Making Sense of the ‘Near-Human’ Google Duplex

CX Today Team

The Google Assistant add-on features a voicebot that makes outbound calls

Google Insights
Making Sense of the ‘Near-Human’ Google Duplex

Google Duplex is an AI-based technology that engages in real-world conversations to accomplish specific tasks, such as scheduling appointments.

This technology makes the conversational experience as realistic as possible, allowing users to communicate naturally, as if chatting with another person, instead of adjusting to a computer.

How Does Google Duplex Work?

Google is working hard to make cellphones smarter with Google Duplex. The feature is a Google Assistant add-on that includes a voicebot that proactively reaches out to contacts.

For example, customers may ask the Assistant to reserve a table at a restaurant. The Assistant then interacts with whoever answers the phone at the restaurant using Duplex.

Duplex’s current function may appear to be only a minor addition – yet it has far-reaching implications. Indeed, when Google Duplex speaks on the phone, its voice sounds remarkably human, complete with pauses and fillers like “uh.”

Also, the Google Duplex system can hold sophisticated conversations and perform most of its tasks without human participation. It is a significant next step in the growth of AI assistants that are entirely self-contained and natural-sounding.

Duplex is available on Google’s Pixel phones and any iPhone running iOS 10 or higher.

3 Critical Google Duplex Features

1. Sounding Natural

Google Duplex combines concatenative text to speech (TTS) software and synthetic TTS software, utilizing Tacotron and WaveNet to offer a natural AI voice.

The system also includes filler words – such as “hmm”s and “uh”s – to sound more genuine, as when humans put together their ideas, this is a typical form of articulation.

Also, the solution matches expectations when it comes to latency. For instance, when people say something simple like “hello?”, they expect a quick response and are more sensitive to delays.

Google Duplex uses faster, low-confidence speech recognition models to determine whether lower latency is required. In extreme cases, it makes quick approximations – usually accompanied by more hesitant responses, as if the individual did not fully understand. Although, it can react with speeds of less than 100 milliseconds.

Surprisingly, adding more delay to a conversation does enable the exchange to feel more authentic in particular circumstances, such as responding to an especially complicated comment.

2. System Training

The Google Duplex system holds task-orientated conversations without the need for human participation. The system also has self-monitoring capabilities, which allow it to identify unsuccessful attempts to carry out tasks. It also sends a message to the user, sharing details of what went wrong.

Fortunately, the system can learn to perform tasks across new domains and industries, using real-time supervised training. Experienced operators perform this coaching as if an instructor watching a student’s work and sharing advice to ensure they accomplish tasks to the optimum standards. These operators can alter the system behaviors in real-time by observing it make phone calls in a new domain. Such a process continues until the system reaches the desired level of quality, at which point monitoring stops, and the system is free to handle the new task.

3. Data Insights

The Google Duplex system constantly learns from millions of interactions. As a result, it picks up valuable insights into how people communicate.

One such insight is that when people talk about something that interests them, they use more diverse language. For example, if someone likes fine dining, they are more likely to talk about restaurants and the food, atmosphere, and services they offer.

Another insight is that people tend to use more hesitations and disfluencies when speaking with AI assistants since they are still getting used to conversing with machines.

Google Duplex Use Cases

Duplex focuses on making restaurant reservations, purchasing movie tickets online, scheduling haircuts, and assisting with lost or forgotten passwords. Yet, Google AI is only limited by time, and the service will likely do more in the future.

The option to verify company hours is another small function, but it is highly relevant during holidays or emergencies when people frequently access Google Search or Maps results.

Benefits of Google Duplex’s Near Human Conversational AI

  • Companies that use Duplex-supported appointment reservations offer another customer engagement route, potentially opening themselves up for more business.
  • Duplex might help minimize no-shows by notifying clients about upcoming appointments in a way that makes canceling and rescheduling simple.
  • Companies can enable employees to focus on activities that drive higher value by having Duplex proactively reach out to customers.
  • Users can benefit from the ability to conduct asynchronous interactions with service providers, such as making reservations in off-hours or when network access is limited.
  • It can assist with accessibility and linguistic hurdles like allowing hearing-impaired or non-native language people to use the phone.

Final Thoughts

The Google Duplex system is a powerful tool for conducting natural conversations with humans. It can handle complex tasks, follow multiple conversation threads, and respond quickly and accurately in real-time.

Such a system has the potential to revolutionize many areas of our lives. Indeed, many people anticipate that these technological advancements will eventually lead to a significant change in communication preferences.

Discover how Google are also innovating in the contact center by reading our article: Google Expands Into CCaaS

 


Join our Weekly Newsletter