Apple managers explain why their AI is different

Apple went all-in on artificial intelligence on Monday as company executives explained the features and considering behind Apple Intelligence, the corporate's recent AI software suite.

But Apple's WWDC kickoff event was rigorously planned to set the iPhone maker other than current AI leaders like Microsoft and Google during a panel discussion on Monday afternoon.

Software chief Craig Federighi and AI chief John Giannandrea said throughout the panel that Apple is taking a special approach to technology than its Silicon Valley rivals. Unlike corporations that develop AI for a broad range of products, Apple is as a substitute focusing only on the devices it sells and the private data an AI might use.

Apple revealed a more limited approach that eschews forward-thinking concerning the technology's potential and as a substitute focuses on small tasks that may be done now without draining the battery.

“We believe the role of AI is not to replace our users, but to give them more power,” Federighi said.

Apple's AI could possibly be the primary that its over 2 billion users interact with. If its AI capabilities can compete with cloud-based competition from Microsoft or Googleit could change the best way billions of dollars price of AI infrastructure is built annually and produce a brand new focus to products that leverage this technology.

Much of the AI ​​development that has attracted the interest of investors and technology corporations has focused on constructing or securing increasingly powerful supercomputers with NVIDIA chips to develop much more power-hungry AI models. In this scenario, users access the AI ​​software by communicating with equally powerful servers over the Internet.

Apple's AI is generally in your device

Apple's vision for AI just isn’t one big model, but a series of smaller models that don't require the identical processing power and memory capability, running on Apple's own devices and chips. When the AI ​​on the phone can't do it, Apple or an app that uses Apple's tools will reach out to the cloud to access a bigger AI model. For example, Apple has partnered with OpenAI to provide users access to ChatGPT when Siri can't provide a solution. And provided that users want it.

Apple executives don’t seek advice from this strategy as using a number of models, but simply as “Apple Intelligence.”

“We believe the right approach is to offer a range of different models and different sizes for different use cases,” said Giannandrea.

Giannandrea said the corporate has been working on making a 3 billion parameter model as a part of Apple Intelligence. For comparison, ChatGPT's 2020 GPT-3 model is significantly larger, with 175 billion parameters. The more parameters, the more memory and processing power is required to run the model.

Apple's approach is quicker than cloud-based options and offers privacy advantages, but can pose problems when the models are too small to get anything done. Apple is banking on its AI having the ability to access personal data concerning the user's appointments, location and activities from the iPhone. One example Federighi cites is that his phone knows who his daughter is.

Apple also says it ensures that its small models only work on tasks they’re good at, reasonably than on an open-ended chatbot interface.

“There is one crucial additional step: We don’t take this teenager and tell him to fly a plane,” Federighi said.

Many of the AI ​​features Apple announced on Monday are much like products announced earlier this yr. Apple's AI can summarize and rewrite documents, generate small images and translate conversations in real time. One notable feature will allow users to make use of AI to create recent emojis without connecting to the web. The recent features might be released in beta this fall.

Apple's approach to data protection

Privacy might be a challenge for Apple because it uses AI. The company has used privacy as one among its key marketing tools for years, emphasizing that Apple's business model doesn’t require targeted promoting and that the corporate has its users' interests in mind, not those of knowledge brokers and spammers.

Other AI corporations collect user data and store it to enhance their software, but this practice doesn't fit with Apple's current privacy policies. Much of Apple's presentation on Monday revolved around steps the corporate has taken to avoid the impression that Apple is collecting user data to enhance its AI.

“We're not going to take that data and send it to a cloud somewhere,” Giannandrea said. “Because we want everything to remain very private, whether it's running locally or on a cloud computing service, and that's how we want it so we can use your most personal data.”

Apple has not disclosed details about what data it used to coach its AI models. Beyond that, the corporate uses files collected from the general public web in addition to licensed data comparable to news archives and stock photos.

For example, Apple said it has developed its own servers using its Apple chips, called Apple Private Cloud, to stop user data sent to an AI server from being stored or reused, allowing third parties to review the software – a notable move for a secrecy-focused company that doesn't typically provide details about its infrastructure.

“Even if a company might make a promise and say, 'Hey, look, we're not going to do anything with the data,' you have no way of verifying that,” Federighi said, explaining why Apple allows auditing of its AI server software.

More AI will follow

At times, Apple representatives looked as if it would downplay how big this shift in the corporate's AI strategy was, saying it was a continuation of the machine learning work the corporate had already done to edit photos, transcribe text or put AI-specific blocks on the corporate's chips.

“It's only recently that others have suddenly started claiming there's a new category,” said Federighi. “But these are things we've been delivering for a long time.”

Apple hasn't put all its eggs in a single approach, nevertheless. It will offer ChatGPT built into its operating systems, allowing users to call the OpenAI model totally free and get a more powerful and bigger AI model. However, OpenAI's ChatGPT might be clearly labeled in Apple's software, telling users that data is being sent to OpenAI servers (which run in Microsoft's cloud) and that responses will even show that they got here from ChatGPT in case they get out of whack.

Apple said it could offer different models in the longer term, suggesting that Apple Intelligence just isn’t the one AI system its customers are prone to use. Federighi said that a few of its customers may at some point wish to have, for instance, a medical AI system or a legal AI model built into Apple products. Or perhaps one among Google's models.

“We look forward to doing integrations with models like Google Gemini in the future. I mean, there's nothing to announce at the moment,” Federighi said. “But that's our direction.”

image credit :