The lines between human and artificial intelligence just grew a little fuzzier. Google Duplex, the company’s new AI assistant – capable of speaking in the dulcet tones of singer John Legend should you so require – booked an appointment for a haircut with an unwitting salon at the Google I/O conference yesterday, even throwing in a few “mmm-hmms” for good measure.
Language is incredibly complex—people ask about something as simple as the weather in over 10,000 ways (“our favourite: ‘Will it be cats and dogs today?’”) Google noted in a blog published to coincide with the event’s keynote by Google CEO Sundar Pichai. But two years after the company unveiled its vision for the Assistant, progress has been blisteringly fast.
In the first demo yesterday at the event in Mountain View, California, a woman calls a hair salon, engaging in chat for almost a minute before agreeing a time that works for a hair appointment. In the second demo, a man calls a restaurant to book a reservation; the woman has a strong accent and isn’t offering clear information so the caller opts to make a new request.
Both callers were AI-powered bots, dispatched through Google Assistant – already available on more than 500 million devices and capable of working with over 5,000 connected home devices and cars from more than 40 brands.
Google says it will be available in more than 30 languages and 80 countries by the end of the year. But its uncanny ability to replicate natural conversation is what stands out – it has been programmed to match expectations around latency – and the speed with which it can learn grows ever faster.
“Up until now, creating a new voice took hundreds of hours in a recording studio. But with advancements in AI and WaveNet technology from DeepMind, we can now create new voices in just a few weeks and are able to capture subtleties like pitch, pace, and all the pauses that convey meaning, so that voices are natural-sounding and unique.”
The assistant is close to recognising when a user is talking to it, rather than conversing with someone else, so follow-up requests won’t need to start with “hey Google”, the company said – one of the top requests of users and can understand complex sentences, fast speech and long remarks. The company sees strong uptake by small businesses.
“Our research shows that 60 percent of small businesses who rely on customer bookings don’t have an online booking system. And many people simply don’t reserve with businesses that don’t take online bookings. With Google Duplex, businesses can operate as they always have; and people can easily book with them through the Google Assistant. We’re just getting started, but we’re excited for how Google Duplex can connect small businesses with Google Assistant users.”
Do say: “Dinner for two please”
Don’t say: “While walking along in desert sand, you suddenly look down and see a tortoise crawling toward you. You reach down and flip it over onto its back. The tortoise lies there, its belly baking in the hot sun, beating its legs, trying to turn itself over, but it cannot do so without your help. You are not helping. Why?”