56

I recently saw a doctor about a minor medical problem. This is the first time I saw this doctor . After the appointment I asked my doctor a follow-up question via my insurance company's website. I got a call back from the doctor but during the entire call I was distracted because all of the doctor's remarks sounded so dull and artificial. Also, he didn't seem to remember anything about me and just kept telling me to make a follow-up appointment with my doctor. I thought to myself that I might have been talking to an AI. That didn't seem too far-fetched to me with the recent advances in AI and doctors' busy schedules. During the entire call I was trying to determine if he was real and almost blurted out "are you a real person??".

In hindsight, I think I was talking to a real person and he just has to deal with so many patients that he starts to talk like that. But I have started getting automated telemarketing calls that try to trick me into thinking a real person is talking to me.

Is there a polite way to ask if someone is a real person? Or if there is a way to determine if I'm talking to a real person (sort of a reverse Turing test).

13 Answers 13

70

I once worked at an IT-Service hotline. I answered a call like I always do and after my standard greetings phrase the customer said nothing so I asked, "Hello? Are you still in the line?" The customer replied, "Oh, wow! You are a real human! I thought it was a band announcement!" Later he explained that my voice was so perfect that he thought it was a remastered recording and that I could easily work as a radio moderator.

So you could say something like:

Oh wow your voice sounds so perfect/good/nice/great. You could work as a radio presenter/moderator.

With this you make him a compliment and don't offend him.

Just for clarification of the meaning: I didn't mean this compliment sarcastically.
The whole sarcasm thing just came up because I additionally mentioned that AIs in general are not so good in detecting sarcasm. Then a user asked in a comment for sources to that claim. That's why I added the sources about AIs and sarcasm.

An AI probably can't answer that (in general). AIs in general are still pretty dumb and struggle when it comes to detecting irony.

Sources and additional information about AIs and sarcasm

  1. Automatic Sarcasm Detection: A Survey

    Coverage of different forms of sarcasm:
    In Section 2, we described four species of sarcasm:
    propositional, lexical, like-prefixed and illocutionary sarcasm. We observe that current approaches are limited in handling the last two forms of sarcasm: like-prefixed and illocutionary. Future work may focus on these forms of sarcasm.

    • please note that this academic paper is about text recognizing
    • written by Aditya Joshi (IITB-Monash Research Academy, India; IIT Bombay, India; Monash University, Australia), Pushpak Bhattacharyya (IIT Bombay, India), Mark James Carman (Monash University, Australia) 10 Feb 2016
  2. #NOT Teaching AI how to be sarcastic is totally the easiest thing ever

    So why is detecting and interpreting sarcasm such a hard task for AI?
    Firstly, because it’s often difficult even for humans to comprehend.

    For example, a machine may not be able to tell whether a smiley face is being used to indicate happiness, humor, or a sarcastic comment. Secondly, there’s the matter of conversational and situational context.

    Thirdly, world knowledge is often needed to understand sarcastic comments.

    • written by Dr. Diana Maynard (Research fellow at the University of Sheffield) October 06, 2016
  • 22
    Why do you have so much about sarcasm in your answer? Is the compliment you suggested supposed to be said sarcastically? – Kat Apr 25 '18 at 17:02
  • 9
    @Kat: Because it is known that AIs respond differently to sarcasm than humans do, making it useful for differentiation. – Ben Voigt Apr 26 '18 at 2:54
  • 1
    @Kat On the one hand because there was comment before asking for source to my claim about AIs difficulties in understanding irony and on the other that what Ben Voigt said – undefined Apr 26 '18 at 6:24
  • 2
    @undefined Rather than focussing on sarcasm (which human are also bad at picking up on), I'd recommend to ask/say something "unexpected", outside the usual scripts of the domain (in this case medicine). Compliments may do nicely, like "Your receptionist was really nice to me during my visit, can you remind me what her name is?" – AllTheKingsHorses Apr 26 '18 at 10:23
  • @AllTheKingsHorses basically my answer wasn't so much focused on sarcasm. It came up after I added sources for some additional sentences I only added because I thought it could be useful to mention that AIs in general are bad in detecting sarcasm and irony. My main focus was the compliment I mentioned in my answer. But I would like to say that almost all sentence/phrases can be played nicely around by a good AI. Almost with some generic answers like "haha you're right" – undefined Apr 26 '18 at 10:44
54

The best strategy I can think of is to ask "can you repeat that?"

If the repeated statement is exactly the same (volume, word choice, inflection, etc.) as the first statement, you may have caught the AI.

I'm sure with some thinking you (and others) can come up with more little tricks like that.

  • 2
    I can tell you firsthand as I implemented a voice-recording quality control tool for a call centre: Those agents have scripts and some do not care to put any modulation into their voice. So humans can also sound quite robotic. How would your strategy cope with that? – user6109 Apr 27 '18 at 7:27
  • 1
    There will still be pretty obvious difference. Context and lead up influences a lot about how words are said even when there's no effort being put in. – Matthew Read Apr 27 '18 at 20:26
18

Ask them something that only a medical doctor could know. For example, pretend to forget the name of a pill your doctor prescribed.

"Oh, yes I did fetch my prescription for... what was it again?"

"Did you say I should rest for two days? or was it three?"

Even just asking why:

"What's the follow up appointment for, again?"

Even if he doesn't remember, an AI won't be able to parse these questions and can't deviate from a set response. A real doctor may answer, or say they don't know or can't remember. Any question that requires them to have medical expertise will do.

  • Wouldn’t a good AI be able to answer something like “I don’t know right now. Let me check” and then go back to the prepared script? – Alvaro Montoro Apr 27 '18 at 17:14
  • 6
    If an AI is good enough to be able to answer these questions, maybe you don't need to talk to the human doctor after all. – Joe Apr 29 '18 at 0:50
13

I am presuming that you didn't recognize the voice as being that of your doctor, as I am unaware of any AI technology in the present that can mimic a specific persons voice.

You can simply say, "I am sorry but I need to verify that you are Dr. Whomever speaking to me in person". Not only could the party on the other end of the line be a machine, it could also be a scammer. If you doctor is offended that you feel the need to ensure you are talking to the right person that really isn't your problem. You have a right to medical privacy and confidentiality.

  • I'll buy that. But how exactly do you verify the other person is who they say they are? It's not like they can show you a photo ID. – jpaugh Apr 26 '18 at 19:44
  • @jpaugh you can call them back – WoJ Apr 26 '18 at 21:25
  • Yes, get their number, use a lookup service to confirm it is likely your doctor's office, and call them. Even then, it is not an impossibility that your call isn't intercepted and you are speaking to a malicious third party, but the chance is small. It wouldn't be terribly inappropriate to ask them to provide information they should know which a 3rd party wouldn't. – Thomas Carlisle Apr 27 '18 at 16:52
9

First of all, spotting a recorded message is easier than spotting AI. By "recorded message" I mean the systems that play a full message and then have some very limited AI to interpret a yes or no response. Such messages are usually delivered very rapidly so that you wait until they are done before responding.

I've had the irritating experience of believing it was a real person inviting me to an event of some kind, with their inquiry ending with "Will you be coming?" When I began to explain why I wasn't coming, I was interrupted with a recorded, "I'm sorry, I didn't understand that. Please say 'yes' or 'no.'" Very annoying.

Rather than waiting for the complete message, now I will interject as they begin their spiel. "Hello, this is Amy calling from ____ to invite you to...." and I just say, "Oh, hi Amy," and if "she" doesn't stop to acknowledge, then I know it's a recording and I can hang up without compunction.


For more complex AI, a similar strategy will work. Just assume it's a person and don't "stick to the script." Don't follow protocol. A real person will understand you anyway; a computer won't.

Even just, "Oh hi, what's your name?" should do it. Add in, "Are you based in the hospital?" Anything that's a perfectly reasonable social question to ask a person, will make it very obvious when you're talking to a machine.

As an added advantage, if you're talking to a "wetware" machine (a person who is robotically following a dull script), adding social conversation may liven up the conversation and end up a much more enjoyable conversation for both of you. (Yes, even mundane transactions should still be enjoyable conversations.)

In other words, whether it's a digital computer or a "mental machine" you're talking to, adding your own LIVE INTEREST to the conversation will "break up" the machinery. :)

6

Did the voice on the phone actually say, "This is Dr Jones"? I think it would be unlikely that anyone would program an AI to lie about its identify. I mean, it might say, "This is Alex calling from Dr Jones office", and try to give the impression that it's a real person. But I'd be very surprised if it falsely claimed to be a specific real person. What would the gain be? If you are fooled, how is the doctor better off than if the AI did not make such a claim? And if you figure it out, you are very likely going to be annoyed about being lied to.

Anyway, the classic way to detect an AI is to change the subject. You could even do it subtly if you're worried that if it is a real person, they will be offended. Like you're describing your medical problem. So say, for example, "My sister came to see me when she heard I was sick. She lives in Georgia. Have you ever been there?" A real person will say yes or no and will likely make some polite comment about relatives or the place. An AI asking about medical conditions will probably not understand the question, so it can't say "yes" or "no". It will have to either admit it didn't understand the question or give an evasive answer.

Frankly, if you really engaged in a back-and-forth conversation and you're not certain that it's an AI, I'd guess it's not an AI.

  • A thought 9 months later: I wouldn't think a person would be offended if you asked whether this was a real person or a computer. I wouldn't be. I suppose if you started going through some long series of questions to prove it, they might get annoyed, but just someone asking, "Or you an actual person or is this a computer?" Why would that be offensive? – Jay Jan 2 at 21:18
5

Is there a polite way to ask if someone is a real person?

Well, you can ask something like,

This may sound a bit strange, but may I know whether I am talking to a real person or not?

Generally, if this is a real person, they won't mind replying. Though, you shouldn't really rely on this approach, because they can also be scripted to tell lies or auto-tuned to give the best response.

Therefore, try something suggested in the next paragraph.

Or if there is a way to determine if I'm talking to a real person (sort of a reverse turing test).

Besides the suggestions made by other users, you can also try to ask something out of medical field. Generally, AIs work through some conditions and try to find the answers written in the script, but they will not be able to answer the questions which are not written in there.

Ask something like,

I liked the [any game] from [yesterday or a particular day]. I am excited that my favorite team won. Which team is your favorite?

You can vary the question for asking their favorite car, or sports etc.

5

Are you Human or AI?


Don't be afraid to ask directly, we are in 2018 and Artificial Intelligence is everywhere.

If there's a human on the other side of the line, be honest and explain that you've been receiving several automated calls and you just want to be sure that you're talking with human being and not with zero's and one's.


Your question falls into the Turing Test category.
CAPTCHA (Completely Automated Public Turing test to tell Computers and Humans Apart) is a good example of a Turing Test.

Drawing of a computer reading a book named "Talk like a human"

Img src and nice reading - "when it Comes to Bots - the Beer Test Might be More Important than the Turing Test"

3

An AI is designed to respond to inputs, but humans do a much better job at filtering those inputs and interpreting them in context.

One thing I sometimes do is begin saying something and while in the middle of the sentence, move the phone away from my mouth and cough a couple of times.

An AI will usually respond fairly quickly with an "I didn't understand that" message (they tend to trigger when they stop detecting sound that decodes to speech). A human will recognize the sound as a cough and let you come back and finish your sentence.

Adding a "pardon me" when returning and then picking up where you left off in the sentence results in an overall sentence that sounds natural enough for a human to understand easily. When a speech-to-text engine tries to parse that sentence, it will get an undecipherable mess and likely ask for clarification.

1

Assume Every Call is a Scam

I get approximately 10 calls a day that are fake. I get 1 call a month (or less) that's legitimate.

Unless the call is someone in my Contact List, most calls I ignore and let go to voicemail (most of those hang up or leave a broken message that started playing before the beep - delete).

If I do answer and don't immediately recognize their voice as soon as the person starts speaking -- I immediately interrupt and say "Is this a recording?".

  • Most of the time the pre-recorded message continues to play and I hang up.

  • Sometimes, it will stop, pause and restart, but not address my question. I hang up.

  • Very rarely, the person will stop talking and say, "No. I'm a real person" or some such acknowledgement, and I will courteously continue the conversation.

Even with a confirmed "real" person, I will not confirm my identity nor give them any information.

  • If they ask for me by name ("Is X available", "Is this X?"), I simply say "What is this in regards to?".
  • If they press on, I say "What company are you with?"
  • If they tell me it's urgent, or ask for any information whatsoever, I tell them "Thank you for the notification. Can you give me a reference number? I'm going to hang up and call the company back, right now, myself."

I will never call any phone number they give me nor go to any website they give. I will look on my statements or cards or lookup the official website myself for the real phone number and call them back. At this point, I will inquire about any issues and then work with them giving them information they ask for, as I now know exactly with whom I am speaking (because I called them).

  • Most of this seems to be answering a completely different question than the one that was asked. – Chris H Apr 30 '18 at 8:24
1

A simple question like:

How are you doing today?

should be sufficient. I commonly ask this question during a phone call to someone, especially if they're waiting for something (e.g. a computer loading). It makes the caller feel appreciated (even if you don't appreciate the call) and can make for a more rewarding interaction.

This response is a lot more natural then some of the answers here, the only problem I can see is if they do not give you a chance to ask the question - i.e. a break in conversation. In that case perhaps interrupting and asking a more direct question would be more suitable.

0

How can I ask (or determine) if someone on the phone is real without offending them?

Programmed systems have a limited series of responses. Mentioning something from the news that struck you or sport, and listen to the response. The problem with limited conversations, is there is little reason to distract into other subjects and AI or not, both could achieve what one requires. Did you see .... it was fantastic? Emotional reflection is hard to fake or humour.

-2

In my experience, AI telephone systems aren't very good at math. In order to make my human check quick, easy and foolproof for both myself and the possible human I'm talking to, I just ask, "What's 32+2?". If it's AI, I get an answer something like, "I'm sorry. I don't have an answer to that question. Please hold the line."

Then I can just disconnect. In order for AI to be programmed with the correct answer to every simple math problem, it would seem we would need technology not currently developed. But I can't say for sure.

I have found that the answers to many predictable questions and comments are programmed into the AI, so I ask a question no one would normally ask. I got the idea from bot checkers online. It doesn't seem to offend anyone, so I continue to use the method. Usually people just laugh. If I hear the correct answer, I just say, "Thanks".

In my my experience, in the United States doctors are rarely available by phone. I've received perhaps 2 phone calls from doctors in my life, besides doctors who are friends.

If I want to speak to someone on the phone, there is usually a nurse on duty to field my call and determine whether or not I need to see the doctor, or if she can recommend something I can do myself.

However, the couple of times I did speak to a doctor on the phone, I'm sure it was a real person.

I realize google.com and search engine optimized AI systems can answer math questions, but telephone AI systems are much less powerful.

  • 2
    If you're going to ask a strange question, why not make it a common question like "How is your day going?", that way you're not going to have to explain why you asked such an odd question. It will also make the caller feel appreciated since you're asking something about them, instead of the other way around. – ChickenFeet Apr 30 '18 at 8:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.