Apple’s AI ambitions: internal friction, counterattacks and challenges

Author丨Hickory

American Editor丨Fisherman

**Source: **Silicon Research Laboratory

Apple's annual autumn "Spring Festival Gala" is getting closer, but under the wave of large-scale models, when Apple will launch its own "Apple GPT" has become another topic of concern.

After all, there was Huawei before, and Xiaomi later. While Chinese mobile phone manufacturers are scrambling to install large models into mobile terminals, Apple still seems to be quite "restrained" in the progress of AI large models.

Not long ago, at the analyst meeting where the latest financial report was released, Apple CEO Cook emphasized the two keywords of "product" and "responsibility" when it came to "AI progress."

Cook said that Apple has been researching generative artificial intelligence (AI) and other models for many years, "They are embedded in almost every product we make." He further added: "We will continue to invest and innovate, responsible Use these technologies responsibly to advance our products to help enrich people's lives."

"Don't expect Apple to talk about artificial intelligence like its peers." An analyst who has long followed Apple's AI progress mentioned. According to CNBC data, compared with the "yelling" of other technology giants, Apple does not seem to want to use AI to tell more stories in this wave of technology.

Is this really the case? When technology giants around the world are actively embracing AI, Apple cannot be willing to lag behind. The reason why the AI layout is kept mysterious is actually the inevitable result of Apple's "vertical integration" and "software and hardware integration" strategies.

It is too early to say that Apple will fall behind in this wave of technology, because Apple is launching a new AI "counterattack." This article will mainly focus on two key issues:

1. Where has Apple's AI layout progressed?

2. Why is Apple so calm about AI?

01 Apple's AI layout, wasted "first-mover advantage"

Most people’s impression of Apple’s AI comes from Siri.

Siri's starting point was brilliant. Backed by international scientific research strength, it was acquired by Jobs at a price of more than 200 million US dollars. "Hey Siri" once became the specific impression of artificial intelligence in that era.

From the beginning, Paranoid Jobs positioned Siri not just as an intelligent voice assistant, but as a bet on a new way of interaction.

Jobs saw that voice-based interaction is the embodiment of human nature. Siri co-founder Norman Winarsky later recalled: "Jobs realized the power of that technology before anyone else, and he realized the uniqueness of having a personal assistant who can interact with you as a real person. value."

This also means that, unlike Google Assistant and other products positioned as search products, Siri’s landscape is actually larger.

It can be said that Jobs established a good first-mover advantage for Apple's AI layout, but since then, the vague positioning, closed ecology and management problems have made Siri gradually fall behind.

How did Siri waste its own advantages and become an "artificial mental retardation" in the minds of users? In summary, there are actually three steps:

The first step is vague positioning, jumping repeatedly between the positioning of "execution engine" and "search product".

A Siri employee once said that the internal hope was to eventually make Siri look like "Samantha" in the movie "Her" and become a virtual intelligent voice butler that assists humans. But after Jobs passed away, Apple poached talents from other technology giants, and Siri also underwent many positioning adjustments and gradually evolved into a search engine.

Bill Stasior, who has witnessed the rise and fall of Siri, was previously in charge of Amazon's search and advertising business. He tended to build Siri into a world-class search engine, so he integrated Siri with Apple's search function resources at that time to improve Siri's performance. But in fact, search engines and intelligent voice assistants do not conflict. The key is to be consistent with their own business and provide users with a better interactive experience.

Take Google's Google Assistant as an example. Although it started late, Google has already devoted almost all its resources and capabilities to Google Assistant as a bridgehead for Google's "AI first". At the same time, based on practicality and combined with its own search advantages, it fully understood the "task-oriented" needs of people in the early days of artificial intelligence.

But that's not the case with Siri. Previously, people expected Siri to be an assistant that was both practical and personalized. But since then, due to positioning as a search, most of Siri's sentences will be converted into search instructions, which cannot accurately identify the user's dialogue needs, and the experience is greatly reduced.

The second step is that excessive internal friction and unsatisfactory management have gradually marginalized Siri.

Bill Stasior may be one of the worst "beating workers" in history. In the 7 years that Apple has been in charge of the Siri project, he has experienced 4 bosses.

• The first boss was Scott Forstall. Within a few weeks of taking office, he was "swept out" by Cook because of the buggy map service.

• The second boss, Eddy Cue, Apple's head of Internet services, didn't make developing the voice assistant his top priority, even occasionally falling asleep in Siri meetings.

• The third boss is Apple software director Craig Federighi. He is mainly responsible for the development and management of Apple's operating system. He pays more attention to security and privacy, and naturally has no time to take care of Siri.

• The fourth boss is John Giannandrea, who is also the current senior vice president of Apple's machine learning and artificial intelligence strategy, but Bill Stasior is no longer involved in Siri's day-to-day management.

In addition to the turmoil in the management, Apple has adopted the method of "mergers and acquisitions + self-research". However, because Apple does not trust external companies, it also makes it difficult to reconcile technologies between different companies, thus exacerbating the loss of talent.

For example, Apple acquired Novauris, a speech recognition technology company, in 2013. However, because Apple wanted to conduct self-research, it formed a new team, hoping that the two parties would cooperate to create synergy. However, this cooperation only lasted for six months before it was interrupted, mainly due to differences in ideas, which evolved into a fight between factions.

The third step is a closed ecosystem. Apple emphasizes the principle of confidentiality. The closed ecosystem of iOS prevents external developers from getting involved, so Siri cannot form the ideal data flywheel.

It was not until the launch of iOS 10 in 2016 that the above-mentioned decline was reversed, and Apple’s new AI journey began.

The reason why iOS 10 is an important turning point in Apple's AI layout is that on the one hand, Apple's AI strategy has become clearer - it no longer places its hope on intelligent voice assistants, but integrates relevant AI achievements into one through the "combination of software and hardware". on an iPhone running iOS 10.

On the other hand, at the data and ecological level, Apple launched SiriKit to connect to third parties, and at the same time activated Apple’s “Differential Privacy”, allowing user data to be uploaded to Apple’s servers in an encrypted manner, in addition to the “privacy protection” that users are concerned about. enough effort.

In terms of development ecology, in 2017, Apple released two major development tools, Core ML and Create ML. The former allows developers to more easily train machine learning models and package them into apps, while the latter is special in that it enables AI computing on mobile devices.

Everything is ready, and Apple has officially launched its own "AI counterattack".

02 Moving towards openness, Apple’s “counterattack”

In fact, compared to Google, Microsoft and other technology giants' positive attitudes towards AI, Apple's calmness towards AI technology is, to some extent, not only a historical reason presented above, but also a kind of "vertical integration" and "software integration". The inevitability brought about by the "hard combination" strategy.

If you look at Apple as a product company or a supply chain company, it is not difficult to understand why Cook rarely talks about AI ambitions.

The reason why Apple can become a technology company that has always been valuable is due to its global industrial layout. The strong "fruit chain" supports Apple's stability and strong cost control. On the other hand, cutting-edge products that are always leading (may not be perfect, but are always controversial) allow Apple to always maximize gross profits and become the person with the right to speak in the value chain.

Under such a business model, if you understand Apple's AI layout, it is not difficult to find that Apple has actually done a lot of things silently.

The first is, vertical integration.

At present, Apple's investment in AI covers semiconductor manufacturing, machine learning, speech recognition, facial recognition, expression tracking, etc. The investment layout in these different fields has helped Apple establish an extensive AI application ecosystem. Apple rarely announces its investment plans to the outside world. The reason is that in a closed ecosystem, these AI companies are firmly controlled by Apple and become part of the "Apple Kingdom".

The second is the combination of soft and hard.

On the one hand, the AI applications of Apple's operating system can improve the performance of related products and create a differentiated experience. For example, Vision Pro will push human-computer interaction into the era of augmented reality by canceling the handles commonly equipped with VR headsets and switching to a multi-modal interaction method of "sight capture + gesture + voice".

At the same time, Apple is also very good at integrating its software capabilities to bring users a more intelligent experience. At the same time, software services are more likely to form a positive growth flywheel due to their low marginal costs.

On the other hand, Apple's AI chip self-development capabilities and hardware innovation capabilities may also promote the implementation of local AI functions on consumer electronics hardware platforms.

In 2014, Apple added a specialized AI chip to the development of its new generation architecture chip SoC. The design idea shifted from AI tasks handled by the CPU and GPU to being processed by a specialized AI chip, and eventually the chip was integrated into the terminal device. Currently, Apple’s A-series chips have been updated to A16, and Apple’s AI computing power continues to evolve. At the same time, in order to efficiently perform more difficult machine learning tasks on terminal devices, Apple launched its neural engine processor (Apple Neural Engine).

Apple's goal is to maximize the use of the Neural Engine, reduce memory consumption, and provide greater speed than the main CPU or GPU. The terminal AI functions supported by the neural engine include natural language processing, computer vision, augmented reality technology, video analysis, image management, etc. These AI functions are integrated into Apple devices. Computing power and algorithm support.

In addition, the realization of AI functions itself also requires extremely high collaborative innovation capabilities of players' software and hardware. Taking facial recognition, an important application in the field of AI, as an example, since iOS 10, Apple has begun to integrate Face ID, a face detection method based on deep learning, into the system. The implementation of Face ID needs to consider both computing power and privacy (facial data needs to be kept in the terminal device) and recognition accuracy, so it puts forward higher requirements for algorithms and hardware, and the reason why Apple can realize the above functions The landing is due to the multiple sensors and optical devices installed on the "notch screen".

In other words, under the strategy of "vertical integration" and "software and hardware integration", Apple strives to make "AI ubiquitous".

03 A more difficult innovation environment and tougher opponents

With its unique business strategy and years of AI layout, Apple seems to value the practicality and usability of AI more than other technology companies, but this does not mean that Apple can sit back and relax.

In the past period of time, it is not difficult to see Apple's anxiety from the internal deployment of "Apple GPT" to the mining of generative AI talents from technology companies such as Meta.

On the one hand, the anxiety comes from the negative effects of the "vertical integration" and "hard-software integration" strategies.

The advantage of vertical integration is that Apple has established its own kingdom, but the disadvantage is that the closed ecosystem magnifies the difficulty of Apple's management. Apple's key R&D division has been experiencing a severe brain drain. According to The Information, many engineers and executives in Apple's chip department have left to look for better opportunities, and even Johnny Srouji, Apple's senior vice president of hardware technology, expressed concerns about this.

What has brought about it is that Apple's hardware and software innovation has fallen into an embarrassing "innovator's dilemma." Taking the A-series chips as an example, in the development of recent generations, Apple's progress in self-developed chips has gradually slowed down, and a series of product function iterations have also been questioned as "toothpaste-squeezing innovation."

As Richard Kramer, a partner at the research firm Arete Research, said: "Apple has entered a stage of gradual reform." Earlier, TF International analyst Ming-Chi Kuo also said that the progress of Apple's generative AI is significantly behind competitors.

Apple's competitors are choosing All in AI in this wave of technology.

The three cloud service giants, Microsoft, Google, and Amazon, are planning to combine each business with generative AI. Amazon CEO Andrew R. Jassy also said that he plans to invest more than $50 billion in cash by 2023, with a focus on AWS technology investment. On the other side, after open-sourcing Llama 2, Meta is also seeking to combine AI with its own business.

Although Apple's latest quarterly report showed stable profitability and the success of Cook's "carrot and stick" strategy, Apple, like most technology companies, must come up with weak terminal demand and a sluggish global mobile phone market. new story.

AI may be one of the few weights in the hands of Apple and Cook. After all, as long as they are at the poker table, Trillions of Apple can always provide its own solutions.

References:

1. Smart things: "AppleGPT, Apple's AI ambitions cannot be accommodated" "Apple launched an AI counterattack! Review of 8-year layout》

2. Ai Faner: "Why is Siri getting stupider and stupider?" These former Apple employees tell the truth"

3. Tianfeng Securities: "Application-based, Apple is expected to lead local AI application innovation in the AIGC era"