The Uncanny Valley of Robocalls


#1

So I’ve been receiving unsolicited junk mail trying to sell me a warranty on a used vehicle I purchased from a private seller, because I had to register the sale to comply with state laws and somehow my mailing address became part of a public record warranty con-artist companies gained access to. Today they demonstrated they also have my phone number, as they called me. Normally I don’t answer numbers I don’t recognize, but I was expecting another call from a company that had a similar number, so I mistakenly answered it. What happened then is the interesting part.

The woman on the line starts the preamble to her pitch and ends with asking if I’d received the mail they sent me. I said, not at all rudely (the rudest thing I do on phone calls is hang up, but I was feeling uncharacteristically hopeful I could make this the last time they tried calling me, so I answered her), yeah, I don’t want to buy a warranty, thanks but please stop calling. She starts off on another canned response from whatever manual she was trained to spout. But she must have rehearsed her responses, because she sounded so much like the canned responses recorded on an automated phone system that I interrupted her and asked not meanly but with genuine uncertainty, is this an automated call? That was the first thing that caused a unscripted response as she seemed momentarily confused at being taken for a robot (yes, insert joke, but that still has to be pretty weird to have someone not be sure you’re real). She chuckled just a little nervously and said no, I’m a real person. Must be a bad connection or something.. At which point I said oh, well, I definitely don’t want to buy a warranty. Thank you and please stop calling. Then I hung up.

But it wasn’t a bad connection at all. I just had a moment when it occurred to me that it would be trivial to write a fairly simple program to toss canned responses at someone based on key words in their replies, and I really wasn’t sure if it was a response tree of well done recordings or an actual human telemarketer.


#2

You still can’t be sure it wasn’t a canned call. This is exactly how I would program my robot to respond.


#3

I don’t get many automatic calls anymore. The most ridiculous ones I have experienced are where the robot tells the that there’s a great deal, and that they need to put me on hold so that I can talk to someone. They call ME, and then they promptly put ME on hold. Does anybody actually wait for the privilege of talking to the telemarketer?


#4

I have no clue how my name and phone number ended up on not sure what list (admittedly it was listed at the time) but we kept getting calls from all sorts of scams saying we explicitly asked for info. One that ended being a real person would actually give me some info gave an obvious throw away email address that I supposedly had used. Every time I didn’t get a robot and asked for contact info so I could send them an official cease and desist letter CLICK.

Anyway after a month or so I dumped the land lane completely as they would only block specific numbers and at a fee and caller ID was more money, etc. etc. When I asked why can’t I get all the cell phone things I get as part of the base package of even $50/month which was advertised for unlimited calling/texts with caller id and voice mail, etc. at the time. They just kinda said sorry we will kill the land line and the telcos wonder why they lose business?


#5

Exactly. It will get harder and harder to tell as time goes on, and this is an area where we’ll reach the Uncanny Valley before things like Skype and physical robots reach it. Chatbots are already well on their way into the Valley. As telemarketers find their jobs automated, I can see a whole new field for psychology applied to intentionally tricking people, albeit in the opposite direction from which I was unintentionally tricked.

The consumer in me was like, damn, I took a telemarketing call. But the geek in me was all huh, that was weird.


#6

IIRC from a few years ago, IBM Watson’s Natural Language Engine could probably handle all of that and not skip a beat. As long as there’s sufficiently accurate voice recognition going on.


#7

Yeah, I feel like accurate voice recognition over a phone line is probably the biggest thing standing in the way of it, but that’s improving steadily.


#8

I use google voice, and when I get a voicemail, if I can understand the audio the automated transcript is 99% correct. Occasionally I get sound-alike substitution where words like “many” can end up transcribed as m’nanny. But that’s gotten quite rare in the last year or two.


#9

The one good thing about telemarketers is that, even if you retire a human by mistake, you’ve still done the world a service. Takes some of the stress out of the process.


#10

Telemarketing has passed the Turing test! All hail our new automated sales overlords!


#11

How well they can parse the expected patterns of human communication isn’t a big deal. The reason why it’s easy is because most people limit themselves to a limited repertoire of social contexts. All you need to do to know if they are paying attention is to drastically alter the script on them. The failure mode may teach you something, or be good for laffs. Parsing speech does not make these systems readily adaptive.

Three years ago, I used to be called by a local newspaper nearly every day with subscription offers. I always declined, and they kept calling. I asked them not to call in the future, to take me off their lists, and they kept calling. Finally, one day I was fed up and I informed them that I was already receiving the paper! The person sounded surprised, said they had no records of me being a customer, and asked if I had it delivered or if I bought it elsewhere. I told them that the paper was carried to me every day in the beak of a flying squid. They never called again! XD It was a human, and they audibly broke down a bit once confronted with the unexpected. If it was an AI, it would have been flustered in some different way.


#12

You haven’t won a free cruise to the Bahamas? This week?


#13


#14

If I don’t get too pissed off, I’ll string phone scammers along as long as I can. As long as they’re wasting time with me, they’re not making money and they aren’t preying on anyone else.

What I’d really like to do is send 140,000 volts at 10 amps down the line and kill them instead.


#15

I’m on board with that.

I suspect lots of these “companies” like the scammers pushing worthless warranties farm their telemarketing out to third-party call centers, or just hire their own temps. The drones may be genuinely clueless or at least willfully ignorant that they’re scamming people. Given that drones staffing a call center aren’t typically tripping over job opportunities, I can’t dig up too much wrath at them personally, though my leniency emphatically does not extend to their employers.

However, my choosing not to get angry has nothing to do with that. I simply won’t let telemarketers darken my day. They’re beneath my contempt.


#16

The trouble with telemarketing drones is similar to the trouble with onrushing conscripts.

Yeah, you have to feel sorry for the poor bastards who end up being the pointy end of a policy made by terrible people immune to the consequences; but they’ll still ruin your day if you don’t stop them.

Now, if there were a way to follow the money and administer a discrete dose of 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine to the offending party; I’d be all over it.


#17

I know. And I honestly have no ill will towards people who are genuinely ignorant. But the thing is, when I get connected to a human on these calls, if I’m angry the first thing I say is: “are you aware that your company has called me illegally?” Usually I get the click from that. Sometimes I don’t, and they play dumb or are genuinely unaware that it’s illegal for them to call people on the do not call list. I then explain: “You, or your employer is committing a felony offense by calling me, since I’m on the do not call list, and have no prior business relationship with you. You should look for a job that isn’t illegal. I hope you can manage that.”

I used to get a lot of calls from “cardholder services”, and when I was connected with a human, I’d ask “which credit card can you reduce my payments on?” They’d usually say “all of them” and I’d have to ask “that’s impossible, what’s the name of your company again?” which always gets the click.

It’s maddening. I used to get called on the land line over 20 times a day from “cardholder services”, and they’re still active. I have no idea why the FCC hasn’t been able to eradicate that scam. The only excuse I can think of is that the FCC just doesn’t want to stop them, or can’t because someone in the FCC or higher up is benefiting from the scam or is even running it, and has ordered or stalled the FCC and FTC from doing anything.


#18

More likely they don’t have the budget to take all the scammers to court, so they only go after the biggest offenders. Though I wouldn’t rule bribery out, I suspect the scammers prefer the path of least resistance and incorporate under false identities they can shed if an investigator shows some drive.


#19

You have much better luck than I do. I end up getting a lot of stuff like (and this is verbatim from an actual one): “Hello, this is well from that’s supposed to call the game for because several dulcimer, I D’s and Eve, blah blah 247. And the server is stuck in the ass at deficit, see.”

To be fair, a lot of my voicemails are from people with a little bit of an accent, but still… I don’t know what they were supposed to be doing with my dulcimer, but I don’t like it! Also, every time I read that last sentence, my brain slips into “cheesy movie mobster” voice…


#20

I’d expect that a computer learning algorithm does better with a single accent, and even better with a single dialect+ accent combination, and still better with a single person’s idiolect and vocabulary. But given a single person’s idolect and vocabulary it wouldn’t necessarily generalize to an accurate model of a language’s sound. So you have to have a lot of dimensions to compare on if you want speech recognition that is accurate for a large population like: all English speakers.

Just spitballing.