This topic that is going to be discussed is clearly one of the most trending news of recent months. Google Duplex is a feature of the already very famous Google Assistant from Google. Duplex was shown at the Google’s Annual I/O Developer Conference this year. The demonstration of the working and capabilities of the Google Duplex are indeed mind-blowing.
If this is all new to you, let’s start from the beginning. Google Assistant is a virtual assistant that was developed by Google and released for public in 2016. One might have heard of Siri by Apple, Cortana by Windows and Alexa by Amazon, these are all virtual assistants developed with the aim to make the life of consumers easy. Google assistant is mainly available on mobile and Smart Home devices running on Android operating system.
Coming on to the topic, Google Duplex is an AI (Artificial Intelligence) enabled feature that can be used for making reservations at a restaurant or an appointment at a hair salon and maybe call to check opening hours of a business. The quality that differentiates it is the real human-like conversation. Duplex perfectly places the fillers in language as a normal person would while thinking such as ‘umm’ and ‘hmm’. This completely changes the way a human interacts with a computer and this is why it is considered being a huge deal. This technology can easily increase productivity by huge factors. It can help people make calls in regions where they are not familiar with the local tongue.
According to the Google AI Blog, “Google Duplex’s conversations sound natural thanks to advances in understanding, interacting, timing, and speaking. At the core of Duplex is a recurrent neural network (RNN) designed to cope with these challenges, built using TensorFlow Extended (TFX). To obtain its high precision, we trained Duplex’s RNN on a corpus of anonymized phone conversation data. 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. 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.”
With a wide array of possibilities and potential for Google Duplex, comes the most common problem, which is, talking to a robot which is pretending to be a human. Some people may find that creepy, scary and disturbing. One more issue that arises is whether calls should be recorded via Duplex or not. At the moment Google is testing it out in a supervised environment and may be initially available at a small scale as a beta test. We cannot expect huge things from it in the beginning but over the time it will eventually get better and be able to perform complex tasks.