AI calls can fool humans! What does Google mean in the end?

lookarticle

AI calls can fool humans! What does Google mean in the end?

Wen/Han Qing

A few days ago Google held a developer conference in 2018, showing a lot of black technology, the most dazzling of which was AI (artificial intelligence) technology.

There aren't too many imaginary "visions", "futures", and no eye-catching VR/AR special effects toys. Get something very close to life directly: Google Voice Assistant!


AI calls can fool humans! What does Google mean in the end?< /p>

Google CEO"Chai Chai Ge"

Meeting in "Boss Talks" For him,

"Pichai" corresponds exactly to the Chinese Pinyin

So there is this nickname.

p>


You might say that it's not just the voice assistant, just a mobile phone, Apple has Siri, Microsoft has small Na, the Android camp has more voice assistants,What is bizarre...


AI calls can fool humans! What does Google mean in the end?

Voice Helper, unpretentious...


This is very common, but Google has developed a thought-provoking function with a new generation of artificial intelligence systems: Google AI assistants called the barbershop to make an appointment with a simulated human voice, and even successfully deceived The young lady at the front desk who answered the phone successfully made an appointment for a haircut. The younger sister did not notice that she was AI!


AI calls can fool humans! What does Google mean in the end?< /p>

Live Broadcasting

Speech via Live Broadcast

This is a Google AI assistant's live call. In the real scene of the barber shop, the person who answered the phone is a real person.


Google AI assistants are not as vocal and mechanical as speech synthesis in previous conversations. Google adopted one. The technique called speech Disfluencies makes these sounds more natural, with almost no difference from real human voices. There are transitions and convergences that sound no different from human speech.


AI calls can fool humans! What does Google mean in the end?


AI asked the clerk when he was available from 10:00 am to 12.00 pm next Tuesday. Miss Sister said letting "it" later, check it out.

This time the Google AI assistant even skinned and responded with a spoken word like “Hum.”

AI calls can fool humans! What does Google mean in the end?


< /p>

Live Video High Energy:



AI can easily distinguish the "fuzzy" and "huh" sounds that are not easily recognizable. It can also immediately think logically, the clerk feedback 10 to 12:00 appointments are full, and ask AI afternoon there is no time.


AI did not accept the clerk's suggestion, instead To: "give me an appointment a little earlier period it?


Google can do this very much. The simple sentence behind the question is a powerful natural language logic processing. System.


For example: Common chat, shopping and other software in our lives comes with some robotic AI customer service, which can answer some simple questions, but only one single The problem, and these low-level AI "intelligence is not enough" can not understand the reciprocal dialogue and branch issues.


AI calls can fool humans! What does Google mean in the end?< /p>

At this stage, many applications with the name "artificial intelligence"

are actually still in the "artificial intellectual disability "(Enjoy)" stage

The picture is from Kizuna AI, spend Q~



Google AI is really bad at this part of natural language logic, he can already understand some simple linear problems, the popular saying is When communicating with people, they can "remember what they said before, record information, and contact the context to ask new questions or answer."


Even if you ask two questions at once in a question, it can understand what you mean. The clerk asked what the AI's name was, and it was reported directly to the owner's name "Lisa."


There was no pause in the whole process. The middle clerk and sister gave AI a lot of selective questions and questions about the spur line. , but all AIs deal withFor example, performance and real people are the same.


AI calls can fool humans! What does Google mean in the end?< /p>

Miss will not think she was calling the robot


Google's artificial intelligence system has made amazing progress in speech synthesis and natural language logic processing.


Maybe you would question that the AI ​​displayed by Google Developers Conference is so arrogant that the scene calls to make a haircut appointment. Will it be a pre-buy actor fraud? Arrange the script so that it can be produced in advance.


This kind of worry is superfluous. Not only did the site have a haircut appointment, Google AI assistants gave some corporate staff a shot. After several rounds of telephone calls, some Chinese restaurants were called to order meals.


Google AI can be used proficiently to use "aha (aha)" as a phonetic modal particle. At one end, Miss Xiaojie, the customer service elder brother, did not suspect that they were talking to a robot.


AI calls can fool humans! What does Google mean in the end?< /p>

The robot first uses speech recognition technology to transfer the text to the text

Reading the text will understand what you said What is it?

Then logicallyAnalysis of how to respond to you

The last is TTS speech synthesis technology

Successful realization of AI and human interaction Dialogue


The most compelling point is: John Hennessy, the new chairman of Google’s parent Alphabet, officially announced that In the area of ​​business appointments, Google’s AI has passed the Turing test!


The famous Turing test is the father of computer science, the father of artificial intelligence, a mathematician, Logician, Alan MacPherson Turing proposed.


AI calls can fool humans! What does Google mean in the end?< /p>


Testers and testees (a person and a machine) are separated from each other by some devices (such as keyboards, Microphone) Ask the testee free to ask questions. After more than one test, if more than 30% of the testers cannot determine whether the testee is a human or a machine, the machine passes the test and is considered to have human intelligence.


AI calls can fool humans! What does Google mean in the end?< /p>

Humans need to make judgments

Whether AI or human is the answer to the question

to pass the "Turing Test"


Summarized in a sentence: This experiment is used to judge whether AI can "tempt the truth" or "cheat humanity".


This is the first time that human beings have been commercialized in this way. The "Turing test" in the realm of life is of great significance. It is a milestone, and the scene of the commercialization of artificial intelligence becomes clearer.


Although artificial intelligence has penetrated into all aspects of life, but the technology is not mature enough, the short board is very obvious, and the real people in many fields There is a long way to go, and there are even jerks such as "artificial intelligence".


AI calls can fool humans! What does Google mean in the end?

The development of new technologies and concepts is as follows

From theory to major breakthrough

Fossilized by capital and public opinion.

Technology development encounters a bottleneck and cannot support foam.

The bubble bursts and people are attracted by the "new term" such as blockchain.

New technology returns to rationality and is used in more practical contexts

Technology reappearsBreakthrough, the concept gradually matures

Marketization, scale, price of cabbage, becomes old technology


This is also the case with the VR/AR technology of the previous big fire, and it is still in a trough. Artificial intelligence technology has come to a recovery period and gradually matures. As for the recently heated block Chain technology is considered to be in the period of overheated bubble-


Google's latest AI technology shows off big tricks and throws in the technology community The next blockbuster also gave us a wake-up call: The first-mover advantage of the United States in artificial intelligence technology is very obvious. How large is the gap between China and the United States?


The University of Oxford published a report in the AI ​​field and first proposed the "National AI Potential Index (AIPI)" This concept mainly examines four areas: hardware, data, algorithms, and commerce to measure a country's overall strength in AI.


In this report, China’s current overall AI strength is actually only about half that of the United States. We still have Long way to go.


AI calls can fool humans! What does Google mean in the end?< /p>

The picture is from the Economist


United States vs. China AI The rapid growth of strength is worrying and I am afraid of losing my leading position. Just last week, the White House announced the creation of an artificial intelligence committee to advise the president and the federal government on artificial intelligence.


Learn more about Zhang Zhaozhong's articles, audio, video , Photo information, please pay attention to "Zhongzuo Zhaozhong" WeChat public number.