This Stands in the Market

As far as the competition goes, Duplex is already miles ahead of Siri and Alexa in terms of being able to mimic natural speech and carry out commands on your behalf. Both Apple and Amazon’s assistants also operate on a command system, meaning you have to beacon them with a “Hey Siri” or “Hey Alexa” to activate them.

google duplex assistant graph flow chart
Incoming sound is processed through an ASR system. This produces text that is analyzed with context data and other inputs to produce a response text that is read aloud through the TTS system.

Duplex is a stunning example of how Google has made their assistant capable of carrying out full conversations, without the need for an activation phrase. This is further accentuated by its ability to use “speech disfluencies,” or the “ahhs” and “umms” we splice in between what we say every day.

Photos via Google (1, 2)