评论
打印
This is Scientific American — 60-Second Science. I'm Larry Greenemeier.
(Booking haircut audio)
The person making the haircut appointment for her boss? She's not a person.
There's a trend rapidly developing in AI—have it do the things we don't want to do. Self-driving cars get a lot of attention in this arena. But Google recently demonstrated AI for more mundane tasks. Their system is called Google Duplex—it's a voice assistant designed to mimic the ways humans speak to one another.
(More haircut-appointment audio)
Amazon, Apple and other tech companies also make voice assistants, of course. What's new with Duplex is that it can carry on a conversation seemingly as a human being—that is, somewhat awkwardly. And that's the point of course—Google doesn't want Duplex to sound like a machine. In the hair salon example, the woman who answers the phone throws Duplex a curveball:
(More haircut appointment audio)
There is a point in the exchange where Duplex shows its true colors, though. Listen carefully and you'll hear a distinct difference between the AI's pronunciation of "10 A.M." and "12 P.M."
(Clip is played)
For just an instant, Duplex's mismatched tones remind you of being on hold with your bank. ("Your wait time is...10 minutes")
A second Duplex clip is a bit more impressive, because the person answering the phone doesn't seem to be fully paying attention. Duplex tells the woman three times that he wants to come by on Wednesday.
(Restaurant-reservation audio)
Duplex has a ways to go before it'll be ready to handle open-ended conversations on random topics. But one day you may be freed from all kinds of unpleasant vocal interactions. Leaving more time for unpleasant social media interactions.
Thanks for listening for Scientific American — 60-Second Science. I'm Larry Greenemeier.
