During the WWDC announcement, Apple focused on how the keyboard autocorrect system in iOS 17 is powered by machine learning based on ‘transformer’ neural networks, with the aim to enhance accuracy and make the corrections feel more personalised to each user.
Having used iOS 17 for a month so far, you can definitely feel the difference. The corrections are better. It feels like it knows what you meant to type far more than any previous version of the software. It also seems more resilient to typing slang. I noticed it can cope with common texting lingo reductions like ‘wut’, opting to leave them alone instead of insisting a correction to the nearest word it finds in the dictionary. In a very unscientific test, I tried typing ‘wut’ on an iOS 16 phone — and it kept changing it to ‘wit’. Overall, the iOS 17 engine is more useful and less obstructive.
But algorithm improvements are only part of the story. Obviously, it still won’t get it right all the time. But in those cases, the experience of managing autocorrect is also improved through a superior UI. When the system does make a mistake, it is far less punishing as the interface now gives you way to quickly revert autocorrect changes. As you type, any corrected words are briefly underlined in blue. This means you can more easily notice when autocorrect changed something, and address it immediately, instead of getting through your whole message and only then spotting an error. Tapping on the underlined word shows a popup menu that lets you undo to what you literally typed, as well as some alternative suggestions to pick from. Word predictions are also much more useful, showing inline as you type. Just hit the spacebar to accept the suggestion and keep typing your message.
The smarter algorithms and smarter UI come together in a very tangible way to offer a meaningfully better experience.