This month, Apple held its developer conference, called WWDC. It’s an annual event that’s typically used to showcase the company’s most important innovations. Many of this year’s presentations were devoted to artificial intelligence (AI), or what Apple calls “Apple Intelligence.”
While Google and Microsoft have made forays into AI with Gemini and OpenAI, respectively, Apple has so far taken a narrower approach. The AI models deployed on iPhone hardware are relatively weak. AI models are evaluated by their many “parameters,” or variables that are adjusted during the training process. OpenAI’s GPT-4 has more than 1.5 trillion parameters, while Apple’s model has 3 billion.
For queries that require more processing power, users can choose to outsource tasks to ChatGPT via the cloud through an enterprise licensing agreement. The licensing agreement appears to be in exchange for access to OpenAI, not payment. In other words, there’s no superintelligent thinking machine at Apple — at least not yet. As a result, the response to the presentations at the conference was somewhat muted. John Herman in New York wrote that this reflects “Apple’s cautious approach,” perhaps concerned about investing too much in technology that isn’t as advanced as I’ve been hyped. Josh Tyrandziel of The Washington Post described Apple Intelligence as “the first rational theory of AI for the masses,” praising the partnership between the established computing company and the startup OpenAI and its limited scope of application. We should celebrate the fact that Apple has not yet fully entered the AI arms race. Google’s rapid pace of catching up with Microsoft’s AI development is already accelerating the decline of Google’s search tools. However, my impression of WWDC is not very optimistic. In a sense, Apple Intelligence, a small model that will eventually be installed in more than 1 billion iPhones worldwide, is a key step in determining its fate. In other words, once AI enters our personal lives, there is no turning back.
《New Personal Intelligence System》
As demonstrated in the presentation, the upcoming iPhone AI can rewrite emails on your behalf, aggregate text messages for active groups, and sort notifications to see which messages appear first, and you can also sort them. Apple CEO Tim Cook described the tool as not just a tool, but a “new personal intelligence system” that acts as a semi-autonomous second brain.
His comments reminded me of Steve Jobs, who declared in 1990 that computers were “intelligent bicycles.”
Apple says its AI tool will be more dynamic and useful than OpenAI’s ChatGPT because it will have access to all your daily information, such as where you travel, what books you read, who you talk to on the phone (ChatGPT, despite its amazing capabilities, still operates in a limited prompt window). “AI tools need to understand the user and be rooted in their personal context, including their daily lives and relationships,” Cook said at WWDC. Apple’s voice assistant Siri, which has been in the iPhone since 2011, can (mostly) understand what you say, set alarms, and check the weather. Meanwhile, Apple Intelligence operates like a turbocharged Siri, more like a ghost inside the machine, controlling the iPhone’s functions.
In the AI community, such tools are called “agents.” With access to all your contacts, messages, and calendars, agents can help you plan your life. Apple gave the example of asking the tool to “play the podcast my wife sent me the other day.” While it would be great to have a machine that can decipher such ambiguous expressions, it also means that your iPhone needs to understand who your wife is and sort out your conversations with her, which I have to think about.
The fact that Apple’s AI is designed to work on the device itself, rather than through the cloud, provides some protection for users’ vulnerable personal data. However, generative AI is still prone to random misunderstandings and “hallucinations.” Hallucination is an AI term that is a euphemism for dramatic errors. AI does not have the ability to determine what is factually accurate or relevant to reality. Cook told The Washington Post in a somewhat uncertain tone that while the tool will never be 100% accurate, “I believe it will be of very high quality.”
It would conceivably take only one freak accident to turn iPhone users away from AI. Another question asked at the conference was “When does my mother land?” If answered incorrectly, it could lead to a huge mistake at the airport pick-up. Even more terrible is the disaster of “reply all.” The AI might misunderstand the word “everyone” and send an email to everyone in your contact list. If you don’t respond to your boss’ message right away, the AI might notice and start hiding it (generative AI is known for affirming your wishes, and they even generate false facts through hallucinations based on fictional sources, such as websites and books).
AI is everywhere in our lives
The Apple event marks a softening of its closed-off attitude toward personal technology. Under Jobs and designer Jonathan Ive, Apple’s position is that its products are perfectly designed and therefore self-sufficient. Apart from the superficial choice of iPhone case color, there is no need for customization, and the operating system and graphical interface are also unified.