Duplicate https://news.ycombinator.com/item?id=17022963

A bot calling for you and a bot answering for salons/restaurant. Both pretending to be human. It’s like having a client/server application speaking English with an accent, without versioning, schema nor content type. What could go wrong.

Well that sounds exactly like humanity. We all know how that turned out.

That's a good point. We don't suffer from all those things mentioned. We have a living API that's meant to be very flexible and adaptive.

We suffer from it all the time. There's a beautiful and tragic poem called "The Charge of the Light Brigade" about the results of a failure to communicate orders well. There's a statue in Rome depicting Moses with horns.

Also the plot of Her.

From the tone deaf product org that brought you google glass and whatever wave was, comes a solution that does not so much pretend to be human for you as it does simply dispense with the need for you pretend you possessed any humanity whatsoever at all.

Next, google Snap! A robot that snaps its fingers at service people to get their attention...

I'd argue that this product isn't as much tone deaf as it is a brilliant way to force businesses to move to using online reservation systems. Phone calls always take longer, and rather than just selecting a time you have to present times where there this takes longer when things are already booked.

>a brilliant way to force businesses to move to using online reservation systems //

For micro businesses this can be a substantial administrative cost. A paper diary is still the best solution for some IMO.

Or they can just start hanging up when this stupid thing calls them.

Seems like a great way to lose business.

You just got called by an AI trying to make a reservation for someone. Do you:

a) Let it book the reservation, and take the customer's money when they show up.

b) Hang up on the AI, potentially losing a sale. "Sorry, I tried making a reservation but they hung up on me."

Some retail businesses don’t accept credit cards.

Point being: stranger things have happened.

You would turn down an appointment/sale because it was scheduled by a machine with a human voice?

Yes, absolutely.

I avoid calling customer service because it forces me to work with a machine.

Yes. I'm perfectly willing to be discerning on who I do business with.

Same—we just draw our lines in different places.

It's been 15+ years since I've worked in the service industry, but from what I remember of it, I think I'd prefer to take reservations & catering orders from a Google Assistant.

And lose customers for no real reason. Great strategy.

I think that it was fairly obvious from the demo that this was demo-ware more than anything else. OTOH, the real use case there seemed to have real value. I can imagine that, in many cases, the request for store hours would be a net benefit.

At some point both sides of the conversation will be done by AIs...

Eh, more likely the AI phone reservation system will also plug into something like OpenTable or it's descendants and Duplex and it's descendant will just check there first.

Maybe.. but voice is a common data format, like piping text streams in UNIX. Don't need gatekeepers like OpenTable...

Yeah and the front of house software we're talking about probably plugs/will plug into multiple similar services. Voice isn't a good analog for unix's text piping because it has to go through encoding and decoding on each end and passes through the lossy phone lines and then through interpretation on top of all that.

Oh, I didn't mean it to be a perfect analogy, because nothing is. Text piping is fragile in other ways because it can be broken trivially by changing the order, formating, padding etc, in ways that voice isn't.

This will break in humorous ways.

Well put. 100% spot on.

edit: Google Glass has it's uses. The problem was that Google marketed it to consumers instead of businesses.

It was never "marketed" except to devs and status signalers. The only one I ever saw in the wild was from a PHB with no use for it.

FWIW, it was great as a status signal! I was in a Panda Express with it once and got asked if I wanted a job by another customer who worked in a nearby building!

I wore it for about a year, I still think it was way better at notifications and stuff than the current smartwatch trend. But the main problem was the company making it and it's insistence on first party control. It didn't become possible to get non-Gmail email notifications on it until near the end of it's life.

I read the first paragraph in Movie Narrator Voice and was not disappointed.

That product reminded me of Facebook’s M, a Messenger-based assistant that wasn’t able to scale. I guess Fb didn’t have two years ago the tech that Google has now.

M’s key uses cases are probably going to prove essential here: aside from the usual flowers-for-my-wife-on-our-anniversary and other difficult-flights-to-book, basically busy executive stuffs, the massive potential was to deal with painful cancellation processes for your ISP, bank, gym, etc. If Google (or a developper using this API) can handle the intentionally convoluted interactions that those imply, this could be very powerful and make using Assistant the best way to handle some painful cases. Imagine applying for a loan, refinancing, switching to a better phone contract whenever there is a good offer: this technology could advise people to move between services with significant usability moats much faster, and greatly improve competition and our experience.

That is unethical, the human on the other side should be made aware that he's talking to a machine instead of a human.

I disagree about the ethics of this. This is no different than having your (human) assistant make an appointment for you.

I'm assuming that if the machine can't determine what to do next, a simple "I'm sorry I'll need to call you back" and notifying the user that they need to call manually is all that will be needed, much like how a (human) assistant that is faced with a question they can't answer alone will need to excuse themselves and call back at a later time with that information.

Now obviously if this ends up wasting a significant amount of time for the business for no gain, I could see an argument about ethics, but making the receiving end aware it's a machine won't solve that problem.

Also, I think making the "user" aware that they are talking to a machine might reduce it's capability significantly. As long as the person on the other end thinks it's human, it seems like it can do a pretty good job. Once the other side knows it's a robot, they will dumb themselves down to how they expect they should talk to a robot, which means they won't say things like "Is 10AM okay?" or "what kind of service do you want done?".

> I disagree about the ethics of this. This is no different than having your (human) assistant make an appointment for you.

Except your assistant is a human, and this is not. I can't quite put my finger on it, but I get extremely upset when I'm called by a robot. (If it's important enough to call me, it's important enough for a human to call me.)

Is it possible the frustration is not because the voice on the other end of the line isn't human, but instead because past attempts at automated phone interaction sucked?

It's a claim on his time that isn't accompanied by the show of good faith of a comparable claim on their time. If we were as cavalier about capturing someone's attention on the phone as we are about sending them emails, we'd have to stop taking phone calls from all unknown numbers full stop.

Ah, okay.

I'm possibly not picking up on that signal since in my personal life, I've stopped taking phone calls from all unknown numbers, full stop.

But we're you called by a robot at your job it was it interrupting your personal time?

I agree with you in the latter case, but the former seems like a win win assuming this system can deliver on what they are showing off here.

And yet this is what every large company has been doing for the last decade plus via call centers etc, Chinese-room style!

In fact, while I don't look forward to whatever collateral damage this unleashes on local businesses, I do relish the obvious application of this technology to dealing with the ever-sprawling "customer service". How fraudy do companies want to make their cancellation scripts when it results in only their own resources being wasted?

It easily could be used unethically. How about hard to get reservations that become available at a certain time or the only way to get one is to keep trying (a real human wouldn't stand a chance against a Google robo-call).

Or being used by a restaurant competitor to keep the phone lines busy with endless reservations and cancellations. Nothing you can't do with a motivated human of course, but the cost and ease of use makes a difference I think.

It seems a very obvious pattern that Google would be able to detect.

And you think Google will be the only ones using this technology for the foreseeable future?

Honestly, handling a realistic voice is incredibly hard; being able to parse context, including from structured, partial information, even more so. I don’t know if the progress in training processors, LSTM structure in the next year, but I doubt many companies will have enough conversations to train their models -- other than the operator of, say, Google Voice.

I am willing to believe that any company who can offer that service as a service will know that they need to monitor for abuse, and have a very sophisticated team to catch those. That might include Apple, Amazon, Baidu (who probably won’t launch it in English). Facebook wasn’t able to see something like that working and they abandoned M four months ago.

Can you explain your views on the ethics?

I think it is a little creepy, but ethically? You're conducting the same activity in the same way you would have done yourself, but now more conveniently. The recipient still has the same experience either way.

My biggest concern about this is that it could be used at scale for "evil" (e.g. junk calls).

My bigger concern would be social engineering attacks. Once you have something that can reliably imitate a human, the potential to do social engineering at scale really takes off.

I don't know if its completely unethical, but I do think its a little dehumanizing to deny someone human interaction, when using means that only humans can employ. i.e. If it sent a clearly marked automated email its one thing, but imagine having this contraption call your mom to wish her on her birthday...

I don’t think that Google has plans to train that system on individual’s voice. I’m not sure what the scale for doing that reliably would be.

1. Making automated, unsolicited robocalls is unethical.

2. This is an automated, unsolicited robocall.

> 1. Making automated, unsolicited robocalls is unethical.

I'd say that any restaurant or salon that treats a customer booking a table or an appointment as 'unsolicited' might not stay in business for very long.

I can see it going the other way:

Robo-booking being used by people who simple put an entry in their own diary - the robot doing the booking and interaction. The human potential client has used minimal resources and because of the ease of use and lack of interaction will not perceive a cost in the making of the booking. I anticipate they'll change, cancel, noshow at a far higher rate than in-person bookers.

Bookings made might then both block other customers and potentially increase the cost to the supplier of administrating the booking [particularly with a micro-business where the supplier end is an actual person].

There's not a great deal more friction involved in making a booking via OpenTable (no financial cost, no deposit etc).

The law and philosophy around automated, unsolicited robocalls is related to telemarketing. It's telemarketing practice that people became so frustrated with that the Do Not Call List was enacted.

This, in contrast, is automated deal-making, targeting phone calls at businesses that want to close a deal. The context is entirely different.

Perhaps a case for robots.txt for phones?

Of course malicious behaviour will not comply, but would at least filter the mainstream products.

Why do you describe the calls as "unsolicited"?

Exactly. Calling a business for the purposes of buying their product is the definition of solicited.

That's literally the reason they have phones.

Because the restaurant did not explicitly opt in to receiving robocalls?

It's as ethical as companies unleashing chatbots and voice-synthesized phone "customer support" gatekeepers on us. Although, I would prefer to have conglomerates taste their own medicine rather than local shops.

Good point. This could potentially let consumers fight back against phone trees and brainless tier 1 support. If I need my cable internet fixed, Id be willing to spend good money to hire one of these bots if it could:

1. Call Comcast

2. Navigate the initial “go away” phone tree

3. Wait on hold for 30 minutes

4. Negotiate with the human tier 1 support drone until they escalate

5. Finally call me back and connect me when it successfully locates an actual T2 support person who can fix my problem

A bot to get your Comcast bill lowered already exists: https://www.theverge.com/2016/11/17/13656264/comcast-bill-ne...

I assume the same is possible for navigating Tier 1 support, though you are really setting a bar for describing people as subhuman at this point. I'm concerned.

I'd argue it's significantly more ethical. There's a world of difference between telemarketing and closing a deal with a business that has an express interest in closing a deal because they want your money.

May be the salon, restaurant also have a bot talking to Google assistant.

And then one day instead of those bots talking through natural language we can have them talking to each other by modulating and demodulating the sounds to create ones and zeros. Maybe even call them MO-DEM bots. And then one day we can make those modems faster and faster, and have those bots communicate using a standard protocol ;)

Amount of compute power wasted for an AI bot to AI bot voice conversation to make a reservations somewhere boggles my mind.

... agreed, only it doesn't boggle my mind as much as the fact that human beings get employed full-time to sit near a phone and do this entirely-automatable task every day.

The ultimate inefficient API.

That’s unlikely. If they had claimed their business and had an active Google Maps account, the request would come to their email for now, and Google will probably develop an integration to their Point-of-Sale software, or surface their OpenTable link.

Possibly illegal (?) if the conversation happens in a jurisdiction which requires both parties need to consent to be recorded.

It's illegal, too - in at least 12 states, you cannot record people without their explicit permission. Impersonation could be a legal issue as well.

Who is getting recorded?

For this to work at all there is at least transitory recording of voice on the call, but it's not necessarily all that different than what had to happen when a computer is part of the community in channel of a call, which is common now with IP telephony. If they aren't doing more and retaining recordings made without consent, I think it quite unlikely this falls afoul of the recording consent laws.

If they do record more than transitorily without the other party’s consent in those jurisdictions, that’s illegal but it's also fairly simple to address.

> but it's also fairly simple to address.

Don't do business in those restrictive jurisdictions?

Many crimes are easy to prevent, yet, still committed!

My point is that if there is a legal problem, it Will be trivial to fix and have little substantial impact on the prospects for the system.

Did you see the IO keynote where they played the recorded call? And said other calls are being recorded?

I assumed those were recorded with consent (or in jurisdictions where unilateral recording is legal).

Unilateral recording is old law; I wouldn't de-facto assume Google is just blowing past that without some much stronger evidence.

Twelve states in the US are two party consent states (both participants must consent), including California.

https://en.m.wikipedia.org/wiki/Telephone_recording_laws

https://codes.findlaw.com/ca/penal-code/pen-sect-632.html

Why is it unethical?

Because the person in the salon or restaurant shouldn't be forced to talk to a bloody computer instead of a human which shows some emotions.

What if you literally can't talk like stephen hawking? Or have crippling social anxiety? Or you don't speak the local language very well? Is it still "unethical" for those scenarios?

It wouldn't be "unethical" if it is plain that there is a machine involved. Google could e.g. say something like 'hi, google assistant here' and clarify the situation (and give a human person the choice to proceed). What do you think?

What if, as the demonstration showed, English is not your first language and you struggle on the phone?

What about foreigners on holiday, could their ask for a reservation in one language to Google Assistant and have it handled properly?

Intonation is not the only way of showing emotion. (Regardless of whether I agree with it being unethical, a robot voice != a talking robot.)

Is it unethical for me to make a reservation using a completely deadpan, monotone voice?

The deadpan voice would fall back on you, a computer won't care. But I agree with what I suppose you want to point out, that ethics is not the relevant criterium.

The ethics of it is the interesting part though, if there's a reason we should not do something it should be discussed.

For me, "ethics" is a too strong word for this discussion.

What is the right word, to me ethics determines if something is good or bad, which I think is what you were trying to say.

So it's ok to force someone to talk to a human, but not a computer?

How much emotion should I be showing when making an appointment. I could see your point for familiar shops like a frequented salon or something.

It's unethical I'm forced to filter junk mail from my important mail, but here we are.

Your own personal robocaller. Your droids we don't server their kind here.

Google is so creepy.

I read that as Saloon at first and now I am so disappointed.

> Anyone else think it's kind of silly that half the front page of HN is google press releases today?

I count exactly 1/3 that concern Google, most of which are not press releases (or other direct-from-Google stories), though the news articles may be heavily based on press releases.

But, no, it's not surprising we tend to get a lot of that on I/O keynote day. MS and Apple dev conferences also produce a lot, especially on the day of the keynote.

In the whole eight days that you've had an HN account, you might not have noticed that every time Google/Apple/Microsoft/et. al. have a dev conference, the front page of HN is loaded with $COMPANY_WHO_IS_HOSTING_CONFERENCE news.

Not to mention things like $famous_ceo_died, or $gov_conspiracy_theories_validated_by_leak.

It's google IO. So no. It's not silly.

I don't need software to pretend to be human for me as I am already human and do not need to pretend.

Corrected title: Google Duplex will call salons, restaurants for you, and pretend to be human.

Please don't post funny jokes on HN.

By all means, post funny jokes. But don't make the same mistake as op00to and post jokes that aren't even remotely funny.