At Apple's developer event WWDC 202, many people were curious about Apple's AI strategy, which was revealed. Apple quickly touched on its artificial intelligence strategy named Apple Intelligence Agot at the event. Hello, this is Jae-won Lee. At this year's Apple's annual developer event, WWDC 20, the focus was on the keynote of the first day. Among global big tech companies, Apple, which has been evaluated as lagging behind in artificial intelligence technology, was expected to introduce services and technologies related to AI, especially generative AI. As expected, Apple brought up AI during the keynote. Of course, this time they did not use AI, in other words, artificial intelligence. Instead, they presented a brand as Apple Intelligence, showing their insistence on providing not just common artificial intelligence, but a unique brand.

It's also a scene, oddly enough, unlike other big tech companies, there was hardly any mention of artificial intelligence itself in the announcements of artificial intelligence companies. So, there was no talk about how many parameters a certain language model or multimodal model had been developed. Instead, as Apple has always done, they focused on what experiences and benefits Apple's customers could have. So, while watching the presentation, I thought very much like Apple. On the other hand, I didn't feel any sense of significant innovation or performance overwhelming other big tech companies. So, there wasn't a feeling that Apple was coming out swinging. Nevertheless, on the other hand, it felt like Apple, which seemed lost after the debut of Siri, finally found the direction it needed to go. Let's briefly look at the content. So, what Apple has introduced.

What characteristics does Apple Intelligence, or AI, have? First of all, AI operates primarily in Apple's devices, numbering over 2.2 billion worldwide, and the apps installed on them. It operates using the data left by over 1 billion Apple users on these devices. So, what does it do? Apple introduced AI capabilities in terms of language, image, action, and context. In simple terms, it understands human language, images, and actions, and based on this, provides services tailored to individual users' contexts. It understands who you are meeting by analyzing the text in emails and calendars, recognizes who a person is in a photo, remembers what you did with that person, schedules appointments, and informs you of tasks to be done. Ultimately, the core of AI is actively utilizing user data floating within the Apple ecosystem to enhance users' artificial intelligence experiences.

It can be said that one of the things that sets Apple apart is that, compared to MSN and Google, its artificial intelligence performance may lag behind, but it plans to show a high level of artificial intelligence service experience by integrating artificial intelligence in detail and perfectly. So, what changes will Apple bring with this AI? Although they have introduced various functions, there are two main points to pay attention to. I named them in my own way: one is the advancement of Siri, and the other is Apple's unique hybrid AI. First, the artificial intelligence assistant Siri, which appeared 13 years ago, will undergo a tremendous upgrade based on AI. The current Siri is said to receive and process around 1.5 billion voice requests a day. However, the problem is that the content is very flat. So, you have to give specific commands like Siri, set an alarm for 7 o'clock or Siri, call my father, and it only performs that specific task. The artificial intelligence...

It's like a secretary, but it's actually a voice recognition app. However, Siri, based on AI, is completely different. It learns and understands the data stored in devices and apps, and when called, it can find the necessary information on its own and perform tasks. It's like being promoted from a subordinate who just did what they were told to oversee the entire app and report the results as a manager. In Apple's words, Siri is now described as a secretary who understands the contents of the screen. Let's take a look at an example mentioned in the presentation. Let's say a meeting is delayed, and your daughter might have difficulty arriving on time for her ballet performance. If you ask Siri what to do if your daughter is late for her ballet practice, Siri, as well as chatbots like GPT-3, offers very general alternatives. When I actually asked GPT-3, it advised me to calculate the departure time correctly and to check the tickets before leaving the house. This is something anyone can do.

Please tell me what you have. Of course, it may be different if we talk for a long time, but it doesn't help much in the current situation. However, let's see how Siri, upgraded AI, is. First, Siri checks my contacts and text messages to figure out who my daughter is and understands. Then, it checks the poster related to the performance my daughter sent me a few days ago, checks the place and time, returns to my calendar, checks when the current meeting started, where this place is, and what time it is now. It predicts the traffic situation when moving to the performance venue and informs me. Well, the scenes from the presentation end here, but if we imagine a little more, it will automatically send a text message to my daughter saying that I might be late, or if I want, it will call a taxi for me. In the midst of busyness, without the need for me to go back and forth between email apps, messaging apps, and taxi apps, now all I have to do is speak.

An artificial intelligence assistant for apps is called an app agent, and the way they operate is collectively referred to as the Large Action Model or LAM. It is a concept where a single artificial intelligence within the app navigates through various information remaining in the app, instead of individual app AIs, and checks it. An app agent learns behavior patterns and takes actions on behalf of the user by moving through one or several apps, which is the concept of LAM. Although this concept was presented as early as 2024, Apple really showcased it this time. Of course, Apple did not use the terms app agent or Large Action Model, but based on Apple's mention of AI as action-packed in the WWDC invitation, using the term action blockbuster in Korean, one can guess how much Apple is emphasizing the weight of actions that Siri, as an app agent, can produce. By incorporating an action button in the iPhone 15 Pro series released last time, Siri can now be launched with a physical button.

It seems that all of this was prepared for this. In addition, Apple introduced Siri's app agent function. When a situation arises where you need to enter your driver's license information, Siri automatically goes into the mobile ID app, retrieves the driver's license information, and comes back. Also, if you say you need to pick up your mother and go to the airport, Siri checks the messages with your mother, confirms the flight, checks the location of the flight in the real-time flight information app, and informs you of the arrival time of the plane. It also turns on the map app to inform you of the time it will take to get there. This app agent function is expected to mainly operate with apps developed directly by Apple in the early stages, but they are planning to release APIs for developers, so the number of apps that Siri can access within the Apple ecosystem will increase. Siri's capabilities will gradually expand, and the number of actions Siri can perform will increase. From this perspective,

When it comes to the recent collaboration between OpenAI and Apple that has garnered attention, there seems to be some ambiguity. Contrary to the speculation that Siri will replace GPT, Apple has actually incorporated GPT-4 as a basis for Siri. GPT is also integrated as one of the options accessible by the app agent Siri, to the extent that users can receive responses from GPT when requesting a search. Therefore, experts suggest that with this approach, services like entertainment and news, including Claude, may also play a role in Apple's ecosystem. Ultimately, companies with robust platforms that serve as gateways to consumers will inevitably wield greater power. We will update on this matter later. Returning to the point, I understand that Siri based on AI is powerful, but I wonder if there are any security concerns if it delves into all my personal information. To address this, Apple has prepared something.

It is Apple's own hybrid AI structure. Apple reportedly processes most AI functions using on-device AI methods. This includes the latest A17 chip installed in iPhones, as well as the M-series chips in MacBooks and iPads, which contain dedicated artificial intelligence processing cores, namely Neural Cores. This explanation states that on-device AI performance can be sufficiently enhanced without data leaving the device, eliminating concerns about security. However, Apple acknowledges that it cannot perform all AI functions solely based on on-device performance. Eventually, there are times when they need to rely on cloud AI power based on servers. Nevertheless, the cloud still raises concerns about security breaches. Therefore, Apple has established a separate server and cloud specifically for Apple Intelligence's inference operations. This so-called private cloud computer, known as Enday, is also built on Apple's proprietary Apple Silicon. Naturally,

Apple's devices and optimized servers for the data sent from those devices will be built in a hybrid AI structure. In this setup, when Apple Intelligence receives a user's request, it determines whether to process it on the device or in the cloud. If the workload exceeds the capacity of the device, it is simply thrown to the cloud. The server responsible for processing does not store data to prevent personal information leakage. Additionally, access to computational tasks is encrypted, making it impossible even for Apple developers to access. It is said to be a structure where even the entity requesting the computation is unknown. By going through rigorous security measures and borrowing computational power in this manner, they have created a safe artificial intelligence. Technical details will be discussed again when more information is released in the future.

While watching the presentation, I got the impression that Apple has put a lot of thought into what is important for AI to be useful to consumers, beyond just the technical quantifiability revealed in numbers. Ultimately, the answer seems to revolve around context, behavior, and personalized conclusions based on them. Apple needs to understand the most personal conversations and behaviors to provide a fully personalized experience, showing that AI is useful. To delve into the most personal areas of users, they have focused more on security. These features are also provided by Microsoft and Google's artificial intelligence, but Apple, based on a solid ecosystem, seems to be able to exert a stronger force. Of course, there are still many aspects to watch out for on Apple's presentation day. Issues like the response speed of Siri transformed into an app agent are expected, and similarly.

In on-device environments, there may be issues with processing speed, as well as heat generation and battery consumption. We will provide more detailed information on these issues after WWDC. That's all for the stories we prepared today. If you enjoyed today's content, please subscribe and like. Thank you.

WWDC 2024 애플 AI 전략을 설명하는 키워드 ‘액션’

Technology(기술), Trend(트렌드), Transformation(혁신)에 관해 
꼭 알아야 할 지식.  

일하는 사람들의 지적 동반자 티타임즈TV. 

www.ttimes.co.kr

Contact
주소  서울시 종로구 청계천로 11(서린동,청계한국빌딩)
e-mail  ttimes6000@gmail.com
Tel  02-724-7718