Google Duplex Blows Automated Assistants Out Of The Water
Artificial Intelligence on the line
Like every other tech leader, Google has invested heavily in Artificial Intelligence. This year’s Google I/O conference, held in Mountain View, California, showcased the accumulation of years of avid AI research. One of the major goals of AI development has been to create systems that can reliably understand and process natural language. For the most part, many of the services available today remain slow and clunky. While automated assistants have certainly reduced the burden on phone operators, their inability to fully handle the nuances of language can frustrate and alienate customers. Thanks to Google Duplex, the days of painfully mechanical AI assistants could be over.
Please leave your artificially intelligent message after the tone…
Revealed in May 2016, Google Assistant is a personal digital helper for everyday life. It can search the internet, schedule events, set alarms, show Google account information and more. However, it can’t make those all important phone calls that make a real PA so valuable. Well, it couldn’t… Until now. Enter Google Duplex, a new feature of Assistant that takes natural language processing (NLP) to the next level. At this year’s I/O conference in May, recordings of two real life Duplex calls were played. One was to a hair salon, in which Duplex successfully booked a haircut for ‘Lisa’. The system even used verbal pauses such as ‘um’ and phonetic responses like ‘mm hmm’. Duplex also made a potentially difficult restaurant booking, handling non standard English and a series of misunderstandings. Computers have gradually been getting better at NLP thanks to deep neural networks. Duplex itself was trained using a recurrent neural network (RNN) and anonymous phone calls.
That isn’t to say that there aren’t still some, er, hang ups. When Duplex can’t understand a conversation, it passes the call to a human operator. Natural language is incredibly complicated. Machines are tasked with comprehending intonation, accents, pauses and colloquialisms, and then they have to use them, too. Far more training is needed before these systems, no matter how advanced they sound, are ready for commercial release. Developers also have to reduce latency – the time taken to process what has been said and deliver an appropriate reply. As of yet, Duplex is only trained to work within specific domains and subject matter. This may sound like a limitation, but confining the technology to a narrow range ensures that expectations are fulfilled before hype takes hold.
Easier said than done?
Automated assistants have a number of challenges to overcome before they are suitable for everyday use. Eventually, people will talk to technology as easily as they type questions into search engines. Interestingly, there’s also another aspect to AI’s ability speak convincingly. Take, for example, a smart boiler. If the boiler is due a service, it could notify an automated assistant connected to the appliances within the home. The assistant could then call a local, well reviewed plumber, and notify the home owner when an appropriate slot has been found. In a nutshell, that’s what chatty AI is about – reducing friction in everyday life. Not only does it save precious time and effort on the part of consumers, but also for businesses as it works to complement existing processes. There’s no need to retrain staff or set up new phone lines. There are obvious benefits to seamless NLP systems, but an artificial system that sounds exactly like a human could easily be viewed as misleading. As such, it’s important for businesses to be very clear about when the technology is used. According to Google’s blog, the company is still working out how best to do this.
It’s hard not to be impressed by Google Duplex. The service is great news for busy individuals and businesses, automating yet another laborious task. It can be used by anybody as part of Google Assistant, contributing to the democratisation of AI. As well as building a high quality and relevant tool for wider AI application, advances in NLP have blurred the lines between artificial and real. To avoid all manner of potential problems, users need to know that they are interacting with a computer system. Chatting to an AI quickly goes from convenient to creepy if you think you’re on the phone to another person. Transparency issues aside, Google has shown that artificial natural language can be achieved to the extent that it’s indistinguishable from a human voice. What have you got to say about that, Siri?
Should we be speaking to technology in the same way that we speak to each other? What does this mean for our future relationship with technology? Is Google winning the AI arms race? Share your thoughts and opinions.
To find out more about natural language processing, sign up for our free newsletter.