Google Assistant iOS app update coming this year, teases Duplex AI phone calls

2»

Comments

  • Reply 21 of 30
    avon b7avon b7 Posts: 4,199member
    eightzero said:
    I'm now seeing some of the demos. Pretty interesting. Given these are in fact just demos, it is worth considering the potential. For perhaps the first time, I'm actually excited and interested in a google product. 

    A few things pop to mind. As a basic premise, there needs to be acknowledgement that this is indeed a robotic system under the control of, and for interaction with humans. Like self driving cars, humans will intentionally screw with them. Some humans, being called by a robot, will either simply hang up (I do that when I reach someone else's robot) or will be tempted to screw with them. Give a fake name, don't book the appointment, then blame the callers robot. Think "we don't serve their kind" in the cantina scenein Star Wars ("IV"). Pretty sure the consumer will need a receipt directly from the other party, not their robotic confirmation. And oh yes, my robot will first inform the other party they are on a recorded line for "quality assurance purposes." 

    These systems will also need to navigate a call to someone else's robot, aka the phone-hell system systems of press 1 to hear how technology doesn't improve anyone's lives etc. Oddly enough, robot to robot stuff should be the easiest part. Take humans out of the loop, and this get dramatically more predictable. Betcha google knows this in spades.

    Many call centers need info to get started. Will your robot be authorized to speak your SSN or other ID (credit card number?) into a call? Because..well...you know...

    This is the first time I've seen something on this AI stuff that gets my intention. Kinda interesting, and I am now (edit) curious what Apple will say about it at WWDC.
    Interesting points with interesting potential safeguards that you have presented.

    No doubt if humans try to deliberately screw with the systems, someone will try AI Bots to screw with the system too. Bot against Bot as it were.

    I suppose there will be something like a receipt (as you suggest) or other confirmation system to keep things on track.
  • Reply 22 of 30
    SpamSandwichSpamSandwich Posts: 31,404member
    eightzero said:
    I'm now seeing some of the demos. Pretty interesting. Given these are in fact just demos, it is worth considering the potential. For perhaps the first time, I'm actually excited and interested in a google product. 

    A few things pop to mind. As a basic premise, there needs to be acknowledgement that this is indeed a robotic system under the control of, and for interaction with humans. Like self driving cars, humans will intentionally screw with them. Some humans, being called by a robot, will either simply hang up (I do that when I reach someone else's robot) or will be tempted to screw with them. Give a fake name, don't book the appointment, then blame the callers robot. Think "we don't serve their kind" in the cantina scenein Star Wars ("IV"). Pretty sure the consumer will need a receipt directly from the other party, not their robotic confirmation. And oh yes, my robot will first inform the other party they are on a recorded line for "quality assurance purposes." 

    These systems will also need to navigate a call to someone else's robot, aka the phone-hell system systems of press 1 to hear how technology doesn't improve anyone's lives etc. Oddly enough, robot to robot stuff should be the easiest part. Take humans out of the loop, and this get dramatically more predictable. Betcha google knows this in spades.

    Many call centers need info to get started. Will your robot be authorized to speak your SSN or other ID (credit card number?) into a call? Because..well...you know...

    This is the first time I've seen something on this AI stuff that gets my intention. Kinda interesting, and I am now (edit) curious what Apple will say about it at WWDC.
    “OK, I’m just going to need a credit card number before I can book your appointment...”
    watto_cobra
  • Reply 23 of 30
    gatorguygatorguy Posts: 21,107member
    eightzero said:
    I'm now seeing some of the demos. Pretty interesting. Given these are in fact just demos, it is worth considering the potential. For perhaps the first time, I'm actually excited and interested in a google product. 

    A few things pop to mind. As a basic premise, there needs to be acknowledgement that this is indeed a robotic system under the control of, and for interaction with humans. Like self driving cars, humans will intentionally screw with them. Some humans, being called by a robot, will either simply hang up (I do that when I reach someone else's robot) or will be tempted to screw with them. Give a fake name, don't book the appointment, then blame the callers robot. Think "we don't serve their kind" in the cantina scenein Star Wars ("IV"). Pretty sure the consumer will need a receipt directly from the other party, not their robotic confirmation. And oh yes, my robot will first inform the other party they are on a recorded line for "quality assurance purposes." 

    These systems will also need to navigate a call to someone else's robot, aka the phone-hell system systems of press 1 to hear how technology doesn't improve anyone's lives etc. Oddly enough, robot to robot stuff should be the easiest part. Take humans out of the loop, and this get dramatically more predictable. Betcha google knows this in spades.

    Many call centers need info to get started. Will your robot be authorized to speak your SSN or other ID (credit card number?) into a call? Because..well...you know...

    This is the first time I've seen something on this AI stuff that gets my intention. Kinda interesting, and I am now (edit) curious what Apple will say about it at WWDC.
    “OK, I’m just going to need a credit card number before I can book your appointment...”
    If so it would only get a unique tokenized one and not the actual card information.
  • Reply 24 of 30
    dysamoriadysamoria Posts: 2,283member
    KITA said:

    In the demonstration calls, Google Duplex used different voices, asking to place an appointment and offering the business relevant details when required. To make the virtual assistant seem more realistic, it included a number of filler sounds, such as "umm" and "ah," while speaking or before responding, with the cadence of voice also varying when appropriate.
    That's really impressive.
    No, it's odd. Filler sounds are an undesirable feature of human speech, to the point where it's virtual (and actual) pathology for some people.

    It's also creepy: when a device does not possess actual intelligence, its designers should absolutely not attempt to leave the user with a false impression that the device has actual intelligence. Have you never dealt with such systems in robocalls? It's frustrating to deal with unwanted calls in the first place, but it's even worse when it's a call by a machine that's designed to fool you into thinking you're talking to an actual human being.

    If software possessed actual intelligence, then my argument would be mostly nullified. That isn't the case. There is no Artificial Intelligence, merely clever algorithms driving scripts and parsers. AI is nothing more than an inappropriately used buzzword.
    watto_cobra
  • Reply 25 of 30
    gatorguygatorguy Posts: 21,107member
    dysamoria said:
    KITA said:

    In the demonstration calls, Google Duplex used different voices, asking to place an appointment and offering the business relevant details when required. To make the virtual assistant seem more realistic, it included a number of filler sounds, such as "umm" and "ah," while speaking or before responding, with the cadence of voice also varying when appropriate.
    That's really impressive.
    It's also creepy: when a device does not possess actual intelligence, its designers should absolutely not attempt to leave the user with a false impression that the device has actual intelligence.
    Hasn't Apple tried to promote the idea of "intelligence" behind Siri? Well yes they have with little humorous responses to some queries, cute comebacks when specific things are said to it, etc. as tho Siri is coming up with it "herself".

    With that out of the way it does appear Google's AI just passed the Turing Test, a landmark in itself. 
    edited May 2018
  • Reply 26 of 30
    dysamoriadysamoria Posts: 2,283member
    Responses to other people aside, I find the notion of enabling software to make calls for people to be an idea that is fraught with [should be obvious] concerns about privacy and control over one's own device. Even if you trusted the developer (and you shouldn't trust one who's business model is to make you their product), you should still be concerned about providing functionality that may be (and usually is) exploited by malignant entities. When cell phones are increasingly being used as methods of identification, inserting APIs that allow software to send messages and make calls to others is a risky endeavor.
    watto_cobra
  • Reply 27 of 30
    gatorguygatorguy Posts: 21,107member
    dysamoria said:
    Responses to other people aside, I find the notion of enabling software to make calls for people to be an idea that is fraught with [should be obvious] concerns about privacy and control over one's own device. Even if you trusted the developer (and you shouldn't trust one who's business model is to make you their product), you should still be concerned about providing functionality that may be (and usually is) exploited by malignant entities. When cell phones are increasingly being used as methods of identification, inserting APIs that allow software to send messages and make calls to others is a risky endeavor.
    ...which is why I expect this to be used in enterprise well before it's offered on the consumer side.  As for automatic messaging/replies we already have that with Android Auto and Carplay. There's also the automatic calls to 911 offered by our smartphones. Baby steps...
    edited May 2018
  • Reply 28 of 30
    eightzeroeightzero Posts: 2,487member
    gatorguy said:
    eightzero said:
    I'm now seeing some of the demos. Pretty interesting. Given these are in fact just demos, it is worth considering the potential. For perhaps the first time, I'm actually excited and interested in a google product. 

    A few things pop to mind. As a basic premise, there needs to be acknowledgement that this is indeed a robotic system under the control of, and for interaction with humans. Like self driving cars, humans will intentionally screw with them. Some humans, being called by a robot, will either simply hang up (I do that when I reach someone else's robot) or will be tempted to screw with them. Give a fake name, don't book the appointment, then blame the callers robot. Think "we don't serve their kind" in the cantina scenein Star Wars ("IV"). Pretty sure the consumer will need a receipt directly from the other party, not their robotic confirmation. And oh yes, my robot will first inform the other party they are on a recorded line for "quality assurance purposes." 

    These systems will also need to navigate a call to someone else's robot, aka the phone-hell system systems of press 1 to hear how technology doesn't improve anyone's lives etc. Oddly enough, robot to robot stuff should be the easiest part. Take humans out of the loop, and this get dramatically more predictable. Betcha google knows this in spades.

    Many call centers need info to get started. Will your robot be authorized to speak your SSN or other ID (credit card number?) into a call? Because..well...you know...

    This is the first time I've seen something on this AI stuff that gets my intention. Kinda interesting, and I am now (edit) curious what Apple will say about it at WWDC.
    “OK, I’m just going to need a credit card number before I can book your appointment...”
    If so it would only get a unique tokenized one and not the actual card information.
    That wold be cool. But of course, you'll need to teach your robot how to handle that too. 
  • Reply 29 of 30
    AI_liasAI_lias Posts: 305member
    chasm said:
    Let's see how it responds when the called entity opens the conversation with "For English, press 1. Our menu options have changed. For Luis, press 7. For Andrea, press 8, for Camelita, press 9. Otherwise, leave a message."
    Funny, but you raise an interesting point. Companies could use Google’s AI to replace their own systems, so both the customer and the business would hand off the drudgery of negotiating details to these systems and then the AI would simply inform both parties after there is agreement or schedules sync. Of course, the real test comes when there is no agreeable conclusion to the negotiation. Does the AI go back to the user and inform them of the failure or does it explain what happened? What happens when a child answers the phone instead of an adult? There are a near infinite number of things that could conceivably go wrong.

    And this raises another interesting point: if that were to happen, there is no need for voice, but the assistants (Google, Alexa, Siri) can talk to each other programmatically, via APIs, and much faster. 
    watto_cobra
  • Reply 30 of 30
    SpamSandwichSpamSandwich Posts: 31,404member
    gatorguy said:
    dysamoria said:
    KITA said:

    In the demonstration calls, Google Duplex used different voices, asking to place an appointment and offering the business relevant details when required. To make the virtual assistant seem more realistic, it included a number of filler sounds, such as "umm" and "ah," while speaking or before responding, with the cadence of voice also varying when appropriate.
    That's really impressive.
    It's also creepy: when a device does not possess actual intelligence, its designers should absolutely not attempt to leave the user with a false impression that the device has actual intelligence.
    Hasn't Apple tried to promote the idea of "intelligence" behind Siri? Well yes they have with little humorous responses to some queries, cute comebacks when specific things are said to it, etc. as tho Siri is coming up with it "herself".

    With that out of the way it does appear Google's AI just passed the Turing Test, a landmark in itself. 
    I'd love to hear a lengthy conversation between a person at a store and this AI to find out at what point the conversation starts to fall apart. Can it engage in idle chit-chat? If something comes up that requires additional information from the person who instructed the AI, what happens then?
    edited May 2018 watto_cobra
Sign In or Register to comment.