A quarter of Google Duplex calls are made by humans

midian182

Posts: 9,741   +121
Staff member
Why it matters: While there have been concerns over Google Duplex’s potential to eventually replace human workers in certain industries, it seems the technology is still far from that level. Google has revealed that when it comes to the AI booking restaurant appointments, about 25 percent of the calls are made by humans.

When Duplex was shown off at Google’s I/O conference last year there was a lot of controversy over the way it tricked people into thinking it was human when booking reservations—it can even use speech disfluencies such as “umm” and “ah.” This led to the company confirming that the system will identify itself when making calls.

Speaking to the New York Times, Google said its AI assistant isn’t yet smart enough to handle every call it makes. Twenty-five percent of these start with a human in a call center, and 15 percent begin with Duplex but get passed on to a real person.

Google said there are several factors behind why it might use a human instead of the AI to place a call, including whether the business takes reservations, or if the user of the assistant might be a spammer. When the Times tried making bookings with Duplex (via Google Assistant), it found three of the four successful reservations were carried out by humans.

Google added that using the human callers helps generate large quantities of data, which is used to train and improve future versions of Duplex.

Last July, Google was reported to be in talks to incorporate Duplex into the call center industry, but the new report suggests workers don’t have to worry about their jobs just yet.

Permalink to story.

 
There are a few industries that even the best AI may never be able to replace and anything that deals with the unpredictable behavior of humans is a good one. While AI can "learn" there are many different ways a human can ask a question, with the expectation of a wide variety of answers. The perfection of AI will be measured by the point when it can argue with the human caller with the appropriate curse word or phrase inserted at exactly the right moment .....
 
There is no such thing as AI. There is nothing intelligent about it.

Here is not the "visual" uncanny valley - it is the uncanny valley of personality. Computers are really stupid and can easily be caught misleading us into thinking there is a human. Why? A simple question can break any computer's programming. I often ask "Do computers have middle names?" and it blows them up. A skunk's white stripes can easily been seen.

Trying to mislead people they are speaking to a real person is pretty sad, and borderlines on what some may consider a lie. It's the same thing as impersonating as a Persian prince that has left you a million dollars...
 
There is no such thing as AI. There is nothing intelligent about it.

Here is not the "visual" uncanny valley - it is the uncanny valley of personality. Computers are really stupid and can easily be caught misleading us into thinking there is a human. Why? A simple question can break any computer's programming. I often ask "Do computers have middle names?" and it blows them up. A skunk's white stripes can easily been seen.

Trying to mislead people they are speaking to a real person is pretty sad, and borderlines on what some may consider a lie. It's the same thing as impersonating as a Persian prince that has left you a million dollars...

Which is why I felt like the field of research should be split into "artificial intelligence" and "artificial consciousness". AI is the ability of a computer program to solve logical problems, and AC would be the ability of a computer to simulate a human personality and stream of thought. Too much of the research gets centered around "if we just make it smarter, it will seem more human", and not enough is done with the context of "if we just make it more human, it will seem smarter".

Adding vocal ticks and speech imperfections doesn't simulate a human personality, and it never will.
 
Back