Google Duplex – AI Machines That Mimic Humans to Book Appointments

Credit: AI Google Blog

"A long-standing goal of human-computer interaction has been to enable people to have a natural conversation with computers, as they would with each other."

- Yaniv Leviathan, Principal Engineer at Google

Google's Duplex is an AI machine that can interact in real time conversations to take care of things like ordering from a restaurant or booking a hair appointment:

Engineers at Google describe the use of neural networks and speech recognition to achieve the flow of conversation between machines and humans:

"At the core of Duplex is a recurrent neural network (RNN)...built using TensorFlow Extended (TFX). To obtain its high precision, we trained Duplex’s RNN on a corpus of anonymized phone conversation data."

For those who haven't run across the term before, a neural network is a network or circuit of neurons. It's an artificial neural network, composed of artificial neurons or nodes.

A neural network can be a biological neural network, made up of real biological neurons, or an artificial neural network, for solving artificial intelligence (AI) problems.

These artificial networks are used in predictive modeling, adaptive control and applications where they are trained through a dataset.

Self-learning resulting from an experience or series of events can occur within networks, which can understand and make conclusions from a complex and seemingly unrelated set of information.

Google's engineers are combining several pieces of tech to create a learning model for AI, to facilitate real time machine and man convos:

"The network uses the output of Google’s automatic speech recognition (ASR) technology, as well as features from the audio, the history of the conversation, the parameters of the conversation (e.g. the desired service for an appointment, or the current time of day) and more."

The idea is to be able to automation many of the functions that are rote or routine for us, with machines.

"We trained our understanding model separately for each task, but leveraged the shared corpus across tasks. Finally, we used hyperparameter optimization from TFX to further improve the model."

This is all combined in the Google Assistant, piece of software that is your robot helper which has the ability to multitask if you wanted to ask what the weather was like in New York City and Austin.  

Here's a model of what a day would look like using the Google Assistant Snapshot powered by Google Duplex, to schedule picking up your groceries from curbside delivery, or notify you when your package has arrived:

Google Duplex was first launched for owners of its Google Pixel, Pixel XL, Pixel 2, Pixel 2 XL, and Pixel 3 and Google Pixel 3 XL phones. More recently, Google Duplex support has started to roll out to any Android devices running version 5.0 or newer. It’s also starting to roll out to Apple’s iPhones that have Google Assistant installed.

Perhaps the Duplex's biggest limitation at this beta stage is that you can't hear the recording of the Google Duplex calls that’s made on your behalf afterward, or get a written transcript. There’s no word yet if Google will allow Duplex users to have that info going forward.

Another draw back is businesses that screen their calls due to the insane amount of spam calls by robo-dialers that are out of control - many businesses won't pick up a call from Duplex / Google Assistant because the caller ID says "Google" instead of a person's name.

Many servers at restaurants that have received calls from Google Duplex describe it as "Creepy but polite."

A lot of Duplex’s work isn’t actually call-based. When you make a request with Google Assistant for a restaurant booking, it searches for vacancies via third-party booking platforms such as OpenTable, Resy, or Yelp.

If Google Duplex can,  it will confirm the time with you and book it, or Duplex will say that time is unavailable and it can't complete the task. The Duplex calling feature only happens for restaurants that aren’t on those sites and who opted into getting calls from Google Assistant on their Google business page.

Gabriel Murphy told The Verge he'd tried the AI out on his staff, monitoring the call in his office. They didn't believe they were talking to a machine.

“None of the staff seemed to have any issue with it, [but] there were plenty of jokes about Skynet and machines taking over,” Murphy says.

Skynet is a fictional artificial neural network-based conscious group mind and artificial general superintelligence system that is the bad guy in the Terminator movie series.

When Skynet gained self-awareness, humans tried to deactivate it, prompting it to retaliate with a nuclear attack, an event known as Judgment Day.

Are we getting closer to that happening in reality? I hope not.

Source: NY Times

In the New York Times article that the gif above comes from, reporters found out that 25 % of the supposed AI-driven calls was actually placed by humans at a Google call center.

The reason why is that Google Duplex just doesn't have enough data yet to operate at 100% machine level. The nature of AI-based-learning in neural networks is you need a lot of data for the machines to parse together accurate responses.

Whether or not we are headed for the Rise of the Machines, as a real threat to human survival, my concerns lie in the cognitive development and social implications that follow when any tech is used to replace human interactions.

Most of the articles we read about new technology that will make our lives easier and automate more of our tasks leave out the affects on social development and the risks of cognitive deficiencies that come with the territory when humans seldom use or stop using a certain function of our brains like empathy and critical thinking that help balance decision making skills.

What does a future look like when everyone is so used to using AI assistants to take care of daily tasks and there's a glitch, or a computer crash, and now everyone is helpless to do basic routine task-management because we're so used to outsourcing daily scheduling to a machine?

Will this take away jobs from personal assistants? Or on the other hand, will this help reduce stress for service workers because they won't get yelled at on the phone if a reservation had to be re-scheduled or canceled?

As always, the bigger picture of how we develop successful interdependent relationships with each other, our friends and family, and the world remains left out of the narrative of bright new shiny technology but still worth pondering as we move forward into a bigger future.

What are your thoughts on machines taking over more of the tasks humans do? Would you use Google Duplex?

               .  .  .