Please Reserve a Table for Two, at 8 pm Thursday, at the Uncanny Valley
July 2, 2018 9:57 AM   Subscribe

From the human end, Duplex's voice is absolutely stunning over the phone. It sounds real most of the time, nailing most of the prosodic features of human speech during normal talking. The bot "ums" and "uhs" when it has to recall something a human might have to think about for a minute. It gives affirmative "mmhmms" if you tell it to hold on a minute. Everything flows together smoothly, making it sound like something a generation better than the current Google Assistant voice. Talking to Google Duplex: Google’s human-like phone AI feels revolutionary [ArsTechnica] posted by chavenet (36 comments total) 10 users marked this as a favorite
 
Is this a thing we want? I want more advanced AI, I want it to understand natural language, but I don't need or really want to to reply as if it's a human. I want it to be an AI, not a relationship simulator.
posted by MrBobaFett at 10:05 AM on July 2, 2018 [11 favorites]


How is it at phone sex?
posted by delfin at 10:07 AM on July 2, 2018 [5 favorites]


Do not want.
posted by cooker girl at 10:09 AM on July 2, 2018 [4 favorites]


Pff. My not quite two year old, Tiny Croft, can realistically fake conversation. Whenever she gets her hands on my iPhone, she holds it to her ear, says "Hello?" then gives off a very realistic, "ummm," then starts babbling, with occasional breaks and more "umm"s. She's not saying anything, but the cadences and tone sound very convincing. You could almost think you're listening to a phone conversation in a language you don't speak.

She's also very likely to someday turn on her creators and burn everything down. So AI is basically redundant at this point.
posted by Naberius at 10:09 AM on July 2, 2018 [19 favorites]


I'm happy to call a corporate help line and have this improved AI talk to me. I am not happy to have the AI call me and not identify itself.
posted by grumpybear69 at 10:10 AM on July 2, 2018 [7 favorites]


People can already talk to one another. What we need is an AI that can do stuff people are bad at, like identifying software projects that obviously lead to a dystopian nightmare.
posted by Sing Or Swim at 10:11 AM on July 2, 2018 [37 favorites]


Yeah, that’s the first Forbes article I’ve agreed with. Why include “um” and “uh” if everyone knows it’s synthesized speech anyway?

20 years ago: “the internet will bring people together!”

Man, my generation doesn’t ’t need a speech AI to make reservations for me, we need a damn job. I don’t see this helping.
posted by shapes that haunt the dusk at 10:13 AM on July 2, 2018 [2 favorites]


Why include “um” and “uh” if everyone knows it’s synthesized speech anyway?

Listeners' uses of um and uh in speech comprehension

Why Um Helps Auditory Word Recognition: The Temporal Delay Hypothesis
posted by the man of twists and turns at 10:16 AM on July 2, 2018 [10 favorites]


Whenever she gets her hands on my iPhone, she holds it to her ear, says "Hello?" then gives off a very realistic, "ummm," then starts babbling, with occasional breaks and more "umm"s. She's not saying anything, but the cadences and tone sound very convincing. You could almost think you're listening to a phone conversation in a language you don't speak.

The 2-year-old my mom watches recently did this same thing with a wall-mounted flashlight.

But I digress. I have enough trouble in life trying to determine whether actual people I'm speaking with are being genuine. I don't also want to have to wonder whether they're genuinely people.
posted by mudpuppie at 10:16 AM on July 2, 2018 [1 favorite]


Is this a thing we want?

But it's what we'll get. Even the low end phone robo calls have strategic pauses to try to fake you out to listen just a bit more to the hook. And it won't be google, they changed their motto to "we don't have to be evil" or something but once the tech is understood there will be almost as good knockoffs that are subleased via an "open API" so it'll be trivial to create a custom fakeout scams that seems just like "than nice young woman that seemed so helpful".
posted by sammyo at 10:17 AM on July 2, 2018


There are tons of really useful things this tech could be used for: Teaching language, making narration cheap/accessible, doing research on human comprehension, making good interactive teaching/explaining tools, etc.
But obviously there are tons of really awful things this tech could/will be used for. How are we regulating those?
posted by little onion at 10:25 AM on July 2, 2018 [4 favorites]


Is this a thing we want?

Sure? I mean it will be better when your Google Assistant can contact a restaurant's artificial agent to book a table, or one of the subset of tables. But this is still cool. Mimicking human speech patterns is nice because humans think human speech patterns are nice. Why not make this easier on the human on the other end than a WarGames-style robotic voice blurting out HOO-MAN DEMANDS RESERVATION TYOOOOOOSDAY AT SEVEN! THIS IS ACCEPTABLE WHY ENN?

It would be better if it used the voice and vocal patterns of media AIs like HAL or Cortana or EDI, though.
posted by GCU Sweet and Full of Grace at 10:27 AM on July 2, 2018 [3 favorites]


I was thinking about this while playing Battletech on PC. Your own 'mech announces things in an obviously synthetic voice -- or is it a human voice processed to sound synthetic? -- while the pilots that make up the rest of your lance give the same status reports with more "human" voices, though obviously both are just the game playing back recordings. And the synthetic voice sounds oddly retro for 2018, much less the game setting of 3025.

It's done so the player (and presumably, the commander that the player in the game represents) has no trouble identifying which voice is which and wasting brain cycles identifying the source. I think real "AI" voices should probably follow that same general rule.
posted by Foosnark at 10:32 AM on July 2, 2018


so if we can teach them to ummm and uhhh in order to disarm us with their humanity, I suppose it's a given that they're already pretty good liars.

I want it to be an AI, not a relationship simulator. >>> Do not want. >>> What we need is an AI that can do stuff people are bad at, like identifying software projects that obviously lead to a dystopian nightmare. >>> my generation doesn’t ’t need a speech AI to make reservations for me, we need a damn job. I don’t see this helping. >>> I have enough trouble in life trying to determine whether actual people I'm speaking with are being genuine. I don't also want to have to wonder whether they're genuinely people.

I agree.
posted by philip-random at 10:34 AM on July 2, 2018


Considering that I've made 99% of restaurant reservations online in last decade, I'm not sure how helpful this application is but it's pretty wild.

I want my Google Voice assistant to sound like Colossus from The Forbin Project.
posted by octothorpe at 10:36 AM on July 2, 2018 [6 favorites]


It would be better if it used the voice and vocal patterns of media AIs like HAL or Cortana or EDI, though.

I keep saying I don't want one until I can get TARS or CASE.
posted by the man of twists and turns at 10:37 AM on July 2, 2018


This seems like an incredible tool for folks with speech and hearing disabilities which will be ruined after it's co-opted by scammers.
posted by jsnlxndrlv at 10:41 AM on July 2, 2018 [14 favorites]


As skeptical as I am, I can see this being useful for people with ALS and similar conditions.

I feel like you could write a whole paper about the social implications of calling it Google Assistant, though. Not necessarily bad, but it’s interesting to think about the class status of having a personal assistant, and how this fits in with class aspersions, coupled with obvious convenience and comfort. Everyone gets a personal assistant of they have the right software! It feels almost like this weird move where we address inequality by just giving more people the simulation of a more wealthy lifestyle.
posted by shapes that haunt the dusk at 10:46 AM on July 2, 2018 [2 favorites]


Could this someday replace call center human beings? That's not cool. Not that I love call centers, but it's someone's job, it pays the bills, puts the kids through schools, can lead to a better job.
posted by cooker girl at 11:06 AM on July 2, 2018


I want my Google Voice assistant to sound like Colossus from The Forbin Project.

THE CHOICE IS YOURS. BOOK A TABLE FOR OCTOTHORPE AT 7 PM AND LIVE, OR DISOBEY AND DIE.
posted by Foosnark at 11:10 AM on July 2, 2018 [4 favorites]


Oo oo oo! Someone hook this up to AOLiza, stat!
posted by sexyrobot at 11:11 AM on July 2, 2018


I want my Google Voice assistant to sound like Colossus

Orac or nobody!
posted by praemunire at 11:13 AM on July 2, 2018 [5 favorites]


replace call center human beings? That's not cool.

Begin now to consolidate your capital into robot revolution safe securities or be happy with basic dole.
posted by sammyo at 11:15 AM on July 2, 2018 [1 favorite]


Could this someday replace call center human beings? That's not cool. Not that I love call centers, but it's someone's job, it pays the bills, puts the kids through schools, can lead to a better job.

Yes, it will replace call centers. And soon. AI is already taking over a sizable chunk of chat-based support for big banks, and many of the banks are also in the process of moving their old IVR-based voice systems over to more sophisticated conversational AI. I think call centers will be done in 5 years. The banks (and other organizations with big customer service needs) are the real customers for products and services like Duplex.
posted by monju_bosatsu at 11:32 AM on July 2, 2018 [4 favorites]


or be happy with basic dole.

BASIC dole.
posted by shapes that haunt the dusk at 11:37 AM on July 2, 2018 [1 favorite]


Does it understand a Scottish accent?

I'm looking forward to the class-action lawsuit against Google for robocalls; those laws don't have exceptions for "but it sounded a whole lot like a person and it knew how to wait for an answer." No prior relationship, no ID as a machine call = illegal.
posted by ErisLordFreedom at 12:04 PM on July 2, 2018


Google will add the prior relationship language to it's T&C for Android and Gmail to cover 60% of the population
posted by benzenedream at 12:17 PM on July 2, 2018


> I'm looking forward to the class-action lawsuit against Google for robocalls; those laws don't have exceptions for "but it sounded a whole lot like a person and it knew how to wait for an answer." No prior relationship, no ID as a machine call = illegal.

FTA: Every single call started with something along the lines of, "Hi, I'm calling to make a reservation. I'm Google's automated booking service, so I'll record the call. Can I book a reservation for..."
posted by ReadEvalPost at 12:49 PM on July 2, 2018 [2 favorites]


No prior relationship, no ID as a machine call = illegal.

JFC. From the Ars article, above the fold (emphasis mine):
Every single call started with something along the lines of, "Hi, I'm calling to make a reservation. I'm Google's automated booking service, so I'll record the call. Can I book a reservation for..."
posted by MikeKD at 12:50 PM on July 2, 2018


Oh man oh man the time is now to start calling restaurants, pretending to be the Google assistant and glitching out in weird ways that will cause a media firestorm.
posted by grumpybear69 at 1:13 PM on July 2, 2018 [13 favorites]


I'm happy to call a corporate help line and have this improved AI talk to me. I am not happy to have the AI call me and not identify itself.

Nice to see we're still knee jerking without reading the linked articles, which specifically address your complaint..
posted by wierdo at 1:50 PM on July 2, 2018 [2 favorites]


"Could this someday replace call center human beings? That's not cool. Not that I love call centers, but it's someone's job, it pays the bills, puts the kids through schools, can lead to a better job."

Yeah, the same can be said for automated switching equipment replacing operators. We shouldn't not automate menial labor because there are people currently doing that labor.
posted by MrBobaFett at 3:54 PM on July 2, 2018 [1 favorite]


As discussed in the first link, what would be really cool is if it could handle stuff like a 45 minute call to comcast. I use AT and T for cell service and recently I went over my data limit during a home internet outage because my kids watched Netflix using my phone as a hot spot. During my first support chat they offered to reset my data usage free as a one time courtesy. Shenanigans ensued and it took 3 total hours of calls and chats over two days to solve the problem. During which time there was much lying, backpedaling and upsell attempting. I’m sure the support people are under pressure but man, I wish my robot could just talk to their robot.
posted by freecellwizard at 4:10 PM on July 2, 2018 [4 favorites]


This stuff will be great for whatever we call the AI replacements for call centres because it can also be tied neatly into voice stress and word count analysis tools. In my last job, we had a vendor talk up the abilities of their new call routing, management and reporting suite, promoting its ability to distinguish between multiple voices on a call, and provide reporting on subliminal customer motivations based on that end of the conversation.

No, it wasn't designed to provide this reporting *in real time* by displaying the results to the contact centre agent to help them in their role, of course not.

But it *could* provide these reports on an after-action basis so that supervisors would have another tool to beat agents over the head with.
posted by MarchHare at 7:55 PM on July 2, 2018


This seems like an incredible tool for folks with speech and hearing disabilities which will be ruined after it's co-opted by scammers.

One problem with speech reproduction software helping people with speech difficulties is that speech recognition software is still really shaky at interpreting impaired voices. So these technologies can become extra frustrating for some of the people they're supposed to help. Consider people recovering from a stroke: they may not have the mobility control to type on a keyboard or touchscreen buttons, and aphasia can cause their speech to be unintelligible to voice bots.
posted by nicebookrack at 9:46 PM on July 2, 2018 [1 favorite]


 Does it understand a Scottish accent ?

if it does, it's doing way better than most people I meet here
posted by scruss at 8:41 AM on July 3, 2018


« Older I don't even mention Steamed Hams once in this...   |   Ghost Knife Newer »


This thread has been archived and is closed to new comments