SwiftKey is on a mission to make the smartest smartphone keyboard in the world, and the only way to do that is through artificial intelligence, apparently. The London-based startup known best for its Android and iOS keyboard apps is getting its hands dirty with some machine learning algorithms. The goal is to bring to market the first ever so-called neural network language model to the smartphone keyboard, and the product is available today as an in-progress Android app called SwiftKey Neural Alpha. It will use the neural network subfield of AI concerned with understanding and mimicking the human brain while churning through lots of data to help better suggest upcoming words.

If it sounds excessive for a keyboard startup to be dabbling in advanced artificial intelligence, SwiftKey marketing chief Joe Braidwood would like you to think of his company as a "language innovation" company instead. That's the core of its business and the reason it has an office in Seoul, South Korea dedicated solely to working with Samsung, the biggest licensee of SwiftKey's predictive language software.

"It's a real breakthrough for us," Braidwood says of Neural Alpha. "We think it's the next frontier in human expression for mobile devices." Neural Alpha represents a subtle shift in the product, but a powerful one the SwiftKey team thinks will change how we think about communicating as we relinquish some elements of control over what we're saying to a learning software brain that can do the work for us.

The key example the company provides is a suggestive cluster when SwiftKey reads the sentence, "I'll meet you at the [blank]." The current SwiftKey model, which is called an n-gram approach, will suggest "end," "moment," or "same" because it analyzes three-word groups and offers up the most commonly used third word from a database. The new neural network model can analyze an entire sentence and come up with responses like "airport," "office," and "hotel" by piecing together the underlying end goal of the sentence. SwiftKey doesn't have to work within the fundamental constraints of the model everyone else uses "because, at a high level, neural networks can understand the meaning of a word," Braidwood says.

What makes Neural Alpha a unique brand of tech, Braidwood says, is that its predictions are being calculated directly on the mobile device. There is no need to communicate with the cloud, as most other AI services like Apple's Siri and Microsoft's Cortana do. That engineering feat is something SwiftKey says only itself and Google has thus far accomplished. Google uses language-based neural networking running on a smartphone to power a feature on its Google Translate app that can analyze text through the device's camera, like a street sign, and translate it into other languages.

"You never want to be in a position where you are waiting for a prediction," Braidwood says of relying on cloud computing. "You want to be able to type at the speed of your thoughts." Though Apple just last year opened up iOS to third-party keyboards, SwiftKey still can't access enough of the underlying technology in the operating system to release Neural Alpha in its current form for iPhones. That said, SwiftKey is aiming to release a finished version, with iOS support or not, by the end of 2016.

SwiftKey was the first company to introduce the suggestion bar at the top of the smartphone keyboard when it launched its Android app in 2010. Since then, a number of companies like Swype, Fleksy, Minuum, and TouchPal have sprouted up with their own breed of algorithmic keyboard. All of them offer a variation on maneuvering the standard letter set, combining the suggestions with quick swiping and other techniques to improve accuracy and speed in ways our thumbs cannot achieve on standard smartphone glass.

So SwiftKey's shift to ever-smarter keyboard tech is a way for the startup to create a technology that changes how we type, but more importantly differentiates it from competitors. More and more keyboard companies are beginning to perform at the level of SwiftKey's product, and Apple, Google, and others are creating their own competing products built directly in their respective operating systems. SwiftKey now needs to offer something smarter.

"We would love to be a definitional power in understanding language," Braidwood says.