Google’s Duplex – was it too real?

Google apparently released 100 new products or features at its big annual conference this week, but there’s one that people can’t stop talking about – the digital assistant that will make calls posing as you.

Some say the demonstration of Google Duplex, which formed part of Google CEO’s Sundar Pichai’s keynote address on Tuesday, passed the Turing Test – fooling a person into thinking the caller was another human on the line, not an artificial intelligence bot.

Others were more hung up on the fact that the AI assistant didn’t introduce itself as a digital representation of its human overlord, but just launched on into the conversation.

Pichai played two recordings on stage – one of a lifelike female voice calling a hair salon to book an appointment. It went like this:

Duplex AI: “Hi. I’m calling to book a women’s haircut for a client. Um, I’m looking for something on May 3rd.”

Salon: “Sure, give me one second.”

Duplex AI: “Mhm”

Salon: “Sure, what time are you looking for, around?”

Duplex AI: “At 12pm.

Salon: “We do not have a 12pm available, the closest we have to that is a 1:15”

Duplex AI: Do you have anything between 10pm and uhh 12pm?

Salon: “It depends on what service she would like. What service is she looking for?”

Duplex AI: “Just a women’s haircut for now.”

A ripple of excited laughter spread through the Shoreline Amphitheatre as the conversation unfolded. If this was a real, unedited conversation it showed that Google’s AI and machine learning technology had managed to pull off an incredibly difficult task – conduct a fluid, naturalistic conversation between human and computer.

We’ve got used to talking to the likes of Siri, Google Assistant and Amazon Alexa, but they are still relatively stilted exchanges.

The second recording, of a “man” calling a restaurant to book a table, was even more impressive –  you can listen to both recordings here. The bot was dealing with background noise, a restaurant worker with a strong accent and a conversation that didn’t quite go as expected – he discovered that the restaurant didn’t take bookings for groups less than five people.

Google staffers I spoke to later were quick to play down Duplex as an early-stage “experiment”that will require a huge amount of development to be useful in the real world and may never even make it into a product. It is aimed at those businesses who aren’t set up to handle online bookings via email or forms, so still get a lot of calls just to make appointments or reservations.

From the caller’s point of view its hugely convenient – once the booking is successfully made on the phone, it is automatically added to Google Calendar. That’s a great application of the technology, but the compelling if fleeting demonstration, left too many questions swirling around Duplex.

A few of us journalists over drinks that night considered some of the “what if” scenarios Duplex gives rise to. What if your digital assistant gets hacked and used to misrepresent you on the phone? What if it makes a mistake and requests a booking you didn’t want, leaving you with a hefty bill?

What if political operatives create bot farms and start calling politicians en masse, making it appear as though there is huge public opposition to an issue? What if marketers replace their easy to spot robocalls with all too human voices trying to coax us into buying things?

We had endless fun going through all the scenarios. But here’s the thing – we’ve no idea how far along Google really is with Duplex beyond those narrow demos – as my female friends pointed out to me, few of them actually call their salon asking for a “women’s haircut”. The conversation is likely to be much more detailed, nuanced and unpredictable. Google’s big fear should be that Duplex will turn out to be a major fizzer, which is a distinct possibility.

But the intense response to it suggests it has triggered something in people – maybe a thrill at the prospect of having a digital slave take on our mundane tasks, maybe fear of a disembodied synthetic voice assuming our identity.

We needed to hear whether AI-produced calls will be introduced as such. Will there be an audit trail so we can go back through the transcripts if there is a dispute over what was ordered, reserved or bought? How can Google prevent this from being misused?

Google underestimated the stir this would cause. It is charging ahead as an “AI-first” company, the biggest of its kind in the world, the pace of progress accelerating.

Its executives have spent the week playing down the fears around artificial intelligence. General AI, as opposed to the narrow AI Duplex represents, is a long way off. By and large, I think they are right – issues around biased algorithms perpetuating racism in the criminal justice system or driverless cars mowing down pedestrians can be addressed now.

What’s harder to deal with is our innate unease at the thought of artificial intelligence dehumanizing us, outperforming us, making us irrelevant, and tricking us with human-like qualities. Google has a way to go to properly socialise the revolution it has kickstarted.

Peter Griffin attended Google I/O in Mountain View, California as guest of Google.

This article originally appeared on