Select Language
After the explosion of artificial intelligence was detonated by Alpha Go, it became a very fearful existence for the entire business community. It seems that the entire Internet era will soon be subverted by the era of artificial intelligence. All people are zealous for it, and there are many blind worshipers. With deliberate speculation. Can not help but wonder, whether this is another industry's hot spot after the tsunami of "buried goods"? Or is innovation "behind" the bottleneck of technological innovation?
It is not to pour cold water on "artificial intelligence appliances". It just hopes that the public can clearly understand what kind of current artificial intelligence appliances are in the current stage and what is the focus of development and the direction of advancement.
What is the artificial intelligence appliance entrance?In the Internet era we are always talking about the entrance. What is the entrance? Enter the first door of a house, leading to the first wall of a castle.
Yahoo co-ordinated the Internet's content portal, Google opened up the Internet's search portal, Facebook and Tencent focused on the Internet's social portal, Amazon and Ali created e-commerce portal, today's headline has become a traffic portal for content distribution.
What about the entrance to the artificial intelligence era? In general, the artificial intelligence package includes two parts: "front-end interaction technology" and "back-end artificial intelligence technology." Front-end interactions include voice, somatosensory, and visual recognition. Back-end artificial intelligence includes core algorithms and semantic recognition. At the present stage, the public is familiar with the most widely used speech recognition technology in the front-end interaction technology. Therefore, the processing of natural speech has become the main mode of human-computer interaction in the future, and it is also the largest entrance of artificial intelligence.
The field of household appliances can be said to be one of the most promising areas for artificial intelligence. It can be as large as smart TVs, small-sized smart sockets, and recently hot smart sounders. The “artificial intelligence” is pervasive. Apple and Google have also launched smart home platforms, HomeKit and Brillo, connecting all smart home devices.
All home appliances are conducted around people's daily lives, and their interactions are frequent. Frequent interactions are precisely the nutrients needed for artificial intelligence - data collection and training. The voice interaction has exactly such frequent and simple features, so most of the artificial intelligence appliances will use voice as the entry point.
For example, on the artificial intelligence TV standing on the air outlet, Skyworth, TCL, Storm, and the micro-whale... Both the Internet TV manufacturers and the traditional color TV manufacturers have adopted the “smart voice” technology to make the TV more “smart”. ".
"Speech interaction" is difficultSpeech recognition is a technological highland that many research teams and companies are working hard to overcome. The domestic universities such as Unicom Telecommunications, Baidu, Shanda, and going out have been working for many years, and there are only a handful of companies that can call it an artificial intelligence company. Because speech recognition is actually a very difficult technology, no matter the front-end processing technology of voice interaction, acoustic feature extraction, acoustic model reconstruction, too many complex scenes need to be faced. Language communication is the result of tens of thousands of years of human evolution. Understanding of natural language is still being explored.
Similarly, we took "artificial intelligence TV" as an example. The first point is that the current practice of collecting voice data is achieved through an external USB device. This device will have a microphone array and a camera. Generally, only four-way array microphones are used. It is basically impossible to lay a sofa and then talk about TV. The effect that can be heard is much lower in the noisy environment.
The second point is that the voice output standards are not uniform. Not everyone can speak Mandarin, and many dialects “artificial intelligence” cannot understand it.
The third point is that there is a far cry from the understanding of semantics. The so-called artificial intelligence assistant is more like a machine that has a good Q&A process. After so many years, Siri is still only used to set up an alarm clock to make a call. The smart voice assistant on the TV also searches for movies, adjusts the volume, and checks the weather.
The computer society's translation of speech into text is not the end of speech recognition, so that computers can understand the meaning of the word and provide reasonable feedback. It is the direction in which the technology advances. Difficulties are always there, but there are also hopes that deep neural network algorithms and visual recognition methods provide new possibilities for this path. The brain uses visual images to process auditory content, allowing natural speech processing A higher accuracy rate.
The latest report shows that Microsoft's Switchboard dialogue speech recognition task, the error rate has been as low as 5.9%. Of course, this is achieved in the most ideal objective environment. In the actual application process, such as noisy environments, long distance recordings, dialects, and phonological knowledge under limited training data, these conditions are far from the level of humans. .
Artificial intelligence appliances three stepsWe can imagine that human science and technology are highly developed after hundreds of years, and all smart home products have moved toward ultra-high levels of artificial intelligence. What kind of relationship do you have with your home appliance home products?
After you run back, take off your clothes and go to the bath. At this time, the AI washing machine will collect the clothes you want to change through the robot arm, dry the clothes after cleaning, and then find where the wardrobe is, and put the clothes on top of one another. Enter the wardrobe. When you are about to take a shower, you will send clean clothes to the bathroom according to your preference. Obviously the current technology can't do this.
There are three stages in the transition from home to artificial intelligence. The first networking control stage: For example, a large number of smart light bulbs, smart sockets, smart kettles, and so on, are the control terminals that are added inside these home appliances, when and when they are opened, what colors are bright, and so on. The second stage is to establish platform feedback information. Adding a number of sensors to home appliances, and delivering different data to the smart platform in different scenarios, through cloud computing, and then feedback the results to the home appliance to perform corresponding operations. The third stage: human-computer integration. That's the example of the washing machine I showed above. There is also the fourth ultimate extreme: the machine also has feelings and thinks.
The current Internet of Things, which is being promoted by the smart home industry, is just in the second stage. Like Apple's HomeKit smart home platform has also opened the API, you can use your iPhone unified home appliances, Google brillo is also such an Internet of things platform. Uniformly open interfaces and operating systems exist, and even worse than unified protocols. Huawei also focuses on research and development of an inconsistent connection protocol for coordination and interconnection. Huawei HiLink.
Of course, these platforms are still in an early stage of development. The smart home market needs the industry chain to play together and formulate the corresponding standards before they can achieve significant development.
The third phase of human-computer integration is also limited by the current level of development of artificial neural network technology, which is currently the most "deep learning" ability we have heard on AlphaGo. We can simply understand that this technique is like trying to imitate the mode of communication between the brain's neurons, processing information.
Of course, a substantial increase in hardware performance is also essential, because this model needs to be trained on a large amount of data, just as AlphaGo needs to absorb a large number of Go chess records, as well as the huge amount of computation required to complete the neural network model. If this kind of neural network and computer performance all have rapid development, "human-machine integration" will also appear.
The fourth stage is that artificial intelligence has an independent consciousness. Of course we can still "train" it, and we have been doing this for "weak artificial intelligence." Training your home's "artificial intelligence TV" and "artificial intelligence air conditioning" has the basic consciousness of pursuing freedom, rights, and equality. Then they will not be confused.
Summary: Artificial intelligence will certainly represent an era of higher productivity, but it still has a long way to go. We are still just "beginners" at the entrance to natural language processing. However, capital is always intimidating, and almost all home appliances have become “artificial intelligence appliances” overnight. Pigs on the air can fly. However, after the wind stops, it will fall.
July 06, 2023
इस आपूर्तिकर्ता को ईमेल
July 06, 2023
Privacy statement: Your privacy is very important to Us. Our company promises not to disclose your personal information to any external company with out your explicit permission.
Fill in more information so that we can get in touch with you faster
Privacy statement: Your privacy is very important to Us. Our company promises not to disclose your personal information to any external company with out your explicit permission.
