Never mention AI, but Apple has become an artificial intelligence company

Image source: Generated by Unbounded AI

After more than a decade of development, the "mistakes" of the autocorrect function may gradually disappear. Apple's much-maligned spelling software is being upgraded with artificial intelligence: Using sophisticated language models, the new AutoCorrect doesn't just check words against a dictionary, but also makes judgments based on the context in which they are in a sentence. In theory, it doesn't misinterpret the consolation the user wants to type into as consolidation, because it knows that the words are not interchangeable.

The next-generation autocorrect feature is one of the minor updates to Apple's iPhone experience announced earlier this month. In addition, the photo app can distinguish your dog from someone else's, automatically recognizing your pet just like a common person in photos; AirPods also adjust background noise more intelligently based on listening conditions.

Apple's new AI feature may just be an "appetizer before the main course."

These features are all powered by artificial intelligence — though you probably won't know this from Apple's demo. Apple didn't mention the word "artificial intelligence" at the conference announcing the updates, although the term has become a buzzword among various tech companies right now. Instead, Apple uses more technical languages such as machine learning or Transformer language models (editor's note: a deep learning model developed by Google on which OpenAI developed GPT). Apple has been silent on the technology — so silent that it is considered lagging behind other tech companies. It's true that ChatGPT can write half-finished business proposals, while Siri can only set your morning alarm and do nothing else. But Apple is still promoting artificial intelligence in a small area, and this gradual promotion may also be the future trend of this technology.

Since ChatGPT launched last fall, tech giants have not shy away from talking about artificial intelligence — for better or worse. Sam Altman, the CEO of OpenAI, tweeted last month that AI is "by far the greatest tool." Microsoft founder Bill Gates called AI "the most important technological advancement since graphical user interfaces." And at a Google conference, Sundar Pichai, CEO of Alphabet, mentioned "artificial intelligence" 27 times in a 15-minute speech. (He has also said that artificial intelligence will be more "profound" than the meaning of humans learning to use fire.) ）

Apple, on the other hand, doesn't even pretend to be big when it comes to artificial intelligence. John Gruber, founder of tech blog Daring Fireball and a veteran Apple watcher, told me he doesn't expect any of the machine learning features Apple announced this year to dramatically change the iPhone user experience. They just make it better in name. "We just want the autocorrect feature to work smoothly." He said in an email, "We only notice it when it goes wrong." ”

The new autocorrect feature will be available in an iOS upgrade later this year, and it may feel like a weakened version of ChatGPT in your pocket. Apple says the software is better able to adjust itself to the way we type and predict the words and phrases we will use next. When you ask ChatGPT a question, you're accessing the same massive language model stored on the cloud that everyone can use. However, smaller and more customized language models that now support autocorrect will be stored on your iPhone. Tatsunori Hashimoto, a computer scientist at Stanford University, said Apple did not disclose specific details of the technology used here. Tatsunori Hashimoto and his research team are working to explore ways to scale out large language models and deploy them to mobile devices.

In addition, AirPods now use "adaptive audio" to analyze the sounds around the user and adjust accordingly. For example, when you start a chat with a barista at a coffee shop, your AirPods may automatically turn down the music and turn up the volume when you're done. Apple says it will use machine learning to understand users' volume preferences and improve their listening experience.

It's typical Apple style, "focusing on what a feature can do, not how it can be done." "Users are more concerned about using the AI technology behind this feature than what programming language the product is developed in." At the same time, he also believes that the new feature emphasizes user privacy, which Apple has always valued, or at least claims to value. Since the company uses an "in-device" model, this may pose less privacy risk than a giant, cloud-based model like ChatGPT. "In a sense, it's private because user data doesn't leave their phone, and models that are fine-tuned for users don't leave their phone." Mohsen Bayati, an AI expert at Stanford Business School, said.

Apple's differences in AI with other tech companies can be explained by their respective business models. Tech giants don't make money in the same way, with Google and Meta accounting for about half of the digital advertising market, and AI-powered chatbots could be another way to attract our consumption. Microsoft has less advertising business, but it wants to counter Google by adding chatbot functionality to search. Amazon's huge cloud services business will also benefit from the use of large language models (they must have a place to live!). ）。 Apple is a high-end brand, and the deeper business is to make your computer and phone more enjoyable to use. "So it's not surprising that Apple is approaching AI cautiously in a product-oriented way." Gruber said.

However, the iPhone may be where many people are first exposed to new advances in AI, in part because how chatbots will appear in our daily lives remains uncertain. ChatGPT was an instant hit, attracting 100 million users within two months, but it's unclear how many people still use it regularly (an OpenAI spokesperson wouldn't say the exact number when asked about the current average monthly user numbers). Many companies are also adding their own chatbot capabilities — Instacart now uses artificial intelligence to deliver recipes, and Salesforce (a company that provides customer relationship management solutions) recently launched "Einstein GPT." But chatbots still have big limitations. They are often fabricated, biased, and a copyright nightmare.

It's easy to get upset with a small technical inconvenience, but there's a reason "No, autocorrect, I'm not talking about ducking" (Editor's note: funking is changed to ducking in autocorrect) has become an internet meme. Better autocorrect would work on billions of phones, tablets, and computers: Most smartphones in the U.S. are now iPhones, while Apple has more than 2 billion active devices worldwide. Other tech giants are also using AI to make small-scale upgrades to existing products. Google recently introduced a feature for drafting Gmail responses through chatbots. The scenario of a human-like chatbot taking over everything isn't the only way AI can change the world. Many small adjustments under our noses can become a big deal. In a sense, they've done it — for years, machine learning has provided us with personalized ads, filtered our social media information, and helped us determine search results.

Like other companies in Silicon Valley, Apple may soon try further. Daniel Ives, a technology analyst at Wedbush Securities, believes Apple's new AI capabilities are just "an appetizer before the main course." His team estimates that the company has spent $8 billion to $10 billion on AI over the past four to five years — the same amount Microsoft invested in OpenAI in January, and Apple is reportedly hiring talent.

So, "Hey Siri, maybe you're running out of time." ”