Apple artificial intelligence can not? You were fooled by its appearance

The outside world has always had such a perception that Apple seems to have fallen behind Google, Facebook, and even Baidu in the field of artificial intelligence.

Until Lei Fengwang (search "Lei Feng network" public number attention) issued "Apple AI is the most complete decryption, iBrain has long been everywhere" article in Apple artificial intelligence in-depth analysis after people reacted to Apple's original AI road has been low-key OK, and it has acquired more than a dozen artificial intelligence companies in the past five years.

At this point, people have another question: Apple's technology accumulation in the AI â€‹â€‹field is also very deep, but why are there few AI products?

In response to this problem, Lei Fengnet consulted the people in the industry in the Kontron Open Course.

After summing up, people in the industry have the following two perspectives:

Â AI just Apple's tools

Apple pays more attention to strengthening the accumulation of artificial intelligence underlying technology so that it can serve existing products instead of launching new artificial intelligence products frequently.

10 years ago, Jobs led Apple to start AI. According to Apple's strength, they can perform AI projects better and faster, but they did not. Apple just sees artificial intelligence as a tool, not an end.

From Apple's acquisition case, it can be concluded that in order to make Siri more perfect, Apple acquired voice recognition company Novauris and VocalIQ based on deep learning to improve the intelligence of human-computer dialogue.

In addition, in order to make Apple's original common application experience better, Apple has acquired nearly ten companies in recent years to apply its neural network and machine learning technology to existing applications, such as news, music, and restaurant recommendations; to identify unknown calls; Detect the user's motion status; list the most likely applications to use after unlocking the phone; scheduling schedules, and more. Although these features may seem insignificant, this is how Apple will apply artificial intelligence.

Apple is using artificial intelligence technology to serve existing products, while other giants are using artificial intelligence technology to launch new artificial intelligence products.

Google: The use of AI technology to develop artificial intelligence surgical robot; the introduction of driverless cars; the development of a new version of the chat robot; open source second-generation deep learning system TensorFlow.

Facebook: development of artificial intelligence assistant code-named M; release of chat robots; introduction of image scene answering system: text description by the machine based on image content; open source support for scientific computing framework of machine learning algorithms.

Baidu: Baidu chopstick search, DuBike, Xiaodu robot, Baidu map, Baidu driverless car, degree secret and other products.

Artificial intelligence applications can be roughly divided into two paths: The first is to cover more users' usage scenarios, such as smart home, autopilot, robots, smart assistants, image entertainment, etc., and to capture more information through artificial intelligence products. Information accumulation and input process.

The second aspect is to accumulate the underlying artificial intelligence technology, develop more advanced algorithms, enhance image recognition, machine learning, and voice recognition capabilities, and collect information from products that are not currently in contact with artificial intelligence to perform better. Processing and feedback belong to the process of information processing and user service output.

Apple is more inclined to the latter.

Â The reason for the lack of AI landing products

Apple's efforts at the bottom-level technology level mean that Apple's innovation in AI products is inferior to other companies. The outside world believes that Apple's lack of color at the AI â€‹â€‹application level is due to insufficient data and conservative use of data.

In response to this problem, the industry insiders of Lei Feng Net create an opinion:

AI landing can be difficult or simple. The focus is on the quality of the final product. Instead of expecting a shiny AI robot product to turn out, it's better to open your eyes and look at the data stack to see what the program is doing and use the data in a more "smart" way. In fact, Apple, Xiaomi, and Huawei face a huge amount of data for users. It is a very painful thing: there is a lot of demand, there is very little work, and the cost of storage is very high. Apple is very cautious about user data privacy, but companies that use large amounts of user data do not seem to have made very good smart applications.

AI circle circulates such a word "artificial intelligence, get the data to get the world." Compared with Google and Facebook, Apple is slightly weaker in terms of data.

However, we found that this sentence is a false proposition in the present, and the data is important, but how much can these data be used?

Earlier, when the media interviewed Yann LeCun, head of the Facebook artificial intelligence laboratory, he asked whether Facebook had made major breakthroughs in the direction of artificial intelligence. Yann LeCun replied with decisiveness "No progress has been made." Later Yann LeCun pointed out that the problem they face is that the amount of data is too large to start. What AI needs is really useful data. Discarding useless data and making effective use of key data is a difficult and highly engineering task.

Some time ago, at the CCF-GAIR conference hosted by Lei Fengnet, Huang Jiangji, vice president of Xiaomi, Fu Sheng, CEO of Cheetah, and Zhang Hongjiang, CEO of Kingsoft, all mentioned this issue: The biggest problem of artificial intelligence is how to use effective data. Millet, Cheetah, and Jinshan produce huge amounts of data every day. This is both exciting and annoying.

Therefore, the amount of data determines the success or failure of AI. At present, this sentence does not seem to hold true for first-tier technology companies such as Apple.

Apple has always been conservative in user data mining. It is reported that Apple only crawls 200MB of cached data for machine learning calls on each user, and does not use data for direct revenue generation. Therefore, all parties agreed that if Apple continues to limit the use of user data, it will seriously hinder the research results of Apple's artificial intelligence.

However, judging from the current research results, Google, Facebook, and Baidu have better data volume than Apple and more AI-enabled products. However, such products with larger data volume support are not perfect and experience is poor. Far from reaching the market's expectations. Returning to Apple, Apple did not say that Apple did not act in the launch of AI. It was better to say that Apple did not develop products that satisfied its own.

Combining these two viewpoints of the above-mentioned people in the industry, Apple's slow performance in the field of artificial intelligence is merely a representation. Apple continues its previous product development style: first, it has to do a good job of the underlying technology, improve the existing product experience, and then try new products internally. The research and development, but in order to ensure quality, research and development cycle and internal elimination rate is significantly higher than that of Google and other companies, resulting in users rarely see Apple's products at the application level.