In the digital arms race between autocorrect and human fallibility, it appears the machines are once again pulling ahead, sharpening their predictive prowess with the help of generative AI and an ever-growing appetite for our typos. Apple, Google, Samsung and their keyboard-wielding cohorts are all quietly overhauling their typing algorithms so that your hurried grocery list no longer includes “eggslpode” and “bannannas,” unless of course you actually meant those things, in which case we have bigger problems than autocorrect.
Apple was first out of the gate last year with what its senior VP of software engineering, Craig Federighi, boldly dubbed “transformer-based language modeling,” which, in layman’s terms, means your iPhone now guesses what you were trying to write by looking not only at the questionable word itself but also at the company it keeps. This is presumably why your phone now assumes that “ducking” is rarely what you meant to say and offers a more colorful alternative, one that feels somehow more honest during moments of rage-induced texting.
Not to be outdone, Google and Samsung have both announced similar improvements powered by on-device AI, a phrase that basically means your phone will now become slightly better at interpreting your scrolling thumbs without outsourcing the job to enormous cloud computers. It’s a bit like hiring a slightly more clever butler who never leaves your pocket and occasionally reads your mind. Though rest assured, this AI does not want your secrets, only your spelling mistakes.
These upgrades are possible because of transformer models, a technology originally cooked up by Google in 2017, which effectively allows your device to learn the nuances of language the same way a toddler learns how not to say embarrassingly accurate things in public. The difference here is your phone’s neural engine can run this complex math at the speed of petulant impatience, meaning suggestions pop up before you’ve even properly maimed the spelling of “restaurant.”
Of course, all this progress does raise faintly alarming philosophical questions. If your device becomes too good at guessing what you mean, do you still have a voice in your own texts or has the autocorrect become the author? Are you still you if every message you send sounds unerringly composed and typo-free, like a novelist with a caffeine IV? And will future generations ever know the thrill of sending a text so disastrously mangled it requires a follow-up apology and a pie chart?
For now, at least, the humans seem content to let the machines handle the spelling as long as the machines steer clear of sarcasm, political opinions and, heaven forbid, emoji recommendations.
Because nothing says “the future is here” quite like your phone quietly judging your grammar and still letting “teh” slide every once in a while, just to keep you humble.