Turing Test passed over the phone

[Puppycow]

https://ai.googleblog.com/2018/05/duplex-ai-system-for-natural-conversation.html

The system is called Google Duplex. It can talk like an actual person. It does not sound like a robot. The people on the other end of the phone are by all indications unaware that they are talking to a robot. Hence, in effect, this is a successful demonstration of a machine passing the Turing test.

Listen to the recorded examples.

True, this is so far only for very simple conversations like making an appointment at a hair salon or a dinner reservation at a restaurant. But, within those limited parameters, it seems to be indistinguishable from a real human being.

 

[casebro]

So it's like talking to Hollywood celebrity?

 

[Puppycow]

To stimulate discussion here's an article by someone who is alarmed by this:

Human or bot? Google Duplex scares me

I'm disturbed by what this technology means and why we even needed to take artificial intelligence this far. It's one thing to be able to understand the quirks of human conversation. But what good does it bring our society to have a robot pose as human?

Because eventually this could lead to genuinely human-like androids. Or intelligent computers like in the movie Her. It would be cool for starters, and the potential uses would be nearly limitless.

 

[3point14]

The % of jobs that can be automated has just shot up massively.

 

[lionking]

The % of jobs that can be automated has just shot up massively.

I think you are right. Scary indeed.

 

[3point14]

I think you are right. Scary indeed.

Yes. The system we have isn't designed to work with that level of automation.

I think there's a reasonable argument that the more that is automated, the fewer hours everyone should work.

The major benefits of massive automation over the last 20 years seem to have accrued only to the very rich*. If that trend continues we're looking at a very dystopian future.




*Cheap tellys don't count as a 'major benefit'

 

[jrhowell]

I don't think this qualifies as passing a real Turing Test. From Wikipedia:

The evaluator would be aware that one of the two partners in conversation is a machine,...

If the people on the phone had suspected that they might be talking to a bot then they could probably have tripped it up fairly easily.

 

[Dr.Sid]

I don't think this qualifies as passing a real Turing Test. From Wikipedia:


If the people on the phone had suspected that they might be talking to a bot then they could probably have tripped it up fairly easily.

No more comments needed .. oh wait ..

 

[Trebuchet]

I saw this on the teevee yesterday. I could practically HEAR the telemarketers salivating.

 

[lomiller]

I don't think this qualifies as passing a real Turing Test. From Wikipedia:

More importantly:

The format of the test allows the interrogator to give the machine a wide variety of intellectual tasks. Turing wrote that "the question and answer method seems to be suitable for introducing almost any one of the fields of human endeavour that we wish to include."[

The range of subject matter for Google Duplex seems to be very limited in any individual conversation.

 

[Pope130]

I agree with lomiller that, given the limited range of topics that this does not meet the Turing Test requirements. Still, a very interesting and useful development.

While I would prefer dealing with a human, even on routine matters like reservations, I would rather deal with a clever program than wait on hold until the battery in my phone runs down.

 

[rjh01]

That sort of thing would be great if you had a simple question you needed to ask an organisation. They could advertise "ring this number, talk to our computer, get your answer without going on hold." Maybe even do this in your choice of language.

It would put a lot of call centres out of existence. The ones that would still exist the staff would need to be highly trained to answer complex questions.

 

[Minoosh]

I've often wondered if I'm chatting with a bot when I've had some interactions with tech support. Maybe things meet in the middle - human customer service agents adopting more robot-like processing, as the robots are adopting more "human" processing.

The link is fascinating; thanks.

 

[Minoosh]

It would put a lot of call centres out of existence. The ones that would still exist the staff would need to be highly trained to answer complex questions.

I'm usually pretty amazed at the level of knowledge I get from (presumably) human agents when I'm making fairly arcane queries. When it's good, it's very good. It seems a LOT of training/experience goes into it.

 

[Skeptic Ginger]

I'm not surprised. I've posted before about robocalls that feign being a real person. When you ask "Are you a bot?" they actually answer they are using a computer to improve their messaging.

The call starts the same every time. "Ginger?" when you answer the bot responds with something like, "I'm so glad I've reached you." Then the sales pitch starts. Next time they call, I'm ready. I'm going to answer,"Mary, it's been so long, I can't believe it's you," to see what hilarity ensues.

A couple days ago I tried "chat" to resolve a Comcast problem with my email. I know from past experience the 'person' you are chatting with is a bot. At the end of the unresolved problem the bot started a sales promotion, complete with all the same tactics some spam call sales people use to keep stringing the caller along. I ended the chat but I can imagine someone convinced there was a real person on the other end might have the same ingrained social response to be polite as they would to a real person and not just hang up.

I have never been so socially correct myself. I hang up on these guys who use the string-the-caller-along tactics all the time.

 

[Skeptic Ginger]

I've often wondered if I'm chatting with a bot when I've had some interactions with tech support. Maybe things meet in the middle - human customer service agents adopting more robot-like processing, as the robots are adopting more "human" processing.

The link is fascinating; thanks.

If you think it's a bot, ask them. Or answer in gibberish and see how they respond. It's all about algorithms and you can foil the bot with a purposeful attempt.

 

[Skeptic Ginger]

I agree with lomiller that, given the limited range of topics that this does not meet the Turing Test requirements. Still, a very interesting and useful development.

While I would prefer dealing with a human, even on routine matters like reservations, I would rather deal with a clever program than wait on hold until the battery in my phone runs down.

Maybe, if the answer is simple. But if it's simple you probably could have found it in the FAQs or some other page of 'try this' answers.

In my recent Comcast experience the bot started with the same crap I had already tired, total waste of more time.

 

[Skeptic Ginger]

More importantly:

The range of subject matter for Google Duplex seems to be very limited in any individual conversation.

It didn't take me long to recognize these bots so I agree, it doesn't pass the Turing test yet.

 

[Skeptic Ginger]

I saw this on the teevee yesterday. I could practically HEAR the telemarketers salivating.

Oh they've been on this well over a year now, maybe three.

 

[Skeptic Ginger]

... Because eventually this could lead to genuinely human-like androids. Or intelligent computers like in the movie Her. It would be cool for starters, and the potential uses would be nearly limitless.

You could design the most sophisticated responsive bot algorithm and you'd still not have independent thinking of a true AI.

I think people forget the independent thought aspect of true AI.