"Whether the current AI functions, which are merely icing on the cake, can truly stimulate a buying spree among consumers remains a big question mark."

@TechNewKnow Original

Looking back at 2024, many mobile phone manufacturers, such as Apple and Honor, have attempted to break through in the AI mobile phone sector. However, to date, how many manufacturers have delivered satisfactory results?

Just last week, Huawei released the latest Mate70 series, which not only features the new Kirin chip but also supports the pure-blood HarmonyOS Next. Significant upgrades have been made, particularly in AI imaging, privacy protection, and intelligent interaction. Especially in AI, Huawei has attempted to showcase its continuous investment in smart technology by introducing nine major functions, including AI dynamic photos, air transmission, and intelligent noise reduction.

However, upon closer examination of these innovations, it is not difficult to see that they are still essentially incremental improvements within the existing technological framework rather than truly groundbreaking breakthroughs. For example, its highly praised AI air transmission function is essentially just a repackaging of existing gesture recognition technology.

Technically speaking, it relies solely on the sensing capabilities of the phone's front camera, combined with basic computer vision algorithms, to achieve simple gesture tracking and recognition. This technology has been attempted by brands such as Samsung and Xiaomi but has gradually been forgotten by the market due to the lack of natural interaction experience.

As for the highly anticipated "Only You Can See" function, its core technology is simply a gaze tracking algorithm based on facial recognition. This technology has existed in the PC field for many years and has now been ported to mobile phones. Moreover, its limitations in practical use are obvious: it requires users to maintain a specific viewing angle and distance, and any deviation will affect the recognition effect.

In addition, other AI functions demonstrated by the Mate70 at this press conference, such as AI call summaries and AI photo editing, are more common "standard technologies" that are generally not very innovative.

These so-called AI innovations seem more like superficial efforts that Huawei had to launch in the absence of substantive AI breakthroughs. In the face of a true AI revolution, these functions are merely superficial embellishments.

However, objectively speaking, such "innovation fatigue" is not unique to Huawei.

01

The Awkward Reality of AI Mobile Phones

In fact, the AI mobile phone sector this year was more like a mundane technology show.

Although Apple's iOS 18 heavily promoted "Apple Intelligence," apart from a small number of cross-application AI functions, most of its other capabilities are merely repackaged versions of existing technologies like ChatGPT, lacking fundamental technological breakthroughs.

At the 2024 MWC (Mobile World Congress), Honor also showcased significant technological breakthroughs in the field of AI, particularly in Agent technology.

On the whole, the automation capabilities demonstrated by its Agent technology do provide a glimpse of a future where "mobile phones are on the verge of entering the era of autonomous driving."

However, behind such visions lie two concerning issues that cannot be ignored.

For example, in the scenario of ordering coffee with one click, when users browse related information or give voice commands, the phone's Agent will directly recognize the intention and connect with relevant external applications (such as payment, maps, food ordering, etc.) to quickly complete the order creation and payment process.

Similarly, when faced with a request to "turn off automatic APP renewal," Honor's AI Agent can automatically find all items with automatic renewal services enabled and ask users if they want to cancel them.

But from the user's perspective, is such "convenience" really a pain-point-solving improvement?

On the small screen of a mobile phone, the operation of various apps has already been simplified to a threshold state. For ordinary consumers, tasks like ordering coffee do not require much effort, and existing manual operations are already convenient enough. Adding automation functions on top of this not only offers low marginal benefits but may even backfire.

The addition of automation functions, while seemingly efficient, may make some users feel that the cost of learning and adaptation is too high.

For example, the current main way to activate Honor's AI Agent is through voice interaction, but from a user experience perspective, this is still a passive function that requires conscious use. Taking the coffee ordering scenario as an example, whenever a user wants to place an order, they need to passively tell the phone with their voice: "I want to order a coffee with XX flavor, XX cups."

Since it's all passive, how is it different from the user spending a few extra seconds to operate with their fingers? Moreover, in public places or noisy scenarios where voice interaction is inconvenient, such an interaction method can make people feel uneasy.

Additionally, although the Agent system emphasizes security and privacy protection, the phone's Agent always needs to access users' personal data during task automation, including contacts, location information, and even login passwords for various apps.

To avoid privacy issues, the Agent system must repeatedly request user "authorization" and "confirmation" when performing tasks (such as turning off automatic deductions). This inevitably leads to a fragmented user experience.

Since the "disturbance" to users is so frequent, what is the point of so-called automation?

Regarding this, former Apple design chief Jony Ive believes that the greatest value of an AI Agent is not to replace user operations but to create a more natural and intuitive way of interaction that aligns with human intuition. In other words, true innovation lies in making technology "invisible."

02

The Costly AI Arms Race

Jony Ive recently collaborated with OpenAI to develop a new generation of AI devices, stating, "We need to rethink the essence of human-computer interaction. AI should not be a tool that requires learning to use; it should exist as naturally as air."

Jony Ive, former Apple design chief

Regarding the concept of "invisible" design, Jony Ive put forth three core viewpoints:

First, natural interaction. When technology truly succeeds, it should disappear into the background. Just as we don't realize the presence of air, the best AI interaction should be effortless. Second, contextual awareness. An AI Agent is not a simple command executor but should understand the user's context and intentions. It should provide assistance at the appropriate time without disrupting the user's thought process. Third, human-centered design. Technology should adapt to humans rather than requiring humans to adapt to technology.

In short, Jony Ive believes that AI should play a more proactive role as a "predictor," preparing before the user's needs arise rather than passively waiting for instructions.

For example, imagine you are driving, and your phone detects, through multidimensional data analysis (GPS location, calendar schedule, real-time traffic conditions), that you might be attending an important meeting at 2 PM.

At this point, the AI system proactively detects that the current road segment is under construction, which is expected to cause a 15-minute delay. It then automatically calculates that you need to leave 10 minutes earlier and pushes the optimal detour route to your in-car system without you even needing to inquire.

This form of interaction fully utilizes the characteristics of a mobile phone as a portable device, organically integrating various scattered data points (location, time, physiological data, calendar, etc.) to proactively create value for users rather than passively waiting for instructions.

This is the true way to apply AI in mobile scenarios, rather than simply porting Copilot or ChatGPT from computers to mobile phones.

By this standard, it is evident that neither Huawei, Honor, nor Apple have reached this "ideal state." Their so-called "automation" merely simplifies already straightforward steps through voice interaction and Agent.

This raises another important question: Are users really willing to pay a high premium for these flashy, icing-on-the-cake "AI functions"?

Regarding this, Dan Ives, a renowned Wall Street technology analyst and one of the "godfathers of tech stocks," commented that mobile phone manufacturers' investments in AI will reach an all-time high in 2024, but there is a significant gap between this investment and actual returns. What we are witnessing is an expensive "AI arms race" that ultimately hurts consumers' wallets.

03

Is the Early Advantage Being Diluted?

Facing the dual challenges of a prolonged mobile phone replacement cycle and rising component costs, Honor's investment in AI over the past year has been aggressive. In 2023, Honor's overall R&D investment accounted for 11.5% of total revenue, with cumulative AI R&D expenses reaching 10 billion.

Although there have been signs of recovery in the mobile phone market since the beginning of 2024, the question remains: Can the current AI functions, which are merely icing on the cake, truly stimulate a buying spree among consumers in the future? This remains a big question mark.

In the more than three years since its independence, Honor has launched four major product lines: Magic series, digital series, V series, and X series, encompassing over 30 mobile phones and covering a price range of 2,000 to 10,000 yuan.

However, the problem is that to gain a foothold in the continuously declining global mobile phone market and even aim for the top five, Honor cannot rely solely on high-margin premium products, which investors favor, to support sales. It still needs a large volume of mass-market models.

This creates a sharp structural contradiction with Honor's high investment in AI mobile phones.

Even more ironically, under the onslaught of specialized large model enterprises, the AI advantage built by Honor and other mobile phone manufacturers through heavy investment is gradually being diluted.

In October this year, Zhipu AI, known as one of China's "six young AI tigers," released its phone intelligence body AutoGLM based on Agent technology, directly enabling various phone functions by simulating human operational behavior.

Zhipu AI's positioning is actually to create a new value layer through technologies like AutoGLM: It is neither an application layer nor a system layer in the traditional sense but a "meta-operation layer."

The entry of large model enterprises may indicate a significant future change in mobile phone AI: AI capabilities are becoming an "infrastructure," much like the internet in its early days. For large model enterprises, future applications are more likely to be "combined services" based on API calls rather than traditional stand-alone applications, so they do not require full system permissions.

As a result, the independent AI systems launched by major mobile phone manufacturers will undoubtedly face more challenges. But frankly, even this state of "questionable practicality" has become an unattainable threshold for many mobile phone manufacturers.

According to the measurement criteria in the "AI Phone White Paper" released by Ipsos, current mobile phone AI in the industry is at the L2 level of crossing the AI Phone threshold and is moving towards L3.

The L3 level requires mobile phones to "recognize human emotions, achieve task-level autonomous disassembly and closed-loop operation, with human-AI collaboration and supervision."

This is the current state of Honor and Apple's Agent technology.

However, even major manufacturers like vivo and OPPO do not fully meet this not-so-high standard.

Taking vivo as an example, the AI system it showcased this year focused more on traditional AI interaction forms, such as "copy with one click" or "drag and drop with one click." Although its OriginOS 5 boasts vivo's self-developed mobile phone intelligence body "PhoneGPT," which can achieve functions like automatic table reservations, no public demonstration has been shown so far.

Similarly, OPPO, which claims to have independently developed a large mobile phone model, in current public demonstrations, has only focused on traditional functions such as AI image recognition, AI one-click screen inquiry, or AI photo editing. Overall, it still gives people the impression of being stuck at the L1 level of AI Phones, which are neither particularly impressive nor indispensable.

Currently, the overall functionality is barely noticeable, the interaction methods are not human-friendly or natural enough, and competition from specialized large model enterprises, among other factors, collectively pose various challenges for mobile phone AI.

Even Apple, as an industry giant, has not yet provided a satisfactory answer to the aforementioned challenges.

For example, in terms of automated operations, the so-called "intelligent agent" function demonstrated by Apple Intelligence is actually an upgraded version of iOS shortcuts. Although the system can understand commands like "help me book a flight to New York," its execution process still requires pre-configured application support.

This situation seems to have placed mobile phone AI in an awkward position: it is neither revolutionary enough to meet market expectations nor sophisticated enough to prove its value.

But if we look back, we will realize that, to some extent, this is because phone manufacturers are repeating the mistakes of the PC era—overemphasizing the accumulation of functions while neglecting the essence of user experience.