Why does Google's new AI wizard tell you that it's a robot – even if it sounds human



[ad_1]

Silicon Valley's quest for artificial intelligence has led to the construction of autonomous cars, drones and robots capable of backflips. But often, this trip boils down to something more prosaic, like ordering a pizza – or booking a restaurant reservation

. That's how I found myself sitting on the phone at the Hummus Shop. Oren, a cafe near the Google campus in Silicon Valley. On Tuesday afternoon, the company chose the first public test of Duplex, its new conversational badistant with artificial intelligence.

Duplex is the virtual badistant of the next generation. When the company first presented it at its developer conference in May, it engaged in such a realistic conversation – with "ums" and breaks – that the person at the other end of the call could not say that the speaker was only a software. Some asked if the interaction was false. (Google said that this was not the case.) Others raised ethical questions as to whether the Google badistant should disclose in a conversation that it is a machine – and that it can be the registration of the call

tasks: book restaurant reservations, call shops to inquire about holiday hours, and make appointments hairdresser. Ask him to make an appointment with the dentist or anything else, and he will be confused. A small group of beta testers will use the wizard as well as pre-screened companies across the United States.

In Oren, a group of reporters relayed each other on the phone, claiming that we were Oren employees responsible for responding to reservation requests.

"Hi, I'm calling to make a reservation," said a very realistic woman's voice when I picked up the phone. "I am Google's automated reservation service, so I will register the call."

"What is it?" I asked.

The software could not hear me at first – even a little background noise can get rid of the AI ​​- and it's turned into a reservation request. I repeated the question more loudly.

"I make a reservation, for a customer, for Monday 2", said the voice, undeterred.

Journalists did their best to trip Duplex, but the technology remained polite and unperturbed – and not jittery cheerfully. Duplex answered all the questions and turned the conversation around, offering to remind later if no places were available.

Google said that the "ums" and "uhs" – parts of the speech called disfluencies – were incorporated into the software after the company. realized that people were more likely to hang up without them. "We started iterating it to make it more natural," said Scott Huffman, vice president of engineering for Google Assistant. "We had a higher success rate." He said that technology can now successfully book a reservation in 4 out of 5 calls that it does. (When he fails, he can ask for help by returning to a person working in a call center of Google.)

I asked Huffman, who had the phone number. used to launch Google search, why it was not going in the opposite direction. the voice is deliberately robotic so that people slow down and become more deliberate, as if they were talking to a computer. Would not that be less misleading and produce clearer results? Huffman explained that, unlike Google Search, which is a destination that people are looking to recover, the goal of the wizard was to integrate it into the natural workflow of people and to be the less disturbing possible.

about who is on the line. If asked, "Are you a robot?" The system responds, "I am an automated system built by Google."

Huffman and a colleague, Nick Fox, stated that they had decided that Google Assistant would disclose this type of information as soon as the experts feared that people have the right to consent before talking to a bot. Executives said that legal considerations were not in their minds when they made the decisions. Two laws – one in Congress and one in California – would require such disclosure.

Duplex is not the first AI to order food effectively by the cat or to be so realistic that it can confuse people by thinking that it is human. Viv, an badistant built by the founders of Siri, was an agile pizza breaker when I saw her in 2016. And Amy, a virtual badistant I used to use to organize my meetings via e-mail, was such a natural conversation that a human secretary had interacted with her once asked me to wish her a happy holiday at the end of the year 2015. (I explained, awkwardly , that Amy was a robot, and I felt embarrbaded about not knowing the right etiquette for this unexplored human-computer interaction.)

Virtual badistant has a competitive edge due to extensive data collection that he uses to train his AI. And it moves the mainstream faster than anyone else.

"We could do robotics and expect to be treated like a robot," said Huffman. "On the other hand, we could reveal that we were a robot and that we could adapt to people's workflows." Over time, he said, "we'll see if we've made the right call. "

© The Washington Post 2018

<! –

->

[ad_2]
Source link