When talking about the capabilities of AI-driven digital assistants, the most talked about issue is their inability to converse in a way a real human does. The robotic tone of the virtual assistants has been limiting them from imitating real humans for a long time. And it’s not just the flat monotone. It’s about understanding the nuances of the language, pitches, intonations, sarcasm, and a lot more.
Now, what if there emerges a technology that is capable of sounding and behaving almost human? Well, look no further, Google Duplex is here to dominate the world of digital assistants.
Google introduced the new Duplex at Google I/O 2018, their annual developer conference, last week. But, what exactly is it?
Google Duplex is a newly added feature to the famed Google assistant. Adding to the capabilities of Google assistant, it is also able to make phone calls for the users, and imitate human natural conversation almost perfectly to get the day-to-day tasks ( such as booking table reservations, hair salon appointments, etc. ) done in an easy manner. It includes pause-fillers and phrases such as “um”, “uh-huh “, and “erm” to make the conversation sound as natural as possible. Don’t believe me? Check out the audio yourself!
Google Duplex booking appointments at a hair salon
Google Duplex making table reservations at a restaurant
The demo call recording video of the assistant and the business employee, presented by Sundar Pichai, Google’s CEO, during the opening keynote, befuddled the entire world about who’s the assistant and who’s the human, making it go noticeably viral.
A lot of questions are buzzing around whether Google Duplex just passed the Turing Test. The Turing Test assesses a machine’s ability to present intelligence closer or equivalent to that of a human being. Did the new human sounding robot assistant pass the Turing test yet? No, but it’s certainly the voice AI that has come closest to passing it.
Now how does Google Duplex work?
It’s quite simple. Google Duplex finds out the information ( you need ) that isn’t out there on the internet by making a direct phone call. For instance, a restaurant has shifted location and the new address is nowhere to be found online. Google Duplex will call the restaurant and check on their new address for you. The system comes with a self-monitoring capability, helping it recognize complex tasks that it cannot accomplish on its own. Such cases are signaled to a human operator, who then takes care of the task.
To get a bit technical, Google Duplex makes use of Recurrent Neural Networks ( RNNs ) which are created using TensorFlow extended ( TFX ), a machine learning platform. Duplex’s RNNs are trained using the data anonymization technique on phone conversation data. Data anonymization helps with protecting the identity of a company or an individual by removing the data sets related to them. The output of Google’s Automatic speech recognition technology, conversation history and different parameters of the conversation are used by the network. The model also makes use of hyperparameter optimization from TFX which further enhances the model.
But, how does it sound natural?
Google uses concatenative text to speech ( TTS ) along with synthesis TTS engine ( using Tacotron and WaveNet ) to control the intonation depending on different circumstances.
Concatenative TTS is a technique that converts normal text into speech by concatenating or linking together the recorded speech pieces. Synthesis TTS engine helps developers modify the speech rate, volume, and pitch of the synthesized output.
Including speech disfluencies ( “hmm”s, “erm”s, and “uh”s ) makes the Duplex sound more human. These speech disfluencies are added when very different sound units are combined in the concatenative TTS or adding synthetic waits. This allows the system to signal in a natural way that it is still processing ( equivalent to what humans do when trying to sort out their thoughts ).
Also, the delay or latency should match people’s expectations. Duplex is capable of figuring out when to give slow or fast responses using low-confidence models or faster approximations. Google also found out that including more latency helps with making the conversation sound more natural.
Some potential applications of Google Duplex for businesses
Now that we’ve covered the what and how of this new technology, let’s look at five potential applications of Google Duplex in the immediate future.
Customer Service
Basic forms of AI using natural language processing ( NLP ), such as chatbots and the existing voice assistants such as Siri and Alexa are already in use within the customer care industry. Google Duplex paves the way for an even more interactive form of engaging customers and gaining information, given its spectacular human sounding capability.
According to Gartner, “By 2018, 30% of our interactions with technology will be through “conversations” with smart machines”.
With Google Duplex, being the latest smart machine introduced to the world, the basic operations of the customer service industry will become easier, more manageable and efficient.
From providing quick solutions to the initial customer support problems and delivering internal services to the employees, Google Duplex perfectly fills the bill. And it will only get better with further advances in NLP.
So far chatbots and digital assistants have been miserable at handling irate customers. I can imagine Google Duplex in John Legend’s smooth voice calming down an angry customer or even making successful sales pitches to potential leads with all its charm and suave! Of course, Duplex must undergo the right customer management training with a massive amount of quality data on what good and bad handling look like before it is ready for such a challenge.
Other areas of customer service where Google Duplex can play a major role is in IT support. Instead of connecting with the human operator, the user will first get connected to Google Duplex. Thus, making the entire experience friendly and personalized from the user perspective and saving major costs for organizations.
HR Department
Google Duplex can also extend a helping hand in the HR department. The preliminary rounds of talent acquisition where hiring executives make phone calls to their respective candidates could be handled by Google Duplex provided it gets the right training.
Making note of the basic qualifications, candidate details, and scheduling interviews are all the functions that Google Duplex should be able to do effectively. The Google Assistant can collect the information and then further rounds can be conducted by the human HR personnel. This could greatly cut down on the time expended by HR executives on the first few rounds of shortlisting. This means they are free to focus their time on other strategically important areas of hiring.
Personal assistants and productivity
As presented at Google I/O 2018, Google Duplex is capable of booking appointments at hair salons, booking table reservations and finding out holiday hours over the phone. It is not a stretch to therefore assume that it can also order takeaway food over a phone call, check with the delivery man regarding the order, cancel appointments, make business inquiries, etc.
Apart from that, it’s a great aid for people with hearing loss issues as well as people who do not speak the local language by allowing them to carry out tasks on phone.
Healthcare Industry
There is already enough talk surrounding the use of Alexa, Siri, and other voice assistants in healthcare. Google Duplex is another new addition to the family.
With its natural way of conversing, Duplex can:
- Let patients know their wait time for emergency rooms.
- Check with the hospital regarding their health appointments.
- Order the necessary equipment for hospital use.
Another allied area is elder care. Google Duplex could help reduce ailments related to loneliness by engaging with the users at a more human level. It could also assist with preventive care and in the management of lifestyle diseases such as diabetes by ensuring patients continue their med intake, keep their appointments, provide emergency first aid help, call 911 etc.
Real Estate Industry
Duplex enabled Google Assistants will help make realtors’ task easy. Duplex can help call potential sellers and buyers, thereby, making it easy for realtors to select the respective customers.
The conversation between Google Duplex ( helping a realtor ) and a customer wanting to buy a house can look something like this:
Google Duplex: Hi! I heard you are house hunting. Are you looking to buy or sell a property?
Customer: Hey, I’m looking to buy a home in the Washington area.
Google Duplex: That’s great! What part of Washington are you looking in for?
Customer: I’m looking for a house in Seattle. 3 bedrooms and 3 baths would be fine.
Google Duplex: Sure, umm, may I know your budget?
Customer: Somewhere between $749,000 to $850,000, is that fine?
Google Duplex: Ahh okay sure, I’ve made a note and I’ll call you once I find the right matches.
Customer: Yeah, sure.
Google Duplex: okay, thanks.
Customer: Thanks, Bye!
Google Duplex then makes a note of the details on the realtor’s phone, thereby, narrowing down the efforts made by realtors on cold calling the potential sellers to a great extent. At the same time, the broker will also receive an email with the consumer’s details and contact information for a follow-up.
Every rose has its thorns. What’s Duplex’s thorny issue?
With all the good hype surrounding Google Duplex, there have been some controversies regarding the ethicality of Google Duplex. Some people have questions and mixed reactions about Google Duplex fooling people of one’s identity as the voice of the Duplex differs significantly from that of a robot. A lot of talk surrounding this issue is trending on several twitter threads.
It has hushed away these questions by saying how ‘transparency in technology’ is important and they are ‘designing this feature with disclosure built-in’ which will help in identifying the system. Google also mentioned how any feedback that people have regarding their new product.
Google successfully managed to awe people across the globe with their new and innovative Google Duplex. But there is a still a long way to go even though Google has already taken a step ahead in an effort to better the human relationships with the machines. If you enjoyed reading this article and want to know more, check out the official Google Duplex blog post.
Read Next
Google’s Android Things, developer preview 8: First look
Google News’ AI revolution strikes balance between personalization and the bigger picture
Android P new features: artificial intelligence, digital wellbeing, and simplicity
