The demonstration of Google Duplex has both stunned and terrified the world. AI can do that.

For some legal insights into this, check out Jay Ward’s discussion on his blog post here.

If you haven’t watched the video yet, its worth the few minutes to understand what all the commotion is about. We play the clip as well in the podcast. The number of questions that an AI discussion with a small business owner creates are too many to enumerate, but we break down some of the more obvious concerns today on this episode of Are You DataSmart?

TRANSCRIPT

Jay: Are You DataSmart? A weekly podcast on data security, information management, and all things related to the data you have, how to protect it, and maximize its value. I’m Jay Ward.

Christian: And I’m Christian Ward. And today we’re gonna tackle, I think, what everyone in the data world has been talking about at Google’s IO Conference, an amazing demonstration of AI, the assistant Google Duplex. Jay, you wrote an article called “Okay Google, Is This Legal,” which had a lot of funny insights into this. And we’re gonna play the portion of the Google keynote where Google Duplex, a newly revealed AI voice capability actually calls a small business to schedule an appointment. So we’re gonna play that for you right now so we can then follow up and discuss it.

Female 1: Hello, how can I help you?

Female 2: Hi, I’m calling to book a woman’s haircut for a client. I’m looking for something on May 3rd.

Female 1: Sure. Give me one second.

Female 2: Mm-hmm.

Female 1: Sure. What time are you looking for around?

Female 2: At 12 p.m.

Female 1: We do not have a 12 p.m. available, but closest we have to that is a 1:15.

Female 2: Do you have anything between 10 a.m. and 12 p.m.?

Female 1: Depending on what service she would like. What service is she looking for?

Female 2: Just a woman’s haircut for now.

Female 1: Okay, we have a 10 o’clock.

Female 2: 10:00 a.m. is fine.

Female 1: Okay. What’s her first name?

Female 2: Her first name is Lisa.

Female 1: Okay. Perfect. So I will see Lisa at 10 o’clock on May 3rd.

Female 2: Okay. Great. Thanks.

Female 1: Great.

Jay: Yes. So the scariest part about that for me is the um and the mm-hmm. And you heard them laugh at it in the audience, which is, you know, I guess that’s a natural human reaction to something that’s terrifying and we can’t understand what we’re seeing. I got that reaction a lot in high school. For me, this is fascinating and powerful and a demonstration of just the incredible advances in technology unthinkable even a few years ago. But it also represents some substantial legal and ethical questions. You know, how are we going to navigate the legality of these calls and the legality of AI interacting with each other, and this is completely uncharted territory.

Christian: Yeah. So I wanna go back to something you said of the uncomfortable laughter in the audience. I really keyed on that because for many of us, as we’re doing with AI, we’re starting to see things that give us that really uncomfortable pause and laughter. And I could feel it in the audience. You could sort of sense that when it says, you know, “Hi, I’d like to schedule…” When it does that, the other thing I realized is, you know the old, it might be a myth it might be true, it’s probably partially true, the old concept of the word QWERTY, Q-W-E-R-T-Y. It was originally designed to have an inhibited speed. And what I mean by that is the typewriters that we originally used and the pattern for the typewriter was slowed down by the placement of the keys. And the idea was that most used keys are actually on the left hand, the statistically less dominant hand.

And so, it was meant to slow down as a typist types, they didn’t want the hammers striking the paper to get jammed together, which happened if you typed too fast. I don’t know if you remember, Jay, but grandma had this Smith Corona Typewriter that we had in the basement for years. I know our other brother Brian now has it in his basement. It’s an antique. But I used to type on that thing, and that actually happens. The mechanical, the mechanism itself jams up when the human goes too fast.

What the arms are in the AI of Google’s Duplex talking to the human is, the AI is slowing down so the human doesn’t jam up, which is really concerning because it was the first time I have witnessed where it was obvious the machine knows what its next question is. It knows what the answer should be. It’s cued up, it’s ready to go. But the human has to sort of gate the conversation. And so, I literally felt like I was watching, and we all were watching Google’s AI QWERTY, a human, instead of what classically has been the other way around.

Jay: Yeah, it was, again, kind of terrifying and interesting at the same time. I wanna know who the coder was, who made the Duplex have the upswing in there at the end of their sentences. “I wanna schedule a hair appointment.” It’s like we’ve decided like we’re gonna go with 13-year-old as the lingo. For me, I was intrigued because this is the definition of a robocall, right.

Christian: Yeah.

Jay: This is a robot calling. And I was wondering as I listened to it, well, where are the disclosures that are necessary? You know, in a two-party state like New York, you need to have the consent of both parties on either end to record the conversation, otherwise, it’s illegal.

Christian: And what is…When you say New York, I know state to state there are different laws regarding recording. Can you just real quickly…So if it’s a robot that calls or it’s recorded, because anything that’s being analyzed by AI, by definition is being recorded at the time. They might make the argument they’re dumping the recording later, but it’s being recorded in order to be analyzed, correct?

Jay: Yeah, and that’s the thing. I mean, there are two streams here to think about. There’s the state law stream of, you know, classically when we’re talking about one party or two party states, it’s I’m on the phone, you know, I turn on a tape recorder and I record our conversation. If you’re in a one-party state, that’s enough because I’ve consented to the recording. If you’re in a two-party state, you would need to get the other party’s consent to record. And if you’re not getting that consent, it’s illegal.

And in some states like California, if you don’t give the warning or a disclaimer when you got, you know, for instance, an automated call that, you know, this call is being monitored for quality assurance, that’s illegal and can come with a $500 mandatory statutory penalty. Yeah. I mean, so if this was being made from Silicon Valley, this call, you know, we’re talking about a potential fine that gets pretty high. And if we’re talking about, you know, a lawsuit related to the recording of someone’s voice. Now we’re talking about litigation where people know because it’s been demonstrated by Google that definitionally, these conversations are recorded because it’s being transcribed and kept either by Google or by the user. And that’s a point that I wanna make here, and that’s really important. Google can say, “No, no we’re gonna give disclaimers. We’re gonna do all of these things to make sure that there’s no issue with complying with the law.” But Google isn’t the one making the call, Google Assistant is making the call.

So unless Google is taking steps to ensure that you can’t use Google, as if you can’t use Duplex without it saying, “By the way, everybody, this is Google Duplex calling,” you’re gonna run afoul potentially of the robocalling requirements, you know, that the FTC enforces really stringently. They do workshops on it all the time. They’re very serious about it. You’re gonna potentially run afoul of those state law requirements on disclosures and disclaimers about having a recorded call. So, you know, my concern here is not necessarily that Google, which has really smart lawyers, is gonna be making use of Duplex appropriately because Google doesn’t need to. It’s when Joe’s flower shop starts using it or when you personally use it that you might be running into these problems. So there’s just a huge array of potential legal issues here.

Christian: Okay. But it’s still really cool. We can agree on that. Right?

Jay: It’s so super cool. It’s so cool.

Christian: So one of my friends I saw tweeted a comment where it said, “Okay, Google, call the top 25 restaurants in our town and get me an appointment for dinner at 7:00 p.m. tomorrow night. And I started looking at that and going, “Holy cow.” There’s not just the legal aspects, there’s also just the operational aspects. So typically there is a, let’s say, an unspoken contract in the effort it takes to schedule a dinner somewhere. So let’s say at 7 o’clock at night. I’m gonna call a bunch of restaurants. If I do that, I’m putting in time, I’m putting in effort.

But if I can ask my AI, like a professional executive assistant to go analyze what restaurant I should go to and scheduling an appointment, I’m transferring that effort to the AI, and subsequently, I’m transferring it to the business. The onus becomes on the business because what starts to happen is, if I can call 25 restaurants, let’s say asynchronously, so I can call them all the same time using the AI, where it initiates 25 phone calls. What’s gonna happen is I’m gonna end up with a sort of different variable set of, “Yes, I got an appointment. No, I didn’t.” And we know that Google can allow for you to maybe do a 6:30 or 7:30 if you give it a range.

So you’re gonna end up with, you’ve got seven different reservations ranging from 6:15 to 7:45 at seven different restaurants. Google’s then gonna go and cancel all of those, which is really damaging to the business owner because it’s one thing, Jay, if you and I decide, “Hey, I’m in Miami visiting. Let’s go to dinner. Let’s take the families out.” When we do that, that’s one instance. But what if in, you know, Coconut Grove, or in Miami, wherever we are, 7,000 people that evening say, “Okay, Google, call the top 25 restaurants.” We have a serious, serious problem, which ultimately, while, it’s cool, it also could be at the level of complete devastation to small businesses where they can’t keep up with the voice calls that are now robotically driven.

Jay: And honestly, I think we’ll see, as a result of that, you know, people use open table now but, you know, get ready for AI reservation management systems that small businesses use to field all of these calls, but you see then what’s gonna happen is, we’re not gonna be having phone calls made, where it’s a person talking to an AI or the other way around, it’s gonna be AI to AI communication. And when that happens, they stop using language that’s understandable to us. I mean, I think you remember the story about Facebook’s AI communicating with itself and reducing hours worth of human conversations like you, I, I, I, everything. It’s not comprehensible to us.

I mean, they could have been giving Roy Batty’s speech from the end of Blade Runner, you know, “Tears in the Rain”, but to us, it looks like they’re making a reservation to go to Chipotle. I’m just kidding, Chipotle doesn’t take reserves. So, in that situation, there’s obviously gonna be substantial displacement when we’re talking about, you know, people taking reservations, and that’s an entire conversation, that’s an entire debate unto itself. But at what point does the handshake between these two AI interactions become a focal point for regulators? Like when are they gonna start saying, “I need to be able to see what’s happening in there. I need to be able to understand these interactions because I can’t see it. It’s not transparent to me.”

Christian: Yeah.

Jay: And transparency is the key for regulators.

Christian: Well, and I think something else that’s gonna happen, and we’re gonna touch on a couple of things. Number one, that it’s impossible to step through what AI is thinking at the highest levels of artificial intelligence. So we’re not even gonna know why the machine elected to go with this reservation, with this restaurant versus another. And that presents problems because if you can’t understand why it made the decision, then quite frankly, transparency into that craziness is not gonna help you. So that’s one thing.

The next thing that we’re gonna have to talk about is, for a business, I don’t know if you remember, there was a, there still is a platform, it’s called GetHuman or there is all these websites where, instead of listening to the monotonous, annoying, “For this press one, for this press two,” it would just tell you, “Hey, if you’re calling Delta Airlines, hit 5521417,” and you get a human. It was really very smart. It was just like a life hack. What you’re talking about is the automation of that at a much higher level, which is, I can have my AI talk to yours and assist me through my life.

I think the real concern, and this is GDPR as well as really any privacy-focused or privacy-oriented company’s viewpoint, is we’ve got to protect people against how this is operating. And realistically, we’re gonna do a podcast soon here on AI Dreaming and the subconscious of AI. If we can’t unpack how decisions are actually being made by the AI in many ways we can’t unpack our own subconscious, that is gonna present these laws with a whole another scope of problems, which, look, I’m excited to see us tackle.

I think the Google demonstration, by the way, is absolutely brilliant. I think it’s amazing. I think it not only changes the whole booking concept or construct, but it also opens up the opportunity for a far more day to day efficiency in our lives where we were hoping every business or small business owner would be totally studious about what availability they have and everything else. We all know that’s not true. They’re busy.

So here’s a way to find a way you sort of bridge the gap where they’re not keeping their calendar set. We could still have the machine talk to a human. But in the end, that is going to put a lot of stress and strain if there aren’t two AIs working this out. And then, at some point, you just have the entire restaurant reservation system of a city, AI’d together, where the AI says, “You know what, for Christian, we know he likes this type of food. Let me only reach out to these five AIs even though he said the top 25 restaurants.” Those decisions without insight into how they are made is the part where the law still isn’t ready for what’s about to come.

Jay: And then we can call that system Skynet, and it’s perfect. We’re all set.

Christian: Yes. Yes. Well, you know, Sky food maybe. Sky restaurant.

Jay: You’re right.

Christian: Sky dinner.

Jay: Sky mall, everything good starts with Sky…

Christian: Exactly. Excellent. Well, thank you, everyone, for joining us for this episode of Are You DataSmart? We’re gonna keep digging into this. And if you haven’t bookmarked that Google Duplex video, do so now because it’s the most terrifying and wonderful thing you’re gonna see this week.

Jay: Thanks again.

Leave a Reply